Multi-level intelligent analysis method and system for cervical images

Abstract

The disclosure provides a cervical image multi-level intelligent analysis method and system, and relates to the field of cervical image multi-level intelligent analysis. The cervical image multi-level intelligent analysis method comprises the following steps: using a preset cervical cancer target detection model to detect cervical cancer targets in a cervical image; if the preset cervical cancer target detection model detects that the cervical image has a cervical cancer lesion, then the cervical image has cervical cancer; if the preset cervical cancer target detection model does not detect that the cervical image has a cervical cancer lesion, then cervical features corresponding to the cervical image are extracted; and using a preset cervical cancer early screening model and the cervical features to perform cervical cancer screening and early warning. The embodiment of the disclosure can realize multi-level intelligent analysis of cervical images.

Description

Technical Field

[0001] This disclosure relates to the field of multi-level intelligent analysis technology of cervical images, and in particular to a method and system for multi-level intelligent analysis of cervical images. Background Technology

[0002] Cervical cancer is one of the most common malignant lesions threatening women's health worldwide, ranking first in both incidence and mortality among female reproductive system lesions, seriously threatening the health of women. Clinically, cervical cancer is most common in women aged 40 to 55, but recent statistics show a trend of increasingly younger onset of new cases. Cervical cancer causes serious harm to women.

[0003] Of all cancers currently known to humans, cervical cancer is the only one that can be completely cured through early prevention and treatment. Cervical cancer has a relatively long, reversible precancerous lesion stage; it generally takes 10 to 15 years, or even 20 years, for a cervical cancer to develop from ordinary cervical inflammation. If it can be diagnosed and treated promptly during the precancerous lesion stage, the development into life-threatening invasive cancer can be prevented. Data published by the National Cancer Center in 2018 showed that in 2014, there were approximately 100,000 new cases of cervical cancer in my country, and 80% of these patients were already diagnosed with invasive cancer, resulting in a poor prognosis. The key to treatment lies in the early detection of precancerous lesions of the cervix, weakening and preventing their development into cervical cancer. Therefore, effective early screening is particularly important for the early prevention and treatment of cervical cancer.

[0004] Artificial intelligence technologies based on deep learning and machine learning have become important tools for disease diagnosis and treatment, and hold promise for application in the early screening of cervical cancer. Deep learning algorithms fully explore the deep relationships hidden in massive amounts of clinical data, obtaining more abstract and higher-level feature representations with high discriminative power through learning from deep networks, thereby improving the accuracy of disease diagnosis. Machine learning algorithms, based on the computer system's self-learning from data, make high-precision predictions and decisions with minimal human intervention. A dual-driven intelligent early screening method for cervical cancer, combining deep learning and machine learning, simulates the pathologist's slide reading process, assisting doctors in diagnosis and significantly improving pathologist efficiency.

[0005] Based on the above, this project team, relying on the Heilongjiang Provincial Natural Science Foundation project "Research on Intelligent Multi-level Screening Method for Early Cervical Cancer Based on Microomics and Deep Convolution Features" (LH2024F044), has previously proposed technical solutions for cervical image processing methods and systems (202510806369X) and intelligent cervical cancer screening methods and systems (2025108063721). The proposed cervical image processing method and system addresses the technical problem that, due to the overlapping clusters of cervical cell nuclei in liquid-based microscopic images, the overlapping state of these clusters leads to inaccurate pixel feature calculations when processing cervical cell nuclei pixels. This improves the accuracy of subsequent intelligent cervical cancer screening models. Among them, a technical solution for a cervical cancer intelligent screening method and system is proposed to solve the problem that the current cervical classification model has limited classification effect. The underlying reason is that it does not deeply mine multimodal cervical images and deep convolution features.

[0006] Although the project team has proposed the above-mentioned technical solutions in the early stage, relying on machine learning to extract cervical abnormality micromics feature information, adopting a data-driven approach to train a deep network model to realize intelligent early screening of cervical cancer. However, the existing technical solutions still have at least one of the following technical problems: (1) The existing technical solutions lack technical solutions corresponding to cervical cancer screening and early warning, which is not conducive to the early diagnosis and treatment of cervical cancer and to improving the level of early diagnosis and treatment of cervical cancer; (2) The existing technical solutions lack the entire process of diagnosis and treatment from screening and early warning to cervical cancer grading and cervical cancer metastasis prediction, which is not conducive to the multi-level intelligent diagnosis and treatment analysis of cervical cancer throughout the entire clinical process; (3) It is difficult to improve the technical problems corresponding to cervical cancer grading and / or cervical cancer metastasis prediction performance. Summary of the Invention

[0007] This disclosure presents a technical solution for a multi-level intelligent analysis method and system for cervical images.

[0008] According to one aspect of this disclosure, a multi-level intelligent analysis method for cervical images is provided, comprising: using a preset cervical cancer target detection model to perform cervical cancer target detection on cervical images; if the preset cervical cancer target detection model detects cervical cancer lesions in the cervical images, then cervical cancer exists; if the preset cervical cancer target detection model does not detect cervical cancer lesions in the cervical images, then extracting cervical features corresponding to the cervical images; and using a preset early cervical cancer screening model and the cervical features to perform cervical cancer screening and early warning.

[0009] Preferably, determining the preset cervical cancer target detection model includes: training a preset deep learning target detection network using a preset number of cervical cancer training images and their corresponding cervical cancer label training images to obtain a preset feature target detection model constructed based on the deep learning target detection network.

[0010] Preferably, determining a preset early cervical cancer screening model includes: training a preset deep learning classification network or a preset machine learning classifier using cervical features corresponding to a preset number of cervical cancer training images and cervical cancer pathology result labels corresponding to the cervical features, to obtain the preset early cervical cancer screening model.

[0011] Preferably, before training the preset deep learning classification network or preset machine learning classifier, the method includes: encoding the corresponding classification features of the cervical features to obtain quantitative features of cervical cancer lesions; and standardizing or normalizing the corresponding values ​​of the cervical features to obtain quantitative features of cervical cancer lesions.

[0012] Preferably, the multi-level intelligent analysis method for cervical images, if cervical cancer is present, further includes: segmenting the cervical cancer lesions in the cervical image to obtain cervical cancer lesion images; extracting quantitative features from the cervical cancer lesions in the cervical cancer lesion images to obtain quantitative features of the cervical cancer lesions; and using the quantitative features of the cervical cancer lesions to perform cervical cancer grading or cervical cancer metastasis prediction.

[0013] Preferably, the segmentation of cervical cancer lesions in the cervical image includes: constructing a binary hybrid location mask training image for each set of cervical cancer training images in a cervical cancer training image set, and training images of the cervix and the corresponding cervical cancer lesions within the cervix; training the preset deep learning segmentation network using the binary hybrid location mask training images, the edge mask training images of the cervical cancer lesions corresponding to the cervical cancer lesions, and the corresponding cervical cancer training images to obtain a cervical cancer lesion segmentation model; and segmenting the cervical cancer lesions in the cervical image to be segmented based on the cervical cancer lesion segmentation model.

[0014] Preferably, the quantitative feature of the cervical cancer lesion is configured as whether the cervical cancer lesion exhibits lobulation; wherein, determining whether the cervical cancer lesion exhibits lobulation includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical image to obtain an eroded cervical cancer lesion mask image; subtracting the eroded cervical cancer lesion mask image from the cervical cancer lesion mask image to obtain a processed cervical cancer lesion mask image; judging the morphology of the cervical cancer lesion in the processed cervical cancer lesion mask image; if the morphology of the cervical cancer lesion is irregular, determining whether the cervical cancer lesion exhibits lobulation based on the processed cervical cancer lesion mask image.

[0015] Preferably, the quantitative feature of the cervical cancer lesion is configured to determine whether the boundary of the cervical cancer lesion is clear; wherein, determining whether the boundary of the cervical cancer lesion is clear includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical image to obtain a cervical cancer lesion mask erosion image; subtracting the cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain a cervical cancer lesion mask-processed image; mapping the cervical cancer lesion location coordinates in the cervical cancer lesion mask-processed image to the cervical image; calculating multiple gradients corresponding to the cervical cancer lesion location coordinates in the cervical image; if the gradient magnitude values ​​corresponding to the multiple cervical cancer lesions are greater than or equal to a preset gradient magnitude value, then the boundary of the cervical cancer lesion corresponding to the preset gradient magnitude value is determined to be clear; otherwise, the boundary of the cervical cancer lesion is determined to be unclear.

[0016] Preferably, the quantitative characteristics of the cervical cancer lesion are configured as one or more of the following: the average gray value of the cervical cancer lesion, the major axis of the largest cross-sectional area of ​​the cervical cancer lesion, the minor axis of the largest cross-sectional area of ​​the cervical cancer lesion, the ratio of the major axis to the minor axis of the largest cross-sectional area of ​​the cervical cancer lesion, the location of the cervical cancer lesion on the cervix, the morphology of the cervical cancer lesion, the presence of a spike sign, the presence of a vacuolar sign, and the proportion of solid cervical cancer lesions to the total number of cervical cancer lesions.

[0017] Preferably, cervical cancer grading using the quantitative features of cervical cancer lesions includes: determining the grading model corresponding to each preset classifier in the preset classifier pool corresponding to the cervical image training set, and the sub-grading model corresponding to each preset classifier in the sub-training set of the cervical image training set; determining the sub-grading probability value set on the sub-validation set using the sub-grading model and the sub-validation set of the cervical image training set; determining the grading probability value set on the validation set using the grading model and the cervical image validation set corresponding to the cervical image training set; and optimizing the grading probability value set using the fusion weights corresponding to the sub-grading probability value set to obtain the corresponding fusion probability set for cervical cancer grading, thus completing the cervical cancer grading.

[0018] Preferably, the method of predicting cervical cancer metastasis using the quantitative features of cervical cancer lesions includes: determining the prediction model corresponding to each preset predictor in the preset predictor pool corresponding to the cervical image training set and the sub-prediction model corresponding to each preset predictor in the sub-training set of the cervical image training set; determining the sub-prediction probability value set on the sub-validation set using the sub-prediction model and the sub-validation set of the cervical image training set; determining the prediction probability value set on the validation set using the prediction model and the cervical image validation set corresponding to the cervical image training set; and optimizing the prediction probability value set using the fusion weights corresponding to the sub-prediction probability value set to obtain the corresponding fusion probability set for cervical cancer prediction, thereby completing the cervical cancer metastasis prediction.

[0019] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a detection unit for performing cervical cancer target detection on cervical images using a preset cervical cancer target detection model; a judgment unit for determining the presence of cervical cancer if the preset cervical cancer target detection model detects cervical cancer lesions in the cervical images; and extracting cervical features corresponding to the cervical images if the preset cervical cancer target detection model does not detect cervical cancer lesions in the cervical images; and a screening unit for performing cervical cancer screening and early warning using a preset early cervical cancer screening model and the cervical cancer features.

[0020] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the aforementioned multi-level intelligent analysis method for cervical images.

[0021] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a computer-readable storage medium storing a computer program / instructions and a bit stream thereon, wherein the computer program / instructions, when executed by a processor, implement the aforementioned multi-level intelligent analysis method for cervical images to generate the bit stream.

[0022] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a computer program product configured with a computer program / instruction, which, when executed by a processor, implements the aforementioned multi-level intelligent analysis method for cervical images.

[0023] In the embodiments of this disclosure, a technical solution for a multi-level intelligent analysis method and system for cervical images is proposed to solve at least one of the following technical problems: (1) Existing technical solutions lack technical solutions corresponding to cervical cancer screening and early warning, which is not conducive to the early diagnosis and treatment of cervical cancer, so as to improve the level of early diagnosis and treatment of cervical cancer; (2) Existing technical solutions lack a full-process diagnosis and treatment process from screening and early warning to cervical cancer grading and cervical cancer metastasis prediction, which is not conducive to the multi-level intelligent diagnosis and treatment analysis of cervical cancer throughout the entire clinical process; (3) It is difficult to improve the technical problem corresponding to cervical cancer grading and / or cervical cancer metastasis prediction performance.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0025] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0026] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the specification, serve to illustrate the technical solutions of this disclosure.

[0027] Figure 1 A flowchart is shown for a multi-level intelligent analysis method for cervical images according to an embodiment of the present disclosure; Figure 2 This is a block diagram illustrating an electronic device 800 according to an exemplary embodiment; Figure 3 This is a block diagram illustrating an electronic device 1900 according to an exemplary embodiment. Detailed Implementation

[0028] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0029] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0030] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0031] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0032] It is understood that the various embodiments of the multi-level intelligent analysis method for cervical images mentioned in this disclosure can be combined with each other to form combined embodiments without violating the principle and logic. Due to space limitations, this disclosure will not elaborate further.

[0033] In addition, this disclosure also provides a multi-level intelligent analysis device or system for cervical images, electronic equipment, computer-readable storage medium, and program. All of the above can be used to implement any of the multi-level intelligent analysis methods for cervical images provided in this disclosure. The corresponding technical solutions and descriptions are described in the relevant section on multi-level intelligent analysis methods for cervical images and will not be repeated here.

[0034] Figure 1 A flowchart illustrating a multi-level intelligent analysis method for cervical images according to an embodiment of this disclosure is shown. Figure 1 As shown, the multi-level intelligent analysis method for cervical images includes: Step S101: using a preset cervical cancer target detection model to perform cervical cancer target detection on the cervical image; Step S102: if the preset cervical cancer target detection model detects cervical cancer lesions in the cervical image, then cervical cancer exists; Step S103: if the preset cervical cancer target detection model does not detect cervical cancer lesions in the cervical image, then extract the cervical features corresponding to the cervical image; Step S104: using a preset early cervical cancer screening model and the cervical cancer features, perform cervical cancer screening and early warning. This addresses the lack of a corresponding technical solution for cervical cancer screening and early warning in existing technologies, which is detrimental to the technical problem of early diagnosis and treatment of cervical cancer, thereby improving the level of early diagnosis and treatment of cervical cancer.

[0035] In the embodiments disclosed herein and other possible embodiments, unlike the prior art, when the preset cervical cancer target detection model does not detect cervical cancer lesions in the cervical image, the cervical features corresponding to the cervical image can be further used for cervical cancer screening and identification to achieve early warning of cervical cancer and improve the level of early diagnosis and treatment of cervical cancer.

[0036] In the embodiments disclosed herein and other possible embodiments, the cervical image may be configured as one or more of the following: a three-dimensional cervical image, an abdominal CT image, an abdominal MRI image, an abdominal PET image, an abdominal CT-PET image, and an abdominal MRI-PET image; the cervical image may also be configured as one or more of the following: a two-dimensional cervical image, an abdominal X-ray image, an abdominal ultrasound image, a cervical microscopic image, an abdominal CT image, an abdominal MRI image, an abdominal PET image, an abdominal CT-PET image, and an abdominal MRI-PET image, corresponding to an abdominal projection image.

[0037] In this embodiment of the disclosure, determining the preset cervical cancer target detection model includes: training a preset deep learning target detection network using a preset number of cervical cancer training images and their corresponding cervical cancer label training images to obtain a preset feature target detection model constructed based on the deep learning target detection network.

[0038] In this embodiment of the disclosure, determining a preset early cervical cancer screening model includes: training a preset deep learning classification network or a preset machine learning classifier using cervical features corresponding to a preset number of cervical cancer training images and cervical cancer pathology result labels corresponding to the cervical features, thereby obtaining the preset early cervical cancer screening model.

[0039] In embodiments of this disclosure and other possible embodiments, before training the preset deep learning target detection network, the method includes: using segmentation and annotation software to delineate cervical cancer bounding boxes on a preset number of cervical images to obtain corresponding cervical cancer training images. For example, annotation software such as CasiaLabeler or Labelme can be used to delineate cervical cancer bounding boxes on a preset number of cervical cancer training images to obtain corresponding cervical cancer label training images.

[0040] In the embodiments disclosed herein and other possible embodiments, those skilled in the art can configure the preset number according to actual needs. For example, the preset number can be configured as 300-500 or other possible values. The preset deep learning object detection network can be configured as a YOLO series or an object detection network based on YOLO.

[0041] In the embodiments disclosed herein and other possible embodiments, the preset deep learning classification network can be configured as a classification network such as CNN, RNN, GCN, etc., and the preset machine learning classifier can be configured as one or more classifiers such as decision tree, Naive Bayes, support vector machine, multilayer perceptron, decision tree, random forest, gradient boosting tree, etc.

[0042] In the embodiments of this disclosure and other possible embodiments, the cervical features can be configured as radiomics features; wherein, the extraction method of the radiomics features can adopt the cervical image processing method and system (202510806369X) and the cervical cancer intelligent screening method and system (2025108063721) of the prior application of this disclosure.

[0043] For example, as disclosed in the prior art of the aforementioned application, the cervical image is configured as a cervical liquid-based microscopic image. A first feature extraction model is used to extract pixel features and morphological features corresponding to cervical cell nuclei from the cervical liquid-based microscopic image to obtain the cervical features. As another example, the cervical image is configured as an abdominal X-ray image. A second feature extraction model is used to extract cervical convolution features from the cervical X-ray image to obtain the cervical features. As yet another example, a second feature extraction model is used to extract cervical convolution features from the cervical X-ray image corresponding to the abdominal X-ray image; the cervical convolution features corresponding to the abdominal X-ray image and the pixel features and morphological features corresponding to the cervical liquid-based microscopic image are then concatenated to obtain the cervical cancer features corresponding to a multimodal feature vector.

[0044] In the embodiments of this disclosure and other possible embodiments, as disclosed in the prior art of the corresponding prior applications, the first feature extraction model is configured as an omics feature extraction model. For example, the omics feature extraction model can be configured as a PyRadiomics model. The second feature extraction model is configured as a cervical convolutional feature extraction model. For example, the cervical convolutional feature extraction model can be configured as a Med3D model or other cervical convolutional feature extraction models (e.g., GoogleNet or DenseNet).

[0045] In embodiments of this disclosure and other possible embodiments, the cervical cancer pathology result label is configured as cervical cancer risk and no cervical cancer (normal / healthy). For example, the cervical cancer risk is configured as mild dysplasia or mild to moderate dysplasia.

[0046] In this embodiment of the disclosure, before training the preset deep learning classification network or the preset machine learning classifier, the method includes: encoding the corresponding classification features of the cervical features to obtain cervical coding quantitative features; standardizing or normalizing the corresponding numerical values ​​of the cervical features to obtain cervical processing features; and using the cervical coding features, the cervical lesion processing features, and the cervical cancer pathology result labels to train the preset deep learning classification network or the preset machine learning classifier to obtain a preset cervical cancer early screening model.

[0047] In this embodiment, the multi-level intelligent analysis method for cervical images, if cervical cancer is present, further includes: segmenting the cervical cancer lesions in the cervical image to obtain cervical cancer lesion images; extracting quantitative features from the cervical cancer lesions in the cervical cancer lesion images to obtain quantitative features of the cervical cancer lesions; and using the quantitative features of the cervical cancer lesions to perform cervical cancer grading or cervical cancer metastasis prediction. This addresses the problem that existing technologies lack a complete diagnostic and treatment process from screening and early warning to cervical cancer grading or cervical cancer metastasis prediction, which is detrimental to the technical problem of multi-level intelligent diagnosis and treatment analysis of cervical cancer throughout the entire clinical process.

[0048] In this embodiment, the segmentation of cervical cancer lesions in the cervical image includes: constructing a binary hybrid location mask training image for each set of cervical cancer training images in a cervical cancer training image set, and training images of the cervix and the corresponding cervical cancer lesions within the cervix; training the preset deep learning segmentation network using the binary hybrid location mask training image, the edge mask training image of the cervical cancer lesion corresponding to the cervical cancer lesion, and the corresponding cervical cancer training image to obtain a cervical cancer lesion segmentation model; and segmenting the cervical cancer lesions in the cervical image to be segmented based on the cervical cancer lesion segmentation model. This addresses at least one of the technical problems arising from the need for manual segmentation of cervical cancer lesions or manual determination of the proportion of cervical cancer lesions, which results in a huge workload for doctors and prevents further surgical procedures or prediction of cervical cancer metastasis.

[0049] In the embodiments disclosed herein and other possible embodiments, unlike the prior art, a binary hybrid location mask training image is constructed for each set of cervical training images in the cervical training image set, corresponding to the cervix and the cervical cancer lesions within the cervix. During the training of the preset deep learning segmentation network, the preset deep learning segmentation network pays more attention to the anatomical relationship between the cervix and the cervical cancer lesions within the cervix, forming a collaborative learning and collaborative response to the cervix and the cervical cancer lesions within the cervix. This enables effective anatomical segmentation of the cervical cancer lesions within the cervix, creating anatomical constraints for the segmentation of the cervical cancer lesions within the cervix.

[0050] In the embodiments disclosed herein and other possible embodiments, the preset deep learning segmentation network is configured as one or more deep learning segmentation networks such as Fully Convolutional Network (FCN), U-Net, DeepLab, and Pyramid Spatial Pooling Network (PSPNet).

[0051] In this embodiment of the disclosure, the construction of the cervix and the binary hybrid location mask training image corresponding to the cervical cancer lesion in each set of cervical training images in the cervical training image set includes: extracting the cervical mask training image with the first mask value corresponding to the cervix and the cervical cancer lesion mask training image with the second mask value corresponding to the cervical cancer lesion in the cervix in each set of cervical training images in the cervical training image set; mapping the second mask value to the cervical position in the cervical mask training image according to the cervical cancer lesion position in the cervical cancer lesion mask training image to construct the binary hybrid location mask training image.

[0052] In this embodiment of the disclosure, determining the cervical cancer lesion edge mask training image corresponding to the cervical cancer lesion includes: performing an erosion operation on the cervical cancer lesion mask training image corresponding to each set of cervical training images in the cervical training image set to obtain a cervical cancer lesion mask erosion training image; and performing a pixel-level subtraction operation on the cervical cancer lesion mask training image and the cervical cancer lesion mask erosion training image to determine the cervical cancer lesion edge mask training image corresponding to the cervical cancer lesion.

[0053] In the embodiments disclosed herein and other possible embodiments, according to the location of the cervical cancer lesion in the cervical cancer lesion mask training image, the second mask value is mapped to the cervical location in the cervical mask training image, and the first mask value corresponding to the cervix is ​​changed to the second mask value corresponding to the cervical cancer lesion in the cervix.

[0054] In the embodiments of this disclosure and other possible embodiments, those skilled in the art can configure the first mask value and the second mask value according to actual needs. For example, the first mask value and the second mask value can be configured as 1 and 2, respectively. However, it is worth noting that, unlike the prior art where the preset mask value only represents a region corresponding to a lesion or area to be segmented, the second mask value in the embodiments of this disclosure represents both the first mask value corresponding to the cervix and the second mask value corresponding to the cervical cancer lesion within the cervix.

[0055] In the embodiments disclosed herein and other possible embodiments, if there are cervical cancer lesions in the cervix, the binary hybrid location mask training image includes a first numerical mask corresponding to the cervix and a second numerical mask corresponding to the cervical cancer lesion; if there are no cervical cancer lesions in the cervix, the binary hybrid location mask training image includes a first numerical mask corresponding to the cervix.

[0056] In the embodiments disclosed herein and other possible embodiments, the extraction of cervical mask training images with first mask values ​​corresponding to the cervix and cervical cancer lesion mask training images with second mask values ​​corresponding to the cervical cancer lesions within the cervix from each set of cervical training images in the cervical training image set includes: using segmentation annotation software CasiaLabeler or Labelme to delineate the cervix and cervical cancer lesions within the cervix in each set of cervical training images of a preset number of cervical training images, thereby obtaining cervical mask training images with first mask values ​​corresponding to the cervix and cervical cancer lesion mask training images with second mask values ​​corresponding to the cervical cancer lesions within the cervix for each set of cervical training images.

[0057] In this embodiment of the disclosure, the step of training the preset deep learning segmentation network using the binary hybrid location mask training image, the cervical cancer lesion edge mask training image corresponding to the cervical cancer lesion, and the corresponding cervical training image to obtain a cervical cancer lesion segmentation model includes: obtaining a binary hybrid location mask image during the training process of the preset deep learning segmentation network using the binary hybrid location mask training image and its corresponding cervical training image; using a preset cervical cancer lesion edge encoder to encode the cervical cancer lesion segmentation mask image corresponding to the cervical cancer lesion edge mask image in the binary hybrid location mask image and the cervical cancer lesion edge mask training image, respectively, to obtain a first edge coding feature map and a second edge coding feature map; calculating a first loss of the binary hybrid location mask image and its corresponding cervical cancer lesion edge mask training image, and calculating a second loss of the first edge coding feature map and the second edge coding feature map, to obtain a cervical cancer lesion segmentation model.

[0058] In the embodiments disclosed herein and other possible embodiments, those skilled in the art can configure the loss functions corresponding to the first loss and the second loss according to actual needs. For example, the loss functions corresponding to the first loss and the second loss can be configured as one or more of the loss functions selected from Cross-Entropy Loss, Dice Loss, Tversky Loss, and Focal Loss.

[0059] In the embodiments disclosed herein and other possible embodiments, the binary hybrid location mask image includes: a mask image corresponding to the cervix and cervical cancer lesions within the cervix; wherein the mask values ​​corresponding to the cervix and cervical cancer lesions within the cervix are different. Furthermore, a cervical cancer lesion segmentation mask image in the binary hybrid location mask image can be determined based on the mask values ​​corresponding to the cervix and cervical cancer lesions within the cervix.

[0060] Furthermore, in the embodiments disclosed herein and other possible embodiments, determining the cervical cancer lesion edge mask image includes: performing an erosion operation on the cervical cancer lesion segmentation mask image to obtain a cervical cancer lesion mask erosion image; performing a pixel-level subtraction operation on the cervical cancer lesion mask image and the cervical cancer lesion mask erosion image to determine the cervical cancer lesion edge mask image corresponding to the cervical cancer lesion.

[0061] In this embodiment of the disclosure, the construction of the preset cervical cancer lesion edge encoder includes: constructing the preset cervical cancer lesion edge encoder using the cervical cancer lesion edge mask training image corresponding to the cervical cancer lesion and its corresponding cervical cancer lesion edge training image.

[0062] In this embodiment of the disclosure, the step of constructing a preset cervical cancer lesion edge encoder using the cervical cancer lesion edge mask training image corresponding to the cervical cancer lesion and its corresponding cervical cancer lesion edge training image includes: extracting the cervical cancer lesion edge training image corresponding to each set of cervical training images in the cervical training image set; and training a preset downsampling convolutional network using the cervical cancer lesion edge training image and its corresponding cervical cancer lesion edge mask training image to obtain the preset cervical cancer lesion edge encoder.

[0063] In this embodiment of the disclosure, the step of extracting the cervical cancer lesion edge training image corresponding to each set of cervical training images in the cervical training image set includes: performing pixel-level multiplication operations on the cervical cancer lesion edge mask training image corresponding to each set of cervical training images in the cervical training image set and its corresponding cervical training image to extract the cervical cancer lesion edge training image.

[0064] In the embodiments disclosed herein and other possible embodiments, unlike the prior art, the preset cervical cancer lesion edge encoder enables the edge corresponding to the cervical cancer lesion within the cervix to participate in the training process of the preset deep learning segmentation network using the binary hybrid position mask training image and its corresponding cervical training image. This makes the edge corresponding to the cervical cancer lesion within the cervix controllable and more conducive to the segmentation of the edge of the cervical cancer lesion within the cervix.

[0065] In the embodiments disclosed herein and other possible embodiments, during the construction of the preset cervical cancer lesion edge encoder, a third loss is calculated between the cervical cancer lesion edge mask training image corresponding to the cervical cancer lesion and its corresponding cervical cancer lesion edge training image. Similarly, those skilled in the art can configure the loss function corresponding to the third loss according to actual needs. For example, the loss function corresponding to the third loss can be configured as one or more of the following loss functions: Cross-Entropy Loss, Dice Loss, Tversky Loss, and Focal Loss.

[0066] In the embodiments disclosed herein and other possible embodiments, the preset downsampling convolutional network can be configured as one or more of the downsampling convolutional networks corresponding to ResNet, EfficientNet, LeNet, AlexNet, YOLOv10, SRFD / DRFD modules, HWD modules, MobileNet, ShuffleNet, etc.; wherein, those skilled in the art can configure the kernel size, stride of the kernel, pooling and other conventional layers of the above-mentioned preset downsampling convolutional network according to actual needs.

[0067] In the embodiments disclosed herein and other possible embodiments, the preset cervical cancer lesion edge encoder corresponding to the preset downsampling convolutional network can respectively encode the cervical cancer lesion segmentation mask image corresponding to the cervical cancer lesion training image in the binary hybrid position mask image and the cervical cancer lesion edge mask image, thereby achieving one or more objectives such as compressing the spatial size of the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image, expanding the receptive field, and enhancing translation invariance. Specifically, the following features are considered: Compressing spatial dimensions: Reducing the height and width of the feature maps corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image reduces subsequent computational load, making the preset cervical cancer lesion edge encoder more efficient; Expanding the receptive field: As the size of the feature maps corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image decreases, neurons in subsequent layers can cover a wider input area, helping to capture the global features corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image; Enhancing translation invariance: The encoder is more robust to small displacements or deformations of the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image, improving the generalization ability of the preset cervical cancer lesion edge encoder.

[0068] For example, the pre-defined downsampling convolutional network can be configured as a hierarchical feature extraction structure or a gradually expanding receptive field structure. For the hierarchical feature extraction structure, it typically consists of multiple convolutional layers and pooling layers stacked alternately. With each downsampling operation, the resolution of the feature maps corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image decreases, but the number of channels may increase to retain richer semantic information. For the gradually expanding receptive field structure, with each downsampling operation on the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image, the input region that the deep neurons of the network can "see" becomes larger and larger, which helps to gradually transition from the local features corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image to the global semantic understanding corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image.

[0069] For example, by setting the stride of the convolution kernel to be greater than 1 (e.g., stride=2), the convolution kernel slides in jumps across the input feature maps corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image, directly reducing the size of the output feature maps corresponding to these images. For instance, if the input size is 5×5, the convolution kernel size is 3×3, and the stride is 2, the output size of the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image will become 2×2. Common max pooling and average pooling downsample the feature maps corresponding to the cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image by selecting the maximum or average value of the local region. For example, 2×2 max pooling will halve the size of the feature map corresponding to the above cervical cancer lesion edge mask image and the cervical cancer lesion edge mask training image.

[0070] In this embodiment of the disclosure, the quantitative feature of the cervical cancer lesion is configured as whether the cervical cancer lesion exhibits lobulation; wherein, determining whether the cervical cancer lesion exhibits lobulation includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical image to obtain a cervical cancer lesion mask erosion image; subtracting the cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain a cervical cancer lesion mask-processed image; judging the morphology of the cervical cancer lesion in the cervical cancer lesion mask-processed image; if the morphology of the cervical cancer lesion is irregular, determining whether the cervical cancer lesion exhibits lobulation based on the cervical cancer lesion mask-processed image.

[0071] In the embodiments disclosed herein and other possible embodiments, before performing the erosion operation on the cervical cancer lesion mask image corresponding to the cervical image, the method includes: using a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network to segment the cervical image into cervical cancer lesions, and determining the cervical cancer lesion mask image of the cervical image.

[0072] In the embodiments disclosed herein and other possible embodiments, the determination of the morphology of cervical cancer lesions in the cervical cancer lesion masking image includes: extracting multiple cervical cancer lesion mask position coordinates corresponding to the cervical cancer lesion mask lines in the cervical cancer lesion masking image; fitting or reconstructing the multiple cervical cancer lesion mask position coordinates to determine the morphology of the cervical cancer lesion; and determining whether the morphology of the cervical cancer lesion is regular based on the morphology of the cervical cancer lesion.

[0073] In the embodiments disclosed herein and other possible embodiments, the step of determining whether there is a lobulated feature of cervical cancer lesions based on the cervical cancer lesion mask image includes: extracting the cervical cancer lesion mask lines respectively; if the shape of the cervical cancer lesion mask lines is an arc-shaped raised connection, then the corresponding cervical cancer lesion has a lobulated feature; otherwise, there is no lobulated feature of cervical cancer lesions.

[0074] In the embodiments disclosed herein and other possible embodiments, the step of processing the image based on the cervical cancer lesion mask to determine whether there is a lobulated feature of the cervical cancer lesion further includes: segmenting the cervical cancer lesion mask line according to its concave part to obtain multiple arc-shaped raised line segments; determining whether the arc-shaped raised part is a tip based on the slope corresponding to each pixel in the multiple arc-shaped raised line segments; if it is a tip, then the corresponding cervical cancer lesion does not have a lobulated feature of the cervical cancer lesion and the cervical cancer lesion has a spike feature; if there is no tip, then the corresponding cervical cancer lesion has a lobulated feature of the cervical cancer lesion.

[0075] In the embodiments disclosed herein and other possible embodiments, determining whether the arc-shaped protrusion is a tip based on the slope corresponding to each pixel point in the multiple arc-shaped protrusion segments includes: mapping the multiple arc-shaped protrusion segments to a Cartesian coordinate system, wherein the coordinate point corresponding to the arc-shaped protrusion of the multiple arc-shaped protrusion segments is the highest pixel point of the arc-shaped protrusion segment in the Cartesian coordinate system; calculating the left derivative and right derivative corresponding to the highest pixel point; if the difference between the left derivative and the right derivative is less than or equal to a set difference, then the arc-shaped protrusion is determined to be a tip; otherwise, the arc-shaped protrusion is determined to be a non-tip.

[0076] In the embodiments disclosed herein and other possible embodiments, those skilled in the art can configure the set difference value as needed. For example, the set difference value can be configured as 1, 2, or other possible values.

[0077] In the embodiments disclosed herein and other possible embodiments, the step of fitting or reconstructing the position coordinates of the multiple cervical cancer lesion masks to determine the morphology of the cervical cancer lesion includes: under the optimal criterion, approximating multiple preset closed curve equations or multiple preset closed surface equations using the position coordinates of the multiple cervical cancer lesion masks to determine the morphology of the cervical cancer lesion.

[0078] In the embodiments disclosed herein and other possible embodiments, the step of approximating a preset closed curve equation or a preset closed surface equation using the multiple cervical cancer lesion mask position coordinates under the optimal criterion to determine the morphology of the cervical cancer lesion includes: under the optimal criterion, minimizing the error or absolute residual between the approximation equations when the multiple cervical cancer lesion mask position coordinates approximate the multiple preset closed curve equations or multiple preset closed surface equations and the multiple near-preset closed curve equations or multiple preset closed surface equations, thereby determining the morphology of the cervical cancer lesion.

[0079] In the embodiments disclosed herein and other possible embodiments, the step of approximating a preset closed curve equation or a preset closed surface equation using the multiple cervical cancer lesion mask position coordinates to determine the morphology of the cervical cancer lesion under the optimal criterion includes: if the cervical image is configured as a two-dimensional abdominal medical image, then under the optimal criterion, approximating a multiple preset closed curve equation using the multiple cervical cancer lesion mask position coordinates to determine the morphology of the cervical cancer lesion; if the cervical image is configured as a three-dimensional abdominal medical image, then under the optimal criterion, approximating a multiple preset closed surface equation using the multiple cervical cancer lesion mask position coordinates to determine the morphology of the cervical cancer lesion.

[0080] In the embodiments disclosed herein and other possible embodiments, the optimal criterion is configured as one or more of the least squares criterion, the minimum absolute residual criterion, etc.

[0081] In the embodiments disclosed herein and other possible embodiments, the optimal criterion is configured as the least squares method criterion. Then, after minimizing the error between the approximation equations when the mask coordinates of the plurality of cervical cancer lesions approximate multiple preset closed curve equations or multiple preset closed surface equations and the cervical cancer lesion mask coordinates of the multiple near-preset closed curve equations or multiple preset closed surface equations, the morphology of the plurality of cervical cancer lesions is obtained. The cervical cancer lesion mask coordinate errors corresponding to the morphologies of the plurality of cervical cancer lesions are extracted respectively. The cervical cancer lesion morphology corresponding to the cervical cancer lesion mask coordinate errors with the smallest error among the morphologies of the plurality of cervical cancer lesions is configured as the final cervical cancer lesion morphology.

[0082] In the embodiments disclosed herein and other possible embodiments, the optimal criterion is configured as the least squares method criterion. Then, after minimizing the absolute residuals of the cervical cancer lesion mask position coordinates obtained by approximating multiple preset closed curve equations or multiple preset closed surface equations to the approximation equations of the multiple cervical cancer lesion mask position coordinates, the morphologies of the multiple cervical cancer lesions are obtained. The absolute residuals of the cervical cancer lesion mask position coordinates corresponding to the morphologies of the multiple cervical cancer lesions are extracted respectively. The cervical cancer lesion morphology corresponding to the morphology with the smallest error in the absolute residuals of the cervical cancer lesion mask position coordinates corresponding to the morphologies of the multiple cervical cancer lesions is configured as the final cervical cancer lesion morphology.

[0083] In the embodiments disclosed herein and other possible embodiments, the preset closed curve equation is configured as one or more of the two-dimensional closed curves corresponding to a first-order polynomial, a second-order polynomial, or a higher-order polynomial, which are used in polynomial fitting. For example, the preset closed curve equation is configured as a closed curve corresponding to a circle, ellipse, or conic section, a closed curve enclosed by line segments corresponding to a triangle, quadrilateral, pentagon, or hexagon, a spline curve, or a polyline closed curve, etc.

[0084] In the embodiments of this disclosure and other possible embodiments, the preset closed surface equation is configured as a bivariate polynomial or a multivariate polynomial. For example, the preset closed surface equation is configured as a closed surface formed by connecting the beginning and end of line segments, such as a circular surface, elliptical surface, conical surface, triangular surface, quadrilateral surface, pentagonal surface, hexagonal surface, etc., corresponding to a closed surface formed by basic geometric closed surfaces, or a closed surface composed of multiple surfaces.

[0085] In the embodiments of this disclosure and other possible embodiments, before determining the cervical cancer lesion mask image corresponding to the cervical cancer lesion in the cervical cancer lesion mask image of the cervical image, the method includes: segmenting the cervical image for cervical cancer lesions using a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network, and determining the cervical cancer lesion mask image of the cervical image. Specifically, before segmenting the cervical image for cervical cancer lesions using the preset segmentation model constructed based on a deep learning segmentation network, the method includes: training the preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images, to obtain the preset cervical cancer lesion segmentation model constructed based on the deep learning segmentation network.

[0086] In the embodiments disclosed herein and other possible embodiments, before training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain a preset cervical cancer lesion segmentation model constructed based on the deep learning segmentation network, the method includes: using segmentation annotation software to delineate cervical cancer lesions on a preset number of cervical images to obtain corresponding cervical cancer lesion mask training images.

[0087] In the embodiments of this disclosure and other possible embodiments, before training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain a preset segmentation model constructed based on the deep learning segmentation network, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate cervical cancer lesions on a preset number of cervical images to obtain corresponding cervical cancer lesion mask training images.

[0088] In the embodiments disclosed herein and other possible embodiments, those skilled in the art can configure the preset number according to actual needs. For example, the preset number can be configured as 300-500. The preset deep learning segmentation network can be configured as one or more deep learning segmentation networks such as Fully Convolutional Network (FCN), U-Net, DeepLab, and Pyramid Spatial Pooling Network (PSPNet).

[0089] In this embodiment of the disclosure, the quantitative feature of the cervical cancer lesion is configured as whether the boundary of the cervical cancer lesion is clear; wherein, determining whether the boundary of the cervical cancer lesion is clear includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical image to obtain a cervical cancer lesion mask erosion image; subtracting the cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain a cervical cancer lesion mask processed image; mapping the cervical cancer lesion position coordinates in the cervical cancer lesion mask processed image to the cervical image; calculating multiple gradients corresponding to the cervical cancer lesion position coordinates in the cervical image respectively; if the gradient magnitude value corresponding to the multiple cervical cancer lesions is greater than or equal to a preset gradient magnitude value, then the boundary of the cervical cancer lesion corresponding to the preset gradient magnitude value is determined to be clear; otherwise, the boundary of the cervical cancer lesion is determined to be unclear. To address at least one of the following technical problems: the current requirement to manually determine the clarity of cervical cancer lesion boundaries is too subjective and lacks intelligence, resulting in a huge workload for doctors and preventing further technical adjustments to surgery, cervical cancer grading, and postoperative cervical cancer metastasis prediction.

[0090] In the embodiments disclosed herein and other possible embodiments, if the gradient magnitude value corresponding to the plurality of cervical cancer lesions is greater than or equal to a preset gradient magnitude value, then the boundary of the cervical cancer lesion corresponding to the location of the cervical cancer lesion that is greater than or equal to the preset gradient magnitude value is determined to be clear; otherwise, the boundary of the cervical cancer lesion is determined to be unclear.

[0091] In the embodiments disclosed herein and other possible embodiments, the step of calculating multiple gradients corresponding to the location coordinates of the cervical cancer lesion in the cervical image includes: determining multiple first position coordinates and multiple second position coordinates on both sides of the location coordinates of the cervical cancer lesion in the cervical image in the x-direction, and multiple third position coordinates and multiple fourth position coordinates in the y-direction, based on the location coordinates of the cervical cancer lesion; calculating the differences between the first pixel values ​​corresponding to the multiple first position coordinates and the second pixel values ​​corresponding to the multiple second position coordinates to obtain multiple x-direction gradients corresponding to the location coordinates of the cervical cancer lesion; calculating the differences between the third pixel values ​​corresponding to the multiple third position coordinates and the fourth pixel values ​​corresponding to the multiple fourth position coordinates to obtain multiple y-direction gradients corresponding to the location coordinates of the cervical cancer lesion; and configuring the multiple x-direction gradients and their corresponding multiple y-direction gradients as multiple gradients corresponding to the location coordinates of the cervical cancer lesion.

[0092] In the embodiments disclosed herein and other possible embodiments, determining multiple first position coordinates and multiple second position coordinates corresponding to both sides of the cervical cancer lesion location coordinates in the x-direction based on the cervical cancer lesion location coordinates in the cervical image includes: using the cervical cancer lesion location coordinates, determining multiple first position coordinates and multiple second position coordinates corresponding to both sides in the x-direction according to a preset distance.

[0093] In the embodiments disclosed herein and other possible embodiments, determining multiple third position coordinates and multiple fourth position coordinates corresponding to both sides of the cervical cancer lesion location coordinates in the cervical image in the y direction based on the cervical cancer lesion location coordinates includes: using the cervical cancer lesion location coordinates, determining multiple third position coordinates and multiple fourth position coordinates corresponding to both sides in the x direction according to a preset distance.

[0094] In this embodiment of the disclosure, the quantitative features of the cervical cancer lesion are configured as one or more of the following: the average gray value of the cervical cancer lesion, the major axis of the largest cross-sectional area of ​​the cervical cancer lesion, the minor axis of the largest cross-sectional area of ​​the cervical cancer lesion, the ratio of the major axis to the minor axis of the largest cross-sectional area of ​​the cervical cancer lesion, the location of the cervical cancer lesion on the cervix, the morphology of the cervical cancer lesion, whether the cervical cancer lesion has a spike sign, whether the cervical cancer lesion has a vacuolar sign, and the proportion of solid cervical cancer lesions to the total number of cervical cancer lesions.

[0095] In the embodiments of this disclosure and other possible embodiments, determining the major axis of the maximum cross-sectional area includes: performing edge detection on the lesion with the largest cross-sectional area in the lesion mask cross-sectional area of ​​each cervical cancer lesion to obtain the edge of the lesion with the largest cross-sectional area; selecting any edge pixel point in the edge of the lesion with the largest cross-sectional area; calculating multiple edge pixel point distances between the arbitrary edge pixel point and other edge pixel points in the edge of the lesion with the largest cross-sectional area; selecting the largest edge pixel point distance from the multiple edge pixel point distances; and configuring the largest edge pixel point distance as the major axis of the lesion with the largest cross-sectional area.

[0096] In the embodiments of this disclosure and other possible embodiments, determining the minor axis of the maximum cross-sectional area includes: performing edge detection on the lesion with the largest cross-sectional area in the lesion mask cross-sectional area of ​​each cervical cancer lesion to obtain the edge of the lesion with the largest cross-sectional area; selecting any edge pixel point in the edge of the lesion with the largest cross-sectional area; calculating multiple edge pixel point distances between the arbitrary edge pixel point and other edge pixel points in the edge of the lesion with the largest cross-sectional area; selecting the minimum edge pixel point distance from the multiple edge pixel point distances; and configuring the minimum edge pixel point distance as the minor axis of the lesion with the largest cross-sectional area.

[0097] In the embodiments of this disclosure and other possible embodiments, determining the ratio of the major axis to the minor axis of the maximum cross-sectional area of ​​the cervical cancer lesion includes: calculating the ratio of the major axis to the minor axis of the maximum cross-sectional area of ​​the lesion to obtain the ratio of the major axis to the minor axis of the maximum cross-sectional area of ​​the cervical cancer lesion.

[0098] In this disclosure and other possible embodiments, determining the location of the cervical cancer lesion on the cervix includes: A pre-defined cervical cancer lesion segmentation model based on a deep learning segmentation network is used to segment cervical cancer lesions in the cervical image to obtain a cervical cancer lesion mask image corresponding to the cervical image; based on the location coordinates of the cervical cancer lesion in the cervical cancer lesion mask image, cervical cancer lesions at preset locations on the cervix are determined.

[0099] In the embodiments of this disclosure and other possible embodiments, determining the cervical cancer lesion at a preset location on the cervix based on the coordinates of the cervical cancer lesion in the cervical cancer lesion mask image includes: dividing the cervical image into cervical regions using a dynamic cervical region segmentation method to obtain a cervical region segmentation network corresponding to each cervical cancer lesion; locating the cervical anatomical region corresponding to each cervical cancer lesion based on its location and the corresponding cervical region segmentation network; determining the cervical cancer lesion at a preset location on the cervix based on the cervical anatomical region corresponding to each cervical cancer lesion; and / or, determining the cervical cancer lesion at a preset location on the cervix based on the cervical anatomical region corresponding to each cervical cancer lesion includes: if the cervical anatomical region corresponding to the cervical cancer lesion is close to the cervical boundary, then the cervical cancer lesion close to the cervical boundary is determined as the cervical cancer lesion at the preset location on the cervix.

[0100] In the embodiments disclosed herein and other possible embodiments, determining whether the cervical anatomical region corresponding to the cervical cancer lesion is close to the cervical boundary includes: calculating multiple cervical cancer lesion boundary distances from each cervical cancer lesion to the cervical boundary in the cervical mask image corresponding to the cervical image; and determining whether the cervical anatomical region corresponding to the cervical cancer lesion is close to the cervical boundary based on the multiple cervical cancer lesion boundary distances and a preset distance.

[0101] In the embodiments disclosed herein and other possible embodiments, determining whether the cervical anatomical region corresponding to the cervical cancer lesion is close to the cervical boundary based on the distance between the cervical boundaries of the plurality of cervical cancer lesions and a preset distance includes: if the distance between the cervical boundaries of the plurality of cervical cancer lesions is less than or equal to the preset distance, then the cervical anatomical region corresponding to the cervical cancer lesion is determined to be close to the cervical boundary; otherwise, the cervical anatomical region corresponding to the cervical cancer lesion is determined to be far from the cervical boundary. Those skilled in the art can determine the preset distance according to actual needs. For example, the preset distance can be configured as 1-10 pixels or other pixel values.

[0102] In the embodiments of this disclosure and other possible embodiments, before segmenting the cervical images using a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network, the method includes: training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network.

[0103] In the embodiments of this disclosure and other possible embodiments, before training the preset deep learning segmentation network, the method includes: using segmentation annotation software to delineate cervical cancer lesions on a preset number of cervical images to obtain corresponding cervical cancer lesion mask training images.

[0104] In the embodiments of this disclosure and other possible embodiments, before training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain a preset segmentation model constructed based on the deep learning segmentation network, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate cervical cancer lesions on the preset number of cervical images to obtain corresponding cervical cancer lesion mask training images. Those skilled in the art can configure the preset number according to actual needs. For example, the preset number can be configured as 300-500.

[0105] In the embodiments disclosed herein and other possible embodiments, the preset deep learning segmentation network is configured as one or more deep learning segmentation networks such as Fully Convolutional Network (FCN), U-Net, DeepLab, and Pyramid Spatial Pooling Network (PSPNet).

[0106] In the embodiments of this disclosure and other possible embodiments, determining the cervical cancer lesion at a preset location on the cervix includes: segmenting the cervix in the cervical image using a preset cervical segmentation model constructed based on a deep learning segmentation network to obtain a cervical mask image corresponding to the cervical image; extracting the cervix from the cervical mask image with a first preset edge width to obtain a cervical image with a first preset edge width; segmenting the cervical cancer lesion in the cervical image with the first preset edge width using a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network to obtain a cervical cancer lesion mask image with a first preset edge width; and determining the cervical cancer lesion in the cervical cancer lesion mask image with the first preset edge width as the cervical cancer lesion at a preset location on the cervix; thereby solving at least one of the technical problems of excessive subjectivity in cervical cancer lesion location positioning, the huge workload for doctors caused by the lack of intelligent cervical cancer lesion location positioning, and the inability to further predict postoperative conditions.

[0107] In the embodiments disclosed herein and other possible embodiments, the step of extracting the cervix from the cervical mask image with a first preset edge width to obtain a cervical image with a first preset edge width includes: performing an erosion operation on the cervix in the cervical mask image using the first preset edge width to obtain a cervical erosion mask image; performing a pixel-level subtraction operation on the cervical mask image and the cervical erosion mask image to obtain a cervical mask image with a first preset edge width; and performing a pixel-level multiplication operation on the cervical image and the first preset edge width cervical mask image to obtain a cervical image with a first preset edge width.

[0108] In this embodiment and other possible embodiments, the mask value corresponding to the cervix in the cervical mask image can be configured to 1. Those skilled in the art can determine the first preset edge width according to actual needs. For example, the first preset edge width can be configured to 1-10 pixels or other pixel values.

[0109] In the embodiments of this disclosure and other possible embodiments, before segmenting the cervical image using the preset cervical segmentation model constructed based on a deep learning segmentation network, the method includes: training the preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical mask training images to obtain the preset cervical segmentation model constructed based on the deep learning segmentation network.

[0110] In the embodiments of this disclosure and other possible embodiments, before training the preset deep learning segmentation network, the method includes: using segmentation annotation software to delineate the cervix on a preset number of cervical images to obtain corresponding cervical mask training images.

[0111] In the embodiments of this disclosure and other possible embodiments, before training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical mask training images to obtain a preset segmentation model constructed based on the deep learning segmentation network, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate the cervix on the preset number of cervical images to obtain corresponding cervical mask training images. Those skilled in the art can configure the preset number according to actual needs.

[0112] In this disclosure and other possible embodiments, determining the peritoneum mask image corresponding to the peritoneum includes: performing a dilation operation on the cervix in the cervical mask image using a second preset edge width to obtain a cervical dilation mask image; and performing pixel-level subtraction operations on the cervical dilation mask image and the cervical mask image to determine the peritoneum mask image corresponding to the peritoneum. This addresses the technical problem that the peritoneum is not clearly displayed on the cervical image, making it difficult to delineate the peritoneum and thus preventing the obtaining of a preset peritoneum segmentation model based on a deep learning segmentation network that meets performance requirements.

[0113] In the embodiments disclosed herein and other possible embodiments, the detection of the peritoneal region image of the cervical cancer lesion at a predetermined location on the cervix and adjacent peritoneum in the cervical image includes: drawing multiple rays corresponding to multiple predetermined directions from the boundary line or surface of the cervical cancer lesion at the predetermined location to the boundary line or surface of the peritoneal mask adjacent to the peritoneum; and defining the image area enclosed by the rays intersecting the peritoneal mask boundary line or surface, the boundary line or surface of the cervical cancer lesion at the predetermined location, and the peritoneal mask boundary line or surface of the peritoneum adjacent to the peritoneum as the peritoneal region image of the cervical cancer lesion.

[0114] In the embodiments disclosed herein and other possible embodiments, determining the image area enclosed by the rays intersecting the peritoneal mask boundary line or peritoneal mask surface, the cervical cancer lesion boundary line or cervical cancer lesion surface at the preset location, and the peritoneal mask boundary line or peritoneal mask surface adjacent to the cervical cancer lesion at the preset location as the peritoneal region image of the cervical cancer lesion includes: determining the image area enclosed by the rays intersecting the rays intersecting the peritoneal mask boundary line or peritoneal mask surface, the cervical cancer lesion boundary line or cervical cancer lesion surface at the preset location, and the peritoneal mask boundary line or peritoneal mask surface adjacent to the peritoneum at the preset location as the peritoneal region mask image of the cervical cancer lesion; performing a pixel-level multiplication operation on the peritoneal region mask image of the cervical cancer lesion and the cervical image to determine the peritoneal region image of the cervical cancer lesion.

[0115] In this disclosure and other possible embodiments, if the cervical image is configured as a two-dimensional abdominal medical image, the detection of the peritoneal region image of the cervical cancer lesion at a preset location on the cervix and adjacent peritoneum in the cervical image includes: drawing multiple rays corresponding to multiple predetermined directions from the cervical cancer lesion boundary line at the preset location to the peritoneal mask boundary line adjacent to the peritoneum at the preset location; and determining the area enclosed by the rays intersecting the peritoneal mask boundary line, the cervical cancer lesion boundary line at the preset location, and the peritoneal mask boundary line adjacent to the peritoneum at the preset location as the peritoneal region image of the cervical cancer lesion.

[0116] In the embodiments disclosed herein and other possible embodiments, the step of drawing multiple rays corresponding to multiple predetermined directions from the boundary line of the cervical cancer lesion at the preset location to the peritoneal mask boundary line adjacent to the peritoneum at the preset location includes: setting an initial radiation point on the boundary line of the cervical cancer lesion at the preset location; determining multiple radiation points on the boundary line of the cervical cancer lesion using the initial radiation point and a preset radiation point step distance; and drawing multiple rays corresponding to multiple predetermined two-dimensional directions, each starting from the multiple radiation points.

[0117] In this disclosure and other possible embodiments, if the cervical image is configured as a three-dimensional abdominal medical image, then detecting the peritoneal region image of the cervical cancer lesion at a preset location on the cervix and adjacent peritoneum in the cervical image includes: drawing multiple rays corresponding to multiple predetermined directions from the surface of the cervical cancer lesion at the preset location towards a peritoneal mask surface adjacent to the peritoneum at the preset location; defining the area enclosed by the rays intersecting the peritoneal mask surface, the surface of the cervical cancer lesion at the preset location, and the peritoneum or peritoneal mask surface adjacent to the peritoneum at the preset location as the peritoneal region image of the cervical cancer lesion; and / or, The process of casting multiple rays in multiple predetermined directions from the surface of the cervical cancer lesion at the predetermined location to a peritoneal mask surface adjacent to the peritoneum at the predetermined location includes: determining a peritoneal mask surface on one side corresponding to the cervical anatomical region adjacent to the cervical cancer lesion at the predetermined location; setting an initial radiation point on the peritoneal mask surface on one side; determining multiple radiation points on the peritoneal mask surface on one side based on the initial radiation point and a predetermined radiation point step distance; and casting multiple rays in multiple predetermined three-dimensional directions starting from the multiple radiation points.

[0118] In this disclosure and other possible embodiments, the corresponding peritoneal mask boundary line or peritoneal mask surface is determined based on the peritoneal mask image corresponding to the peritoneum. Specifically, determining the corresponding peritoneal mask boundary line or peritoneal mask surface based on the peritoneal mask image corresponding to the peritoneum includes: if the cervical image is configured as a two-dimensional abdominal medical image, then edge detection is performed on the peritoneal mask image corresponding to the peritoneum to determine the corresponding peritoneal mask boundary line; if the cervical image is configured as a three-dimensional abdominal medical image, then edge detection is performed on each layer of the peritoneal mask image corresponding to the peritoneum to obtain the peritoneal mask edge image of each layer; and three-dimensional reconstruction is performed on each peritoneal mask edge image to determine the corresponding peritoneal mask surface.

[0119] In this disclosure and other possible embodiments, determining the peritoneal mask surface corresponding to the cervical anatomical region adjacent to the preset location cervical cancer lesion includes: dividing the peritoneum into a left peritoneum and a right peritoneum according to the cervical anatomical region of the preset location cervical cancer lesion; wherein the left peritoneum and the right peritoneum correspond to the left peritoneal mask surface and the right peritoneal mask surface, respectively. Therefore, before determining the peritoneal mask surface corresponding to the cervical anatomical region adjacent to the preset location cervical cancer lesion, the method includes: dividing the peritoneum into multiple cervical anatomical peritoneal mask surfaces according to the cervical anatomical region of the preset location cervical cancer lesion; determining the cervical anatomical peritoneal mask surface corresponding to the cervical anatomical region adjacent to the preset location cervical cancer lesion according to the multiple cervical anatomical peritoneal mask surfaces; setting initial radiation points on the cervical anatomical peritoneal mask surface; and determining multiple radiation points on the cervical mask surface using the initial radiation points and preset radiation point step distances.

[0120] For example, based on the cervical anatomical region of the cervical cancer lesion at the preset location, the peritoneum is divided into left peritoneum and right peritoneum; a left peritoneal mask surface or a right peritoneal mask surface corresponding to the cervical anatomical region adjacent to the cervical cancer lesion at the preset location is determined; an initial radiation point is set on the left peritoneal mask surface or the right peritoneal mask surface; multiple radiation points on the peritoneal mask surface are determined based on the initial radiation point and the preset radiation point step distance.

[0121] In this disclosure and other possible embodiments, if the cervical anatomical region of the cervical cancer lesion at the preset location is close to the left anterior peritoneum, then the left peritoneal mask surface corresponding to the left anterior peritoneum is retained; if the cervical anatomical region of the cervical cancer lesion at the preset location is close to the right anterior peritoneum, then the right peritoneal mask surface corresponding to the left anterior peritoneum is retained. Based on the above, those skilled in the art can determine the specific peritoneal mask surface according to the actual anatomical structure of the cervix. Therefore, the peritoneal mask surface can also be configured as a peritoneal mask surface corresponding to different peritoneal locations. For example, a peritoneal mask surface at any location on the right peritoneum.

[0122] In the embodiments disclosed herein and other possible embodiments, those skilled in the art can determine the second preset edge width according to actual needs. For example, the second preset edge width can be configured to 1-5 pixels or other pixel values.

[0123] In the embodiments of this disclosure and other possible embodiments, determining whether there is a peritoneal traction sign for cervical cancer lesions in the cervical image based on the peritoneal region image of the cervical cancer lesion includes: detecting whether there are cord-like or linear features between the peritoneum adjacent to the cervical cancer lesion at the preset location in the peritoneal region image of the cervical cancer lesion; if so, determining that there is a peritoneal traction sign for cervical cancer lesions in the cervical image; otherwise, determining that there is no peritoneal traction sign for cervical cancer lesions in the cervical image.

[0124] In the embodiments of this disclosure and other possible embodiments, the step of detecting whether there are cord-like or linear features between the peritoneum adjacent to the cervical cancer lesion at the preset location in the peritoneal region image of the cervical cancer lesion includes: using a preset feature target detection model based on deep learning to perform feature target detection on the peritoneal region image of the cervical cancer lesion, and determining whether there are cord-like or linear features between the peritoneum adjacent to the cervical cancer lesion at the preset location in the peritoneal region image of the cervical cancer lesion.

[0125] In the embodiments of this disclosure and other possible embodiments, before performing feature target detection on the peritoneal region image of the cervical cancer lesion using a preset feature target detection model based on deep learning, the method includes: training feature point training images using a preset number of cervical training images and their corresponding cord-like features and / or linear features, training a preset deep learning target detection network, and obtaining a preset feature target detection model constructed based on the deep learning target detection network.

[0126] In the embodiments disclosed herein and other possible embodiments, before training the preset deep learning object detection network, the method includes: using segmentation and annotation software to delineate a preset number of cervical images with rectangular bounding boxes representing linear and / or linear features, thereby obtaining corresponding linear and / or linear feature training images. For example, annotation software such as CasiaLabeler or Labelme can be used to delineate the preset number of cervical training images with rectangular bounding boxes representing linear and / or linear features, thereby obtaining corresponding linear and / or linear feature training images. Those skilled in the art can configure the preset number according to actual needs. For example, the preset number can be configured as 300-500 or other possible values.

[0127] In the embodiments disclosed herein and other possible embodiments, the preset deep learning object detection network may be configured as a YOLO series or an object detection network based on YOLO.

[0128] In this embodiment of the disclosure, determining the proportion of solid cervical cancer lesions includes: segmenting the cervical image using a deep learning-based solid cervical cancer lesion segmentation model to obtain a cervical cancer lesion segmentation mask image and a solid cervical cancer lesion segmentation mask image; and determining the proportion of solid cervical cancer lesions within the cervical cancer lesion based on the cervical cancer lesion segmentation mask image and the solid cervical cancer lesion segmentation mask image.

[0129] In the embodiments of this disclosure and other possible embodiments, the step of segmenting the cervical image using a deep learning-based solid cervical cancer lesion segmentation model to obtain a cervical cancer lesion segmentation mask image and a solid cervical cancer lesion segmentation mask image includes: constructing cervical cancer lesions in each set of cervical training images in a cervical training image set and binary mixed location mask training images corresponding to solid cervical cancer lesions within the cervical cancer lesions; training the preset deep learning segmentation network using the binary mixed location mask training images, the solid cervical cancer lesion edge mask training images corresponding to the solid cervical cancer lesions, and the corresponding cervical training images to obtain a solid cervical cancer lesion segmentation model; and segmenting the cervical image to be segmented based on the solid cervical cancer lesion segmentation model to obtain a cervical cancer lesion segmentation mask image and a solid cervical cancer lesion segmentation mask image. This addresses at least one of the technical problems of the existing technology requiring manual segmentation of solid cervical cancer lesions, which results in a huge workload for doctors and prevents further prediction of surgery and postoperative metastasis.

[0130] In the embodiments disclosed herein and other possible embodiments, unlike the prior art, a binary mixed location mask training image is constructed for each set of cervical training images in the cervical training image set, corresponding to cervical cancer lesions and solid cervical cancer lesions within the cervical cancer lesions. During the training of the preset deep learning segmentation network, the preset deep learning segmentation network pays more attention to the anatomical relationship between the cervical cancer lesions and solid cervical cancer lesions within the cervical cancer lesions, forming a collaborative learning and collaborative response to the cervical cancer lesions and solid cervical cancer lesions within the cervical cancer lesions. This enables effective anatomical segmentation of solid cervical cancer lesions within the cervical cancer lesions, creating anatomical constraints on the segmentation of solid cervical cancer lesions within the cervical cancer lesions.

[0131] In the embodiments disclosed herein and other possible embodiments, when each set of cervical training images and the cervical images to be segmented in the cervical training image set can be configured as two-dimensional cervical images, each set of cervical training images may be one cervical training image corresponding to a subject at a specific moment, or multiple cervical training images corresponding to multiple subjects at multiple specific moments. For example, when the two-dimensional cervical images are configured as abdominal X-ray images or abdominal ultrasound images, each set of cervical training images may be one cervical training image corresponding to a subject at a specific moment corresponding to breath-holding, deep breathing, or deep inspiration, or multiple cervical training images corresponding to a subject at multiple specific moments corresponding to the breathing process or breath-holding process.

[0132] In the embodiments disclosed herein and other possible embodiments, when each set of cervical training images and the cervical images to be segmented in the cervical training image set can be configured as three-dimensional cervical images, each set of cervical training images may be multiple three-dimensional cervical training images corresponding to the examinee at a specific time or multiple four-dimensional cervical training images corresponding to the examinee at multiple specific times. For example, when the three-dimensional cervical images can be configured as one or more of abdominal CT images, abdominal MRI images, abdominal PET images, abdominal CT-PET images, and abdominal MRI-PET images, each set of cervical training images may be multiple three-dimensional cervical training images corresponding to the examinee at a specific time corresponding to breath-holding or deep breathing or deep inhalation, or multiple four-dimensional cervical training images corresponding to the examinee at multiple specific times corresponding to the breathing process or breath-holding process.

[0133] In the embodiments disclosed herein and other possible embodiments, the preset deep learning segmentation network is configured as one or more deep learning segmentation networks such as Fully Convolutional Network (FCN), U-Net, DeepLab, and Pyramid Spatial Pooling Network (PSPNet).

[0134] In the embodiments disclosed herein and other possible embodiments, each set of cervical training images may include multiple cervical cancer lesions, and each cervical cancer lesion may include solid cervical cancer lesions. That is, the multiple cervical cancer lesions in each set of cervical training images may or may not contain solid cervical cancer lesions.

[0135] In the embodiments disclosed herein and other possible embodiments, the construction of cervical cancer lesions and binary mixed location mask training images corresponding to solid cervical cancer lesions within each set of cervical training images in the cervical training image set includes: extracting cervical cancer lesion mask training images with first mask values ​​corresponding to cervical cancer lesions and solid cervical cancer lesion mask training images with second mask values ​​corresponding to solid cervical cancer lesions within the cervical cancer lesions in each set of cervical training images in the cervical training image set; mapping the second mask values ​​to the cervical cancer lesion positions in the solid cervical cancer lesion mask training images according to the positions of solid cervical cancer lesions in the solid cervical cancer lesion mask training images, thereby constructing binary mixed location mask training images.

[0136] In the embodiments of this disclosure and other possible embodiments, determining the solid cervical cancer lesion edge mask training image corresponding to the solid cervical cancer lesion includes: performing an erosion operation on the solid cervical cancer lesion mask training image corresponding to each set of cervical training images in the cervical training image set to obtain a solid cervical cancer lesion mask erosion training image; performing a pixel-level subtraction operation on the solid cervical cancer lesion mask training image and the solid cervical cancer lesion mask erosion training image to determine the solid cervical cancer lesion edge mask training image corresponding to the solid cervical cancer lesion.

[0137] In this disclosure and other possible embodiments, determining the morphology of the cervical cancer lesion includes: determining a cervical cancer lesion mask processing image corresponding to each cervical cancer lesion in a cervical cancer lesion mask image of the cervical image; extracting multiple cervical cancer lesion mask position coordinates corresponding to the cervical cancer lesion mask lines in the cervical cancer lesion mask processing image corresponding to each cervical cancer lesion; and fitting or reconstructing the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion to determine the morphology of the cervical cancer lesion corresponding to each cervical cancer lesion. This addresses at least one of the technical problems arising from the inability to accurately and intelligently determine the morphology of cervical cancer lesions or the excessive workload for doctors caused by manually determining the morphology of cervical cancer lesions, and the inability to further predict cervical cancer metastasis after surgery.

[0138] In the embodiments disclosed herein and other possible embodiments, the step of fitting or reconstructing the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion to determine the cervical cancer lesion morphology includes: under the optimal criterion, approximating multiple preset closed curve equations or multiple preset closed surface equations using the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion to determine the cervical cancer lesion morphology corresponding to each cervical cancer lesion.

[0139] In the embodiments disclosed herein and other possible embodiments, the step of approximating a preset closed curve equation or a preset closed surface equation using the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion under the optimal criterion, and determining the cervical cancer lesion morphology corresponding to each cervical cancer lesion, includes: under the optimal criterion, minimizing the error or absolute residual between the approximation equation and the multiple near-preset closed curve equation or multiple preset closed surface equation when the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion are approximated, and determining the cervical cancer lesion morphology corresponding to each cervical cancer lesion.

[0140] In the embodiments disclosed herein and other possible embodiments, the step of approximating a preset closed curve equation or a preset closed surface equation using the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion under the optimal criterion, and determining the morphology of the cervical cancer lesion corresponding to each cervical cancer lesion, includes: if the cervical image is configured as a two-dimensional abdominal medical image, then under the optimal criterion, approximating a multiple preset closed curve equation using the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion, and determining the morphology of the cervical cancer lesion corresponding to each cervical cancer lesion; if the cervical image is configured as a three-dimensional abdominal medical image, then under the optimal criterion, approximating a multiple preset closed surface equation using the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion, and determining the morphology of the cervical cancer lesion corresponding to each cervical cancer lesion.

[0141] In the embodiments disclosed herein and other possible embodiments, the optimal criterion is configured as one or more of the least squares criterion, the minimum absolute residual criterion, etc.

[0142] In the embodiments disclosed herein and other possible embodiments, the optimal criterion is configured as the least squares criterion. Then, after minimizing the error between the approximation equations and the approximation equations of the cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion when approximating multiple preset closed curve equations or multiple preset closed surface equations, the morphology of multiple cervical cancer lesions corresponding to each cervical cancer lesion is obtained. The cervical cancer lesion mask position coordinate errors corresponding to the multiple cervical cancer lesion morphologies are extracted respectively. The cervical cancer lesion morphology corresponding to the cervical cancer lesion mask position coordinate errors with the smallest error among the multiple cervical cancer lesion morphologies is configured as the final cervical cancer lesion morphology corresponding to each cervical cancer lesion.

[0143] In the embodiments disclosed herein and other possible embodiments, the optimal criterion is configured as the least squares criterion. Then, after minimizing the absolute residuals of the cervical cancer lesion mask position coordinates obtained by approximating multiple preset closed curve equations or multiple preset closed surface equations to the multiple cervical cancer lesion mask position coordinates corresponding to each cervical cancer lesion, the morphology of multiple cervical cancer lesions corresponding to each cervical cancer lesion is obtained. The absolute residuals of the cervical cancer lesion mask position coordinates corresponding to the multiple cervical cancer lesion morphologies are extracted respectively. The cervical cancer lesion morphology corresponding to the cervical cancer lesion mask position coordinates with the smallest error in the absolute residuals of the multiple cervical cancer lesion mask position coordinates is configured as the final cervical cancer lesion morphology corresponding to each cervical cancer lesion.

[0144] In the embodiments disclosed herein and other possible embodiments, the preset closed curve equation is configured as one or more of the two-dimensional closed curves corresponding to a first-order polynomial, a second-order polynomial, or a higher-order polynomial, which are used in polynomial fitting. For example, the preset closed curve equation is configured as a closed curve corresponding to a circle, ellipse, or conic section, a closed curve enclosed by line segments corresponding to a triangle, quadrilateral, pentagon, or hexagon, a spline curve, or a polyline closed curve, etc.

[0145] In the embodiments of this disclosure and other possible embodiments, the preset closed surface equation is configured as a bivariate polynomial or a multivariate polynomial. For example, the preset closed surface equation is configured as a closed surface formed by connecting the beginning and end of line segments, such as a circular surface, elliptical surface, conical surface, triangular surface, quadrilateral surface, pentagonal surface, hexagonal surface, etc., corresponding to a closed surface formed by basic geometric closed surfaces, or a closed surface composed of multiple surfaces.

[0146] In the embodiments disclosed herein and other possible embodiments, determining the cervical cancer lesion mask image includes: performing an erosion operation on the cervical cancer lesion mask image to obtain a cervical cancer lesion mask erosion image; and subtracting the cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain a cervical cancer lesion mask processed image.

[0147] In the embodiments of this disclosure and other possible embodiments, before determining the cervical cancer lesion mask image corresponding to the cervical cancer lesion in the cervical cancer lesion mask image of the cervical image, the method includes: using a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network to segment the cervical image for cervical cancer lesions, and determining the cervical cancer lesion mask image of the cervical image.

[0148] In the embodiments of this disclosure and other possible embodiments, before segmenting cervical cancer lesions in cervical images using a preset segmentation model constructed based on a deep learning segmentation network, the method includes: training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain a preset cervical cancer lesion segmentation model constructed based on the deep learning segmentation network.

[0149] In the embodiments disclosed herein and other possible embodiments, before training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain a preset cervical cancer lesion segmentation model constructed based on the deep learning segmentation network, the method includes: using segmentation annotation software to delineate cervical cancer lesions on a preset number of cervical images to obtain corresponding cervical cancer lesion mask training images.

[0150] In the embodiments of this disclosure and other possible embodiments, before training a preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain a preset segmentation model constructed based on the deep learning segmentation network, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate cervical cancer lesions on the preset number of cervical images to obtain corresponding cervical cancer lesion mask training images. Those skilled in the art can configure the preset number according to actual needs. For example, the preset number can be configured as 300-500.

[0151] In the embodiments disclosed herein and other possible embodiments, the preset deep learning segmentation network is configured as one or more deep learning segmentation networks such as Fully Convolutional Network (FCN), U-Net, DeepLab, and Pyramid Spatial Pooling Network (PSPNet).

[0152] In this disclosed embodiment and other possible embodiments, determining whether the cervical cancer lesion exhibits a spike-like feature involves calculating the ratio of the long axis to the short axis of each cervical cancer lesion in the cervical cancer lesion mask image corresponding to the cervical image. The cervical image is configured as a two-dimensional abdominal medical image. If the ratio of the long axis to the short axis is greater than or equal to a first preset ratio, then the corresponding cervical cancer lesion is determined to have a spike-like feature. This addresses at least one of the technical problems currently present: the need for manual determination of whether a cervical cancer lesion edge exhibits a spike-like feature, which is overly subjective and lacks intelligent methods, resulting in a heavy workload for doctors and hindering subsequent cervical cancer grading and metastasis prediction.

[0153] In this disclosure and other possible embodiments, a method for detecting a spike-like feature is also proposed, comprising: calculating multiple ratios of the long and short diameters of each three-dimensional cervical cancer lesion in a cervical cancer lesion mask image corresponding to the cervical image; if any one of the ratios is greater than or equal to a first preset ratio, then the corresponding cervical cancer lesion is determined to have a spike-like feature; wherein the cervical image is configured as a three-dimensional abdominal medical image. This addresses at least one of the technical problems currently present in the manual determination of whether a spike-like feature exists at the edge of a cervical cancer lesion, including excessive subjectivity, lack of intelligent determination resulting in a huge workload for doctors, and inability to further perform cervical cancer grading and cervical cancer metastasis prediction.

[0154] In embodiments and other possible embodiments of this disclosure, a method for detecting a spike-like feature is also proposed, comprising: determining whether the cervical image is a two-dimensional abdominal medical image or a three-dimensional abdominal medical image; if the cervical image is a two-dimensional abdominal medical image, calculating the ratio of the long axis to the short axis of each cervical cancer lesion in the cervical cancer lesion mask image corresponding to the cervical image; if the ratio of the long axis to the short axis is greater than or equal to a first preset ratio of the long axis to the short axis, determining that the corresponding cervical cancer lesion has a spike-like feature; otherwise, determining that the corresponding cervical cancer lesion does not have a spike-like feature; if the cervical image is a three-dimensional abdominal medical image, calculating multiple ratios of the long axis to the short axis of each three-dimensional cervical cancer lesion in the cervical cancer lesion mask image corresponding to the cervical image; if any one of the ratios of the long axis to the short axis is greater than or equal to the first preset ratio of the long axis to the short axis, determining that the corresponding cervical cancer lesion has a spike-like feature.

[0155] In the embodiments of this disclosure and other possible embodiments, determining the major axis corresponding to each three-dimensional cervical cancer lesion includes: performing three-dimensional reconstruction on each cervical cancer lesion mask in the first cervical cancer lesion mask image to obtain the three-dimensional mask cervical cancer lesion corresponding to each cervical cancer lesion mask; performing edge detection on the three-dimensional mask cervical cancer lesion of each cervical cancer lesion mask to obtain the three-dimensional cervical cancer lesion mask edge; selecting any three-dimensional edge pixel point in the three-dimensional cervical cancer lesion mask edge; calculating multiple three-dimensional edge pixel point distances between the arbitrary three-dimensional edge pixel point and other three-dimensional edge pixel points; selecting the maximum three-dimensional edge pixel point distance from the multiple three-dimensional edge pixel point distances; and configuring the maximum three-dimensional edge pixel point distance as the major axis corresponding to each three-dimensional cervical cancer lesion.

[0156] In the embodiments of this disclosure and other possible embodiments, determining the short axis corresponding to each cervical cancer lesion includes: performing three-dimensional reconstruction on each cervical cancer lesion mask in the first cervical cancer lesion mask image to obtain a three-dimensional mask cervical cancer lesion corresponding to each cervical cancer lesion mask; performing edge detection on the three-dimensional mask cervical cancer lesion of each cervical cancer lesion mask to obtain the three-dimensional cervical cancer lesion mask edge; selecting any three-dimensional edge pixel point in the three-dimensional cervical cancer lesion mask edge; calculating multiple three-dimensional edge pixel point distances between the arbitrary three-dimensional edge pixel point and other three-dimensional edge pixel points; selecting the minimum three-dimensional edge pixel point distance from the multiple three-dimensional edge pixel point distances; configuring the minimum three-dimensional edge pixel point distance as the long axis corresponding to each three-dimensional lesion; if the ratio of the long axis to the short axis is less than the first preset ratio of the long axis to the short axis, then it is determined that the corresponding cervical cancer lesion does not have a spike sign.

[0157] In this embodiment and other possible embodiments, determining the cervical cancer lesion mask image corresponding to the cervical image includes: training a first preset deep learning segmentation network using a preset number of cervical training images and their corresponding cervical cancer lesion mask training images to obtain a preset cervical cancer lesion segmentation model; segmenting the cervical cancer lesions in the cervical image using the preset cervical cancer lesion segmentation model to obtain the corresponding cervical cancer lesion mask image; and segmenting the arteries using the preset artery segmentation model to obtain the corresponding artery segmentation mask image. The cervical cancer lesions in the cervical image are configured as cervical cancer lesions awaiting chemotherapy or cervical cancer lesions awaiting surgery.

[0158] In the embodiments disclosed herein and other possible embodiments, the mask value corresponding to each cervical cancer lesion region in the cervical cancer lesion mask image and the cervical cancer lesion mask training image is configured as 1, and the mask value corresponding to non-cervical cancer lesion regions is configured as 0.

[0159] In the embodiments disclosed herein and other possible embodiments, before training the first preset deep learning segmentation network using a preset number of cervical training images and their corresponding cervical cancer lesion mask training images, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate cervical cancer lesions on the preset number of cervical training images to obtain corresponding cervical cancer lesion mask training images.

[0160] In this embodiment of the disclosure, if the cervical cancer lesion has a spike pattern, then based on the cervical cancer lesion mask image and arterial vessel mask image corresponding to the cervical image, it is determined whether the spike corresponding to the cervical cancer lesion is adjacent to an arterial vessel; if the spike corresponding to the cervical cancer lesion is adjacent to an arterial vessel, then the spike is determined to be a dangerous spike.

[0161] In this embodiment of the disclosure, if the cervical cancer lesion has a spike pattern, then based on the cervical cancer lesion mask and arterial vessel mask image corresponding to the cervical image, it is determined whether the spike corresponding to the cervical cancer lesion is adjacent to an arterial vessel; if the spike corresponding to the cervical cancer lesion is adjacent to an arterial vessel, then the spike is determined to be a dangerous spike.

[0162] In this embodiment of the disclosure, determining whether the spinous process corresponding to the cervical cancer lesion is adjacent to an artery based on the cervical cancer lesion mask image and the artery mask image corresponding to the cervical image includes: extracting multiple spinous process position coordinates corresponding to multiple spinous processes in the cervical cancer lesion mask image and artery mask edge position coordinates corresponding to the artery mask edge in the artery mask image; calculating multiple distances between the multiple spinous process position coordinates and the artery mask edge position coordinates; if the minimum distance among the multiple distances is less than a preset distance, then it is determined that the spinous process corresponding to the cervical cancer lesion is adjacent to an artery.

[0163] In this embodiment of the disclosure, determining whether the spinous process corresponding to the cervical cancer lesion is adjacent to an artery based on the cervical cancer lesion mask and artery mask image corresponding to the cervical image includes: extracting multiple spinous process position coordinates corresponding to multiple spinous processes and artery mask edge position coordinates corresponding to the artery mask edge in the cervical cancer lesion mask and artery mask images; calculating multiple distances between the multiple spinous process position coordinates and the artery mask edge position coordinates respectively; if the minimum distance among the multiple distances is less than a preset distance, then it is determined that the spinous process corresponding to the cervical cancer lesion is adjacent to an artery.

[0164] In the embodiments of this disclosure and other possible embodiments, determining the spike includes: detecting the edge of the cervical cancer lesion mask in the cervical cancer lesion mask image to obtain the edge line of the cervical cancer lesion mask; calculating the first pixel distance and the second pixel distance between any cervical cancer lesion mask edge pixel on the edge line of the cervical cancer lesion mask and its adjacent cervical cancer lesion mask edge pixels respectively; if the first pixel distance and the second pixel distance are greater than a set pixel distance, then the arbitrary cervical cancer lesion mask edge pixel is configured as a spike.

[0165] In this disclosure and other possible embodiments, determining whether the location of the cervical cancer lesion to be treated or operated on, based on the cervical cancer lesion mask image and the arterial vessel mask image, is adjacent to an arterial vessel includes: calculating multiple positional distances between the arterial vessel edge mask position coordinates corresponding to the arterial vessel mask image and the cervical cancer lesion edge mask position coordinates corresponding to the cervical cancer lesion mask image; if the minimum positional distance among the multiple positional distances is less than or equal to a preset positional distance, then the cervical cancer lesion is determined to be adjacent to an arterial vessel; otherwise, the cervical cancer lesion is determined to be far from an arterial vessel. The arterial vessel edge mask position coordinates and the cervical cancer lesion edge mask position coordinates are respectively configured as the position coordinates of the arterial vessel edge in the arterial vessel edge mask image corresponding to the arterial vessel mask image and the position coordinates of the cervical cancer lesion edge mask in the cervical cancer lesion edge mask image corresponding to the cervical cancer lesion mask image.

[0166] In this disclosure and other possible embodiments, determining whether the location of the cervical cancer lesion to be treated or operated on, based on the cervical cancer lesion mask and arterial vessel mask images, is adjacent to an artery includes: calculating multiple positional distances between the arterial vessel edge mask position coordinates and the cervical cancer lesion edge mask position coordinates corresponding to the cervical cancer lesion mask and arterial vessel mask images; if the minimum positional distance among the multiple positional distances is less than or equal to a preset positional distance, then the cervical cancer lesion is determined to be adjacent to an arterial vessel; otherwise, the cervical cancer lesion is determined to be far from an arterial vessel. The arterial vessel edge mask position coordinates and the cervical cancer lesion edge mask position coordinates are respectively configured as the position coordinates of the arterial vessel edge in the arterial vessel edge mask and the position coordinates of the cervical cancer lesion edge mask in the cervical cancer lesion mask and arterial vessel mask images.

[0167] In the embodiments disclosed herein and other possible embodiments, those skilled in the art can configure the preset position distance according to actual needs. For example, the preset position distance can be any value between 0 and 10 pixels or other values.

[0168] In the embodiments disclosed herein and other possible embodiments, the location of cervical cancer lesions and their proximity to arteries are of significant clinical importance for indicating disease type, assisting in differential diagnosis, and facilitating subsequent chemotherapy and surgery. It is necessary to propose a technical solution for determining the location of cervical cancer lesions and for intelligent processing of abdominal medical images, in order to address at least one of the following technical problems: the current reliance on manual determination of the location of cervical cancer lesions and their proximity to arteries leads to excessive subjectivity; the lack of intelligent determination results in a heavy workload for doctors; and it prevents further cervical cancer grading and metastasis prediction.

[0169] In the embodiments disclosed herein and other possible embodiments, the step of using cervical images to determine the cervical cancer lesion mask image and arterial vessel mask image corresponding to the cervical cancer lesion to be treated with chemotherapy or to be operated on includes: training a first preset deep learning segmentation network using a preset number of cervical training images and their corresponding cervical cancer lesion mask training images to obtain a preset cervical cancer lesion segmentation model; training a second preset deep learning segmentation network using a preset number of cervical training images and their corresponding arterial vessel mask training images to obtain a preset arterial vessel segmentation model; segmenting the cervical cancer lesion to be treated with chemotherapy or to be operated on in the cervical image using the preset cervical cancer lesion segmentation model to obtain the corresponding cervical cancer lesion mask image; and segmenting the arterial vessels using the preset arterial vessel segmentation model to obtain the corresponding arterial vessel mask image.

[0170] In the embodiments disclosed herein and other possible embodiments, the step of using cervical images to determine the cervical cancer lesion mask and arterial vessel mask images corresponding to the cervical cancer lesions to be treated with chemotherapy or to be operated on includes: training a preset deep learning segmentation network using a preset number of cervical training images and their corresponding cervical cancer lesion mask and arterial vessel mask training images to obtain a preset cervical cancer lesion and arterial vessel segmentation model; and segmenting the cervical cancer lesion to be treated with chemotherapy or the cervical cancer lesion and arterial vessels to be operated on using the preset cervical cancer lesion and arterial vessel segmentation model to obtain the corresponding cervical cancer lesion mask and arterial vessel mask images.

[0171] In the embodiments disclosed herein and other possible embodiments, the mask value corresponding to each cervical cancer lesion region in the cervical cancer lesion mask image and the cervical cancer lesion mask training image is configured as 1, and the mask value corresponding to the non-cervical cancer lesion region is configured as 0; similarly, the mask value corresponding to the arterial blood vessel in the arterial blood vessel mask image and the arterial blood vessel mask training image is configured as 1, and the mask value corresponding to the non-arterial blood vessel region is configured as 0.

[0172] In the embodiments disclosed herein and other possible embodiments, the mask value corresponding to each cervical cancer lesion region in the cervical cancer lesion mask and artery mask image and the cervical cancer lesion mask and artery mask training image is configured as 1, the mask value corresponding to the artery is configured as 2, and the mask value corresponding to the non-cervical cancer lesion region and the non-artery region is configured as 0.

[0173] In the embodiments disclosed herein and other possible embodiments, before training the first preset deep learning segmentation network using a preset number of cervical training images and their corresponding cervical cancer lesion mask training images, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate cervical cancer lesions on the preset number of cervical training images to obtain corresponding cervical cancer lesion mask training images.

[0174] In the embodiments disclosed herein and other possible embodiments, before training the second preset deep learning segmentation network using a preset number of cervical training images and their corresponding arterial mask training images, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate the arteries in the preset number of cervical training images to obtain the corresponding arterial mask training images.

[0175] In the embodiments disclosed herein and other possible embodiments, before training the preset deep learning segmentation network using a preset number of cervical training images and their corresponding cervical cancer lesion masks and arterial vessel masks, the method includes: using segmentation annotation software CasiaLabeler or Labelme to delineate cervical cancer lesions and arterial vessels on the preset number of cervical training images to obtain corresponding cervical cancer lesion masks and arterial vessel mask training images.

[0176] In the embodiments disclosed herein and other possible embodiments, the preset deep learning segmentation network, the first preset deep learning segmentation network and the second preset deep learning segmentation network are configured as one or more deep learning segmentation networks such as Fully Convolutional Network (FCN), U-Net, DeepLab, Pyramid Spatial Pooling Network (PSPNet).

[0177] In the embodiments disclosed herein and other possible embodiments, if the edge of the cervical cancer lesion has a spike-like sign, determining whether the cervical cancer lesion in the cervical image is a cervical cancer lesion to be treated with chemotherapy or a cervical cancer lesion to be operated on includes: detecting whether the boundary of the cervical cancer lesion in the cervical cancer lesion mask image corresponding to the cervical image is blurred; if the boundary of the cervical cancer lesion in the cervical cancer lesion mask image is blurred, then configuring the cervical cancer lesion as a cervical cancer lesion to be treated with chemotherapy or a cervical cancer lesion to be operated on.

[0178] In the embodiments disclosed herein and other possible embodiments, if the edge of the cervical cancer lesion has a spike-like feature, determining whether the cervical cancer lesion in the cervical image is a cervical cancer lesion awaiting chemotherapy or a cervical cancer lesion awaiting surgery includes: detecting whether the boundary of the cervical cancer lesion in the cervical image corresponding to the cervical image is blurred and whether the edge of the cervical cancer lesion has a spike-like feature; if the boundary of the cervical cancer lesion in the cervical image and the arterial mask image is blurred and the edge of the cervical cancer lesion has a spike-like feature, then the cervical cancer lesion is configured as a cervical cancer lesion awaiting chemotherapy or a cervical cancer lesion awaiting surgery.

[0179] In this disclosure and other possible embodiments, determining whether the cervical cancer lesion exhibits vacuole characteristics includes: performing vacuole characteristic target detection on each cervical cancer lesion in the cervical image; configuring cervical cancer lesions corresponding to detected vacuole characteristics as having vacuole characteristics; extracting the average pixel values ​​of the edges of cervical cancer lesions for which no vacuole characteristics were detected; if the average pixel value corresponding to the cervical cancer lesions for which no vacuole characteristics were detected is less than the corresponding average pixel value of the cervical cancer lesion edge, then it is determined that the cervical cancer lesions for which no vacuole characteristics were detected have vacuole characteristics; otherwise, it is determined that the cervical cancer lesions for which no vacuole characteristics were detected do not have vacuole characteristics. This addresses at least one of the technical problems of existing methods that require manual determination of the presence of vacuole characteristics in cervical cancer lesions, which are too subjective and lack intelligent determination, resulting in a huge workload for doctors and hindering further surgical and postoperative prediction.

[0180] In the embodiments and other possible embodiments of this disclosure, unlike the prior art, the technical solution proposed in this disclosure, by performing target detection of cervical cancer lesions with vacuolar signs on each cervical cancer lesion in the cervical image, and using the average pixel value of the edge of the cervical cancer lesion corresponding to which vacuolar signs were not detected, further processing of the cervical cancer lesions for which vacuolar signs were not detected, can effectively solve the technical problem of missed detection of cervical cancer lesions with vacuolar signs. Utilizing the average pixel value of the edge of the cervical cancer lesion corresponding to which vacuolar signs were not detected improves the robustness of the method for determining whether cervical cancer lesions with vacuolar signs exist to a certain extent. At the same time, it avoids at least one of the technical problems of existing methods that require manual determination of the presence of vacuolar signs, which are too subjective and lack intelligent determination, resulting in a huge workload for doctors and preventing further technical analysis such as cervical cancer grading and cervical cancer metastasis prediction.

[0181] In the embodiments disclosed herein and other possible embodiments, the step of detecting vacuolar signs of cervical cancer lesions in each cervical cancer lesion in the cervical image includes: using a deep learning-based vacuolar sign detection model to detect vacuolar signs of cervical cancer lesions in each cervical cancer lesion in the cervical image.

[0182] In the embodiments of this disclosure and other possible embodiments, before performing cervical cancer lesion vacuolation target detection on each cervical cancer lesion in the cervical image using a deep learning-based cervical cancer lesion vacuolation target detection model, the method includes: training a preset deep learning target detection network to obtain a deep learning-based cervical cancer lesion vacuolation target detection model.

[0183] In the embodiments disclosed herein and other possible embodiments, the step of training a preset deep learning target detection network to obtain a deep learning-based target detection model for cervical cancer lesions with vacuolar signs includes: training the preset deep learning target detection network using a preset number of cervical training images and their corresponding vacuolar sign target training images to obtain a deep learning-based target detection model for cervical cancer lesions with vacuolar signs.

[0184] In the embodiments disclosed herein and other possible embodiments, before training the preset deep learning target detection network, the method includes: using segmentation annotation software to delineate the vacuolar signs of cervical cancer lesions using rectangular bounding boxes on a preset number of cervical images, thereby obtaining corresponding training images of the vacuolar signs. For example, annotation software such as CasiaLabeler or Labelme can be used to delineate the vacuolar signs of cervical cancer lesions on a preset number of cervical images, thereby obtaining corresponding training images of the vacuolar signs. Those skilled in the art can configure the preset number according to actual needs. For example, the preset number can be configured as 300-500 or other possible values.

[0185] In the embodiments disclosed herein and other possible embodiments, the preset deep learning object detection network may be configured as a YOLO series or an object detection network based on YOLO.

[0186] In the embodiments of this disclosure and other possible embodiments, determining the total number of cervical cancer lesions corresponding to the cervical image includes: segmenting the cervical cancer lesions or detecting cervical cancer lesions in the cervical image to determine the total number of cervical cancer lesions corresponding to the cervical image. Specifically, segmenting the cervical cancer lesions in the cervical image includes: segmenting the cervical cancer lesions in the cervical image using a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network. Detecting cervical cancer lesions in the cervical image includes: detecting cervical cancer lesions in the cervical image using a deep learning-based cervical cancer lesion target detection model.

[0187] In the embodiments of this disclosure and other possible embodiments, a preset cervical cancer lesion segmentation model based on a deep learning segmentation network is used to segment the cervical image into cervical cancer lesions to obtain a corresponding cervical cancer lesion mask image; the number of cervical cancer lesion masks corresponding to the cervical cancer lesion mask image is counted to determine the total number of cervical cancer lesions corresponding to the cervical image.

[0188] In the embodiments of this disclosure and other possible embodiments, a cervical cancer lesion target detection model based on deep learning is used to detect cervical cancer lesions in the cervical image to obtain a corresponding cervical cancer lesion detection image; the number of cervical cancer lesion detection boxes corresponding to the cervical cancer lesion detection image is counted to determine the total number of cervical cancer lesions corresponding to the cervical image.

[0189] In the embodiments of this disclosure and other possible embodiments, before segmenting the cervical images using a preset cervical cancer lesion segmentation model constructed based on a deep learning segmentation network, the process includes: training the preset deep learning segmentation network using a preset number of cervical images and their corresponding cervical cancer lesion mask training images to obtain the preset cervical cancer lesion segmentation model constructed based on the deep learning segmentation network.

[0190] In the embodiments of this disclosure and other possible embodiments, before training the preset deep learning segmentation network, the method includes: using segmentation annotation software to delineate cervical cancer lesions on a preset number of cervical images to obtain corresponding cervical cancer lesion mask training images.

[0191] In the embodiments of this disclosure and other possible embodiments, the step of extracting the edge pixel values ​​of cervical cancer lesions corresponding to cervical cancer lesions for which no vacuolar signs were detected includes: performing edge detection on the cervical cancer lesion mask image corresponding to the cervical cancer lesion for which no vacuolar signs were detected to obtain a cervical cancer lesion edge image; performing pixel-level multiplication operations on the cervical cancer lesion edge image and its corresponding cervical image to obtain a cervical cancer lesion edge image; and extracting the pixel values ​​corresponding to the edges of cervical cancer lesions within the cervical cancer lesion edge image to obtain the corresponding cervical cancer lesion edge pixel values.

[0192] In the embodiments of this disclosure and other possible embodiments, the step of performing edge detection on the cervical cancer lesion mask image corresponding to the cervical cancer lesion for which no vacuolar sign was detected, to obtain the cervical cancer lesion edge image, includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical cancer lesion for which no vacuolar sign was detected, to obtain a cervical cancer lesion mask erosion image; and subtracting the corresponding cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain the cervical cancer lesion edge image.

[0193] In the embodiments of this disclosure and other possible embodiments, the step of extracting the average pixel value of the edge of the cervical cancer lesion corresponding to the cervical cancer lesion for which no vacuolar sign was detected includes: performing edge detection on the cervical cancer lesion mask image corresponding to the cervical cancer lesion for which no vacuolar sign was detected to obtain the edge image of the cervical cancer lesion; performing pixel-level multiplication on the edge image of the cervical cancer lesion and its corresponding cervical image to obtain the edge image of the cervical cancer lesion; extracting the pixel value corresponding to the edge of the cervical cancer lesion in the edge image of the cervical cancer lesion to obtain the corresponding edge pixel value of the cervical cancer lesion; and calculating the average value of the edge pixel values ​​of the cervical cancer lesion to obtain the average pixel value of the edge of the cervical cancer lesion.

[0194] In the embodiments of this disclosure and other possible embodiments, the step of performing edge detection on the cervical cancer lesion mask image corresponding to the cervical cancer lesion for which no vacuolar sign was detected, to obtain the cervical cancer lesion edge image, includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical cancer lesion for which no vacuolar sign was detected, to obtain a cervical cancer lesion mask erosion image; and subtracting the corresponding cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain the cervical cancer lesion edge image.

[0195] In the embodiments of this disclosure and other possible embodiments, determining whether cervical cancer lesions without detected vacuolar signs exist based on the average pixel value corresponding to the cervical cancer lesions without detected vacuolar signs and the average pixel value of the corresponding edge of the cervical cancer lesions includes: if the average pixel value corresponding to the cervical cancer lesions without detected vacuolar signs is less than the average pixel value of the corresponding edge of the cervical cancer lesions, then it is determined that cervical cancer lesions without detected vacuolar signs exist; otherwise, it is determined that cervical cancer lesions without detected vacuolar signs do not exist.

[0196] In this embodiment, cervical cancer grading is performed using the quantitative features of cervical cancer lesions, including: determining the grading model corresponding to each preset classifier in the preset classifier pool corresponding to the cervical image training set, and the sub-grading model corresponding to each preset classifier in the sub-training set of the cervical image training set; determining the sub-grading probability value set on the sub-validation set using the sub-grading model and the sub-validation set of the cervical image training set; determining the grading probability value set on the validation set using the grading model and the cervical image validation set corresponding to the cervical image training set; and optimizing the grading probability value set using the fusion weights corresponding to the sub-grading probability value set to obtain the corresponding fusion probability set for cervical cancer grading, thus completing the cervical cancer grading. This addresses the technical problem of difficulty in improving cervical cancer grading performance with minimal human intervention.

[0197] In the embodiments disclosed herein and other possible embodiments, the features within the cervical image training set, the cervical image validation set, the sub-training set, and the sub-validation set are configured as quantitative features of cervical cancer lesions. For a supervised, preset deep learning classification network or a preset machine learning classifier, the features within the cervical image training set, the cervical image validation set, the sub-training set, and the sub-validation set are configured as quantitative features of cervical cancer lesions and corresponding cervical cancer grading labels.

[0198] In this embodiment of the disclosure, before training the preset deep learning classification network or the preset machine learning classifier, the method includes: encoding the corresponding classification features of the quantitative features of the cervical cancer lesions to obtain coded quantitative features of the cervical cancer lesions; standardizing or normalizing the corresponding values ​​of the quantitative features of the cervical cancer lesions to obtain processed quantitative features of the cervical cancer lesions; and using the coded quantitative features of the cervical cancer lesions, the processed quantitative features of the cervical cancer lesions, and the cervical cancer pathology result labels to train the preset deep learning classification network or the preset machine learning classifier to obtain a preset early cervical cancer screening model.

[0199] For example, in encoding the classification features (classification features) corresponding to whether the boundary of the cervical cancer lesion is clear, whether the cervical cancer lesion has a spike-like sign, whether the cervical cancer lesion has a lobulated sign, and whether the cervical cancer lesion has a vacuolar sign, the clear boundary of the cervical cancer lesion, the presence of a spike-like sign, the presence of a lobulated sign, and the presence of a vacuolar sign are configured as 1, and the unclear boundary of the cervical cancer lesion, the absence of a spike-like sign, the absence of a lobulated sign, and the absence of a vacuolar sign are configured as 0. As another example, the location of the cervical cancer lesion on the cervix can be configured as different positional encoding features corresponding to a first numerical queue (0, 1, 2, ... or 0, 2, 4, ...); the morphology of the cervical cancer lesion can be configured as different morphological encoding features corresponding to a second numerical queue (0, 1, 2, ... or 1, 3, 5, ...). The cervical cancer grading corresponding to the pathological results of cervical cancer grading can be configured as different grading coding features corresponding to the first numerical queue (0, 1, 2, ...).

[0200] In the embodiments of this disclosure and other possible embodiments, the sub-grading probability value set corresponding to each preset grader on the sub-validation set is determined using the corresponding sub-grading model of each preset grader and the sub-validation set of the cervical image training set. Similarly, the grading probability value set corresponding to each preset grader on the validation set is determined using the corresponding grading model of each preset grader and the cervical image validation set corresponding to the cervical image training set.

[0201] In this disclosure and other possible embodiments, constructing the preset classifier pool includes: obtaining a preset classifier with a set number ≥ 2; and constructing the preset classifier pool using the preset classifier with the set number ≥ 2. Simultaneously, constructing a preset clinical guidance pool includes: obtaining preset clinical guidance corresponding to the area under the subject operating characteristic curve and the coordinate axes, recall, and precision; and constructing the preset clinical guidance pool using the preset clinical guidance corresponding to the area under the subject operating characteristic curve and the coordinate axes. The preset classifiers can be configured as deep learning-based classification networks corresponding to CNN, RNN, and GCN, random forest (RF), gradient boosting decision tree (GBDT), K-nearest neighbors (KNN), AdaBoost (Ada), and machine learning-based classifiers corresponding to linear discriminant analysis (LDA).

[0202] In the embodiments of this disclosure and other possible embodiments, before determining the corresponding hierarchical model of each preset classifier in the preset classifier pool corresponding to the cervical image training set and the corresponding sub-hierarchical model of each preset classifier in the sub-training set of the cervical image training set, the method includes: initializing the list of fusion weights to be processed corresponding to each preset clinical guide in the preset clinical guide pool, initializing the list of probability value sets corresponding to the sub-hierarchical probability value set and the hierarchical probability value set, initializing the fusion probability set corresponding to each preset clinical guide, initializing the list of fusion weights to be processed corresponding to each preset clinical guide in the preset clinical guide pool, initializing the list of fusion probability value sets corresponding to each preset clinical guide, and initializing the list of probability value sets corresponding to the sub-hierarchical probability value set and the hierarchical probability value set.

[0203] In the embodiments of this disclosure and other possible embodiments, the initialization of the list of fusion weights to be processed corresponding to each preset clinical guide in the preset clinical guide pool includes: initializing the list of fusion weights to be processed corresponding to each preset clinical guide in the preset clinical guide pool.

[0204] In the embodiments disclosed herein and other possible embodiments, the cervical image set is divided according to a first preset ratio corresponding to the number of folds in cross-validation, resulting in a cervical image training set and a cervical image validation set. Simultaneously, the cervical image training set is divided according to a second preset ratio corresponding to stratified sampling, resulting in a sub-training set and a sub-validation set. Those skilled in the art can configure the first preset ratio and the second preset ratio according to actual needs. For example, the first preset ratio can be configured as 4:1, and the second preset ratio as 9:1.

[0205] According to the first preset ratio corresponding to the number of folds in cross-validation, the quantitative features of cervical cancer lesions and their corresponding pathological grading results corresponding to the cervical image set are divided to obtain a cervical image training set and a cervical image validation set. According to the second preset ratio corresponding to stratified sampling, the cervical image training set is divided to obtain a sub-training set and a sub-validation set.

[0206] The step of determining the hierarchical model corresponding to each preset classifier in the preset classifier pool corresponding to the cervical image training set and the sub-hierarchical model corresponding to each preset classifier in the sub-training set of the cervical image training set includes: determining the hierarchical model corresponding to each preset classifier in the preset classifier pool corresponding to the cervical image training set and the multiple sub-hierarchical models corresponding to each preset classifier in the preset classifier pool in the sub-training set of the cervical image training set.

[0207] Determining the corresponding sub-grading model for each preset classifier in the preset classifier pool corresponding to a sub-training set of the cervical image training set includes: using the quantitative features of cervical cancer lesions corresponding to the sub-training set of the cervical image training set, and training each preset classifier in the preset classifier pool using the quantitative features of cervical cancer lesions and their corresponding pathological grading results to obtain the corresponding sub-grading model.

[0208] In this disclosure and other possible embodiments, the method further includes: dividing the cervical image set according to a first preset ratio to obtain a cervical image training set and a cervical image validation set. It also includes: dividing the cervical image training set according to a second preset ratio to obtain a sub-training set and a sub-validation set; the preset classifier is configured as a deep learning-based hierarchical network or a machine learning-based classifier.

[0209] For example, in embodiments of this disclosure and other possible embodiments, the preset classifier may be configured as a deep learning-based classifier network (e.g., CNN, RNN, GCN, etc.) or a machine learning-based classifier (e.g., decision tree, Naive Bayes, support vector machine, multilayer perceptron, decision tree, random forest, gradient boosting tree, etc., one or more).

[0210] In this disclosure and other possible embodiments, determining the fusion weight list corresponding to the sub-level probability value set includes: using a preset weight function to perform weight processing on the fusion weight list to be processed corresponding to the sub-level probability value set to obtain the corresponding fusion weight list. The preset weight function is configured as one or more of a linear weight function, an exponential weight function, a power weight function, and a classifier ranking weight function.

[0211] In the embodiments of this disclosure and other possible embodiments, determining the list of fusion weights to be processed includes: calculating the area, recall, and precision under the subject operating characteristic curve corresponding to the preset clinical guidance and the coordinate axis using the sub-prediction probability value set respectively; and configuring the area, recall, and precision under the subject operating characteristic curve and the coordinate axis as the list of fusion weights to be processed.

[0212] Applying the linear weight function to the list of fusion weights to be processed includes: summing the fusion weights in the list to be processed to obtain a first weight sum; dividing each fusion weight by the first weight sum to obtain a corresponding fusion weight list. Applying the power weight function to the list of fusion weights to be processed includes: using a first adjustment parameter (configured to 2 or other possible values), exponentializing each fusion weight in the list to be processed (the first adjustment parameter is the power) to obtain processed fusion power weights; summing the processed fusion power weights to obtain a second weight sum; dividing each processed fusion power weight by the second weight sum to obtain a corresponding fusion weight list. The process of applying the exponential weight function to the list of fusion weights to be processed includes: multiplying the fusion weights in the list of fusion weights to be processed using a second adjustment parameter (configured to 5 or other possible values) to obtain adjusted fusion weights; exponentially processing the adjusted fusion weights (using the adjusted fusion weights as powers of the exponent) to obtain exponential weights; summing the exponential weights to obtain a third weight sum; and dividing the exponential weights by the third weight sum to obtain the corresponding fusion weight list. The process of applying the prediction model ranking weight function to the list of fusion weights to be processed includes: determining the ranking of the sub-prediction model corresponding to each predictor based on the area enclosed by the receiver operating characteristic curve and the coordinate axis, recall, and precision; summing the rankings corresponding to each preset clinical guidance to obtain a fourth weight sum; and dividing the rankings corresponding to each preset clinical guidance by the corresponding fourth weight sum to obtain the corresponding fusion weight list.

[0213] In the embodiments of this disclosure and other possible embodiments, the step of determining the corresponding grading model of each preset grader in the preset grader pool corresponding to the cervical image training set and the corresponding sub-grading model of each preset grader in the preset grader pool corresponding to the sub-training set of the cervical image training set includes: determining the corresponding grading model of each preset grader in the preset grader pool corresponding to the cervical image training set and the corresponding multiple sub-grading models of each preset grader in the preset grader pool corresponding to the sub-training set of the cervical image training set.

[0214] In the embodiments of this disclosure and other possible embodiments, determining the corresponding grading model of each preset grader in the preset grader pool corresponding to the cervical image training set includes: extracting features from the cervical image training set to obtain the corresponding quantitative features of cervical cancer lesions, and training each preset grader in the preset grader pool using the quantitative features of cervical cancer lesions and their cervical cancer pathological grading result labels to obtain multiple corresponding grading models.

[0215] In the embodiments of this disclosure and other possible embodiments, determining the corresponding sub-grading model of each preset grader in the preset grader pool corresponding to the sub-training set of the cervical image training set includes: using the quantitative features of cervical cancer lesions corresponding to the sub-training set of the cervical image training set, and training each preset grader in the preset grader pool using the quantitative features of cervical cancer lesions and their cervical cancer pathological grading result labels to obtain the corresponding sub-grading model.

[0216] In this embodiment, cervical cancer metastasis prediction is performed using the quantitative features of cervical cancer lesions, including: determining the prediction model corresponding to each preset predictor in a preset predictor pool corresponding to the cervical image training set, and the sub-prediction model corresponding to each preset predictor in a sub-training set of the cervical image training set; determining a sub-prediction probability value set on the sub-validation set using the sub-prediction model and the sub-validation set of the cervical image training set; determining a prediction probability value set on the validation set using the prediction model and the cervical image validation set corresponding to the cervical image training set; and optimizing the prediction probability value set using the fusion weights corresponding to the sub-prediction probability value set to obtain a corresponding fusion probability set for cervical cancer prediction, thus completing the cervical cancer metastasis prediction. This addresses the technical problem of difficulty in improving the performance of cervical cancer metastasis prediction with minimal human intervention.

[0217] In the embodiments disclosed herein and other possible embodiments, the features within the cervical image training set, the cervical image validation set, the sub-training set, and the sub-validation set are configured as quantitative features of cervical cancer lesions. For a supervised, preset deep learning prediction network or a preset machine learning predictor, the features within the cervical image training set, the cervical image validation set, the sub-training set, and the sub-validation set are configured as quantitative features of cervical cancer lesions and corresponding cervical cancer metastasis prediction labels.

[0218] Similar to the cervical cancer grading described above, in this embodiment and other possible embodiments, the sub-prediction probability value set corresponding to each preset predictor on the sub-validation set is determined using the corresponding sub-prediction model of each preset predictor and the sub-validation set of the cervical image training set. Similarly, the prediction probability value set corresponding to each preset predictor on the validation set is determined using the corresponding prediction model of each preset predictor and the cervical image validation set corresponding to the cervical image training set. Furthermore, the postoperative pathological result indicating cervical cancer metastasis is configured as 1 if cervical cancer metastasis is present, and as 0 if cervical cancer metastasis is not present.

[0219] Similar to the cervical cancer grading described above, in this disclosure and other possible embodiments, constructing the preset predictor pool includes: obtaining a preset predictor pool of at least two numbers; and constructing the preset predictor pool using the preset predictor pool of at least two numbers. Simultaneously, constructing a preset clinical guidance pool includes: obtaining preset clinical guidance corresponding to the area under the receiver operating characteristic curve (ROC) and the coordinate axes, recall, and precision; and constructing the preset clinical guidance pool using the preset clinical guidance corresponding to the area under the ROC curve and the coordinate axes. The preset predictor can be configured as a deep learning-based prediction network corresponding to CNN, RNN, GCN, random forest (RF), gradient boosting decision tree (GBDT), K-nearest neighbors (KNN), AdaBoost (Ada), and a machine learning-based predictor corresponding to linear discriminant analysis (LDA).

[0220] Similar to the cervical cancer grading described above, in the embodiments of this disclosure and other possible embodiments, before determining the prediction model corresponding to each preset predictor in the preset predictor pool corresponding to the cervical image training set and the sub-prediction model corresponding to each preset predictor in the sub-training set of the cervical image training set, the method includes: initializing the list of unprocessed fusion weights corresponding to each preset clinical guide in the preset clinical guide pool, initializing the list of probability value sets corresponding to the sub-prediction probability value set and the prediction probability value set, initializing the fusion probability set corresponding to each preset clinical guide, initializing the list of unprocessed fusion weights corresponding to each preset clinical guide in the preset clinical guide pool, initializing the list of fusion probability value sets corresponding to each preset clinical guide, and initializing the list of probability value sets corresponding to the sub-prediction probability value set and the prediction probability value set.

[0221] Similar to the cervical cancer grading described above, in the embodiments of this disclosure and other possible embodiments, the initialization of the list of fusion weights to be processed corresponding to each preset clinical guide in the preset clinical guide pool includes: initializing the fusion weights to be processed corresponding to each preset clinical guide in the preset clinical guide pool.

[0222] Similar to the cervical cancer grading described above, in this embodiment and other possible embodiments, the cervical image set is divided according to a first preset ratio corresponding to the cross-validation fold number to obtain a cervical image training set and a cervical image validation set. Simultaneously, the cervical image training set is divided according to a second preset ratio corresponding to stratified sampling to obtain a sub-training set and a sub-validation set. Those skilled in the art can configure the first preset ratio and the second preset ratio according to actual needs. For example, the first preset ratio can be configured as 4:1, and the second preset ratio as 9:1.

[0223] Similar to the cervical cancer grading described above, the quantitative features of cervical cancer lesions and their corresponding pathological prediction results corresponding to the cross-validation folds are divided according to a first preset ratio to obtain a cervical image training set and a cervical image validation set. The cervical image training set is then divided according to a second preset ratio corresponding to stratified sampling to obtain a sub-training set and a sub-validation set.

[0224] Similar to the above-mentioned cervical cancer grading, the step of determining the prediction model corresponding to each preset predictor in the preset predictor pool corresponding to the cervical image training set and the sub-prediction model corresponding to each preset predictor in the sub-training set of the cervical image training set includes: determining the prediction model corresponding to each preset predictor in the preset predictor pool corresponding to the preoperative cervical image training set and the multiple sub-prediction models corresponding to each preset predictor in the preset predictor pool corresponding to the sub-training set of the preoperative cervical image training set.

[0225] Similar to the cervical cancer grading method described above, determining the corresponding sub-prediction model for each preset predictor in the preset predictor pool corresponding to a sub-training set of the preoperatively acquired cervical image training set includes: quantifying the cervical cancer lesions corresponding to the sub-training set of the preoperatively acquired cervical image training set, and training each preset predictor in the preset predictor pool using the quantitative features of the cervical cancer lesions and their corresponding pathological prediction results to obtain the corresponding sub-prediction model.

[0226] Similar to the cervical cancer grading described above, in this embodiment and other possible embodiments, the method further includes: dividing the cervical image set according to a first preset ratio to obtain a cervical image training set and a cervical image validation set. It also includes: dividing the cervical image training set according to a second preset ratio to obtain a sub-training set and a sub-validation set; the preset predictor is configured as a deep learning-based prediction network or a machine learning-based predictor.

[0227] Similar to the cervical cancer grading described above, for example, in the embodiments of this disclosure and other possible embodiments, the preset predictor may be configured as a prediction network based on deep learning (e.g., CNN, RNN, GCN, etc.) or a predictor based on machine learning (e.g., decision tree, Naive Bayes, support vector machine, multilayer perceptron, decision tree, random forest, gradient boosting tree, etc., one or more).

[0228] Similar to the cervical cancer grading described above, in this disclosure and other possible embodiments, determining the fusion weight list corresponding to the sub-predicted probability value set includes: using a preset weight function to perform weight processing on the fusion weight list to be processed corresponding to the sub-predicted probability value set to obtain the corresponding fusion weight list. The preset weight function is configured as one or more of a linear weight function, an exponential weight function, a power weight function, and a predictor ranking weight function.

[0229] Similar to the cervical cancer grading described above, in the embodiments of this disclosure and other possible embodiments, determining the list of fusion weights to be processed includes: calculating the area, recall, and precision enclosed by the receiver operating characteristic curve and the coordinate axis corresponding to the preset clinical guidance using the sub-prediction probability value set; and configuring the area, recall, and precision enclosed by the receiver operating characteristic curve and the coordinate axis as the list of fusion weights to be processed.

[0230] Similar to the cervical cancer grading described above, the linear weighting function is used to process the list of fusion weights to be processed, including: summing the fusion weights to be processed in the list to be processed to obtain a first weight sum; dividing each fusion weight to be processed by the first weight sum to obtain a corresponding fusion weight list. The power weighting function is used to process the list of fusion weights to be processed, including: using a first adjustment parameter (configured to 2 or other possible values), exponentiating each fusion weight to be processed in the list to be processed (the first adjustment parameter is used as the power) to obtain processed fusion power weights; summing the processed fusion power weights to obtain a second weight sum; dividing each processed fusion power weight by the second weight sum to obtain a corresponding fusion weight list. The process of applying the exponential weight function to the list of fusion weights to be processed includes: multiplying the fusion weights in the list of fusion weights to be processed using a second adjustment parameter (configured to 5 or other possible values) to obtain adjusted fusion weights; exponentially processing the adjusted fusion weights (using the adjusted fusion weights as powers of the exponent) to obtain exponential weights; summing the exponential weights to obtain a third weight sum; and dividing the exponential weights by the third weight sum to obtain the corresponding fusion weight list. The process of applying the prediction model ranking weight function to the list of fusion weights to be processed includes: determining the ranking of the sub-prediction model corresponding to each predictor based on the area enclosed by the receiver operating characteristic curve and the coordinate axis, recall, and precision; summing the rankings corresponding to each preset clinical guidance to obtain a fourth weight sum; and dividing the rankings corresponding to each preset clinical guidance by the corresponding fourth weight sum to obtain the corresponding fusion weight list.

[0231] Similar to the cervical cancer grading described above, in the embodiments of this disclosure and other possible embodiments, the step of determining the prediction model corresponding to each preset predictor in the preset predictor pool corresponding to the cervical image training set and the sub-prediction model corresponding to each preset predictor in the sub-training set of the cervical image training set includes: determining the prediction model corresponding to each preset predictor in the preset predictor pool corresponding to the preoperatively acquired cervical image training set and the multiple sub-prediction models corresponding to each preset predictor in the preset predictor pool corresponding to the sub-training set of the preoperatively acquired cervical image training set.

[0232] Similar to the cervical cancer grading described above, in the embodiments of this disclosure and other possible embodiments, determining the corresponding prediction model for each preset predictor in the preset predictor pool corresponding to the preoperative cervical image training set includes: extracting features from the preoperative cervical image training set to obtain the corresponding quantitative features of cervical cancer lesions, and training each preset predictor in the preset predictor pool using the quantitative features of cervical cancer lesions and their postoperative cervical cancer metastasis pathology results labels to obtain multiple corresponding prediction models.

[0233] Similar to the cervical cancer grading described above, in the embodiments of this disclosure and other possible embodiments, determining the corresponding sub-prediction model of each preset predictor in the preset predictor pool corresponding to the sub-training set of the preoperatively acquired cervical image training set includes: quantifying the cervical cancer lesions corresponding to the sub-training set of the preoperatively acquired cervical image training set, and training each preset predictor in the preset predictor pool using the quantitative features of the cervical cancer lesions and their postoperative cervical cancer metastasis pathology results labels to obtain the corresponding sub-prediction model.

[0234] Constructing a cervical cancer metastasis prediction model based on the area under the receiver operating characteristic curve (ROC) and coordinate axes, along with its recall and precision, is more suitable for physicians in diagnosing and treating cervical cancer. Specifically, most researchers currently strive for the highest cervical cancer metastasis prediction model across all evaluation metrics. While achieving the highest prediction across all metrics may be ideal, in practice, it is difficult to achieve this across all evaluation metrics.

[0235] Therefore, it is necessary to construct separate cervical cancer metastasis prediction models corresponding to the assessment indicators that physicians are concerned about. For example, if physicians are concerned about the risk of cervical cancer metastasis (e.g., needing to expand the resection range of cervical cancer lesions in the high-risk group for metastasis), a cervical cancer metastasis prediction model with the highest area under the curve (AUC) should be recommended. If they are concerned about the risk of cervical cancer recurrence, a cervical cancer metastasis prediction model with the highest recall should be recommended. If they are concerned about overtreatment of cervical cancer, a cervical cancer metastasis prediction model with the highest accuracy should be recommended. Simultaneously, a novel clinically goal-oriented integrated machine learning algorithm is proposed, which is used throughout the entire process of constructing cervical cancer grading and metastasis prediction models, eliminating the difficult technical problems of selecting the base learner using existing methods. Furthermore, the three recommended cervical cancer metastasis prediction models, based on an integrated machine learning algorithm with specific clinical goals, provide a solution to ongoing clinical challenges and have the potential to become indispensable quantitative analysis tools for cervical cancer grading and metastasis prediction in the clinical setting.

[0236] In the embodiments of this disclosure and other possible embodiments, before acquiring the cervical image, the process includes: acquiring multiple positioning trigger signals corresponding to the positioning mating component of the foot pedal connected to the bed frame during its movement towards multiple preset positioning positions corresponding to the bed frame; and positioning the foot pedal and the bed frame at preset positioning positions corresponding to the preset positioning signals based on the multiple positioning trigger signals and preset positioning signals. This addresses at least one of the technical problems currently encountered with foot pedals or digital X-ray machines (bedside X-ray machines), such as the difficulty in quickly releasing the foot pedal and bed frame from positioning and quickly assembling them during handling or assembly.

[0237] In the embodiments disclosed herein and other possible embodiments, the bed frame includes: a first bed frame and / or a second bed frame disposed on the opposite side of the first bed frame; footrests are respectively disposed at one end of the first bed frame and one end of the second bed frame. The first bed frame is disposed parallel to the second bed frame. That is, the first bed frame and the second bed frame can be configured to be parallel or, depending on actual needs, non-parallel, but one end of the first bed frame and one end of the second bed frame are respectively connected to both ends of the footrests.

[0238] In the embodiments disclosed herein and other possible embodiments, the foot pedal and the bed frame are positioned at the preset positioning positions corresponding to the preset positioning signals according to the plurality of positioning trigger signals and preset positioning signals, and the relative positions between the bed frame and the foot pedal are adjusted so as to realize that the foot pedal and the bed frame are difficult to quickly release the positioning during the transportation process or difficult to quickly assemble and position during the assembly process.

[0239] In the embodiments disclosed herein and other possible embodiments, the positioning body corresponding to the preset position to be positioned is configured as a groove, and the positioning mating component or the positioning mating part corresponding to the positioning mating component is configured as a protrusion that matches the groove.

[0240] In the embodiments disclosed herein and other possible embodiments, when a force is applied to the positioning and mating component, the applied force acts on the positioning and mating component, and the positioning and mating component drives the foot pedal to move along the position of the positioning body corresponding to the preset position to be positioned. When the positioning and mating component moves to the position of the positioning body, the positioning body connects with the positioning and mating component and makes a "click" sound, thereby indicating that the relative position between the bed frame and the foot pedal has been adjusted to the preset relative position.

[0241] In this embodiment of the disclosure, before obtaining the multiple positioning trigger signals corresponding to the positioning mating component connected to the bed frame during its movement to the multiple preset positioning positions corresponding to the bed frame, the method includes: applying a first force to the positioning mating component during its movement to the multiple preset positioning positions corresponding to the bed frame and applying a second force to the positioning mating component to generate the multiple positioning trigger signals; under the action of the first force and the second force, determining the multiple positioning trigger signals corresponding to the positioning mating component connected to the bed frame during its movement to the multiple preset positioning positions corresponding to the bed frame.

[0242] In this disclosed embodiment and other possible embodiments, when a force (first force) is applied to the positioning and mating component, the force along the positioning body corresponding to the preset positioning position drives the foot pedal to move along the position of the positioning body corresponding to the preset positioning position; the force acting on the positioning and mating component (second force) generates the plurality of positioning trigger signals. In this embodiment of the disclosure, positioning the footrest and the bed frame at a preset positioning position corresponding to a preset positioning signal includes: determining the cumulative number of triggers corresponding to the plurality of positioning trigger signals; and positioning the footrest and the bed frame at the preset positioning position corresponding to the preset positioning signal based on the cumulative number of triggers and the preset cumulative number of positioning trigger signals corresponding to the preset positioning signal.

[0243] In the embodiments disclosed herein and other possible embodiments, the positioning body corresponding to the preset positioning position is configured as multiple positioning bodies corresponding to different preset positioning positions; the positioning and mating component can sequentially contact the multiple positioning bodies corresponding to the different preset positioning positions to generate the multiple positioning trigger signals.

[0244] In this embodiment of the disclosure, the step of positioning the footrest and the bed frame at the preset positioning position corresponding to the preset positioning signal based on the cumulative number of triggers and the preset cumulative number of positioning triggers corresponding to the preset positioning signal includes: if the cumulative number of triggers is equal to the preset cumulative number of positioning triggers, then the footrest and the bed frame are positioned at the preset positioning position corresponding to the preset positioning signal.

[0245] For example, in the embodiments disclosed herein and other possible embodiments, the number of preset positioning cumulative trigger signals corresponding to the preset positioning signal is configured to be 2. The positioning coordination component can sequentially contact the second positioning body corresponding to the different preset positioning positions to generate 2 positioning trigger signals. At this time, the cumulative trigger number corresponding to the 2 positioning trigger signals is configured to be 2, and the positioning foot pedal and the bed frame move to the preset positioning position corresponding to the preset positioning signal.

[0246] In this embodiment of the disclosure, determining multiple positioning trigger signals corresponding to the positioning mating component connected to the footrest of the bed frame during its movement toward multiple preset positioning positions corresponding to the bed frame includes: acquiring an audio signal corresponding to the positioning mating component connected to the footrest of the bed frame during its movement toward multiple preset positioning positions corresponding to the bed frame; and determining multiple positioning trigger signals corresponding to the positioning mating component connected to the footrest of the bed frame during its movement toward multiple preset positioning positions corresponding to the bed frame based on the audio signal.

[0247] In the embodiments disclosed herein and other possible embodiments, an audio acquisition unit (e.g., a recorder or microphone) can be used to acquire the audio signals corresponding to the positioning and mating components of the footrest connected to the bed frame as they move toward multiple preset positioning positions corresponding to the bed frame.

[0248] In this embodiment of the disclosure, determining multiple positioning trigger signals corresponding to the positioning mating component connected to the bed frame during its movement toward multiple preset positioning positions corresponding to the bed frame, based on the audio signal, includes: acquiring a preset spectrum corresponding to the contact mating component with the positioning body corresponding to the multiple preset positioning positions; and determining multiple positioning trigger signals corresponding to the preset spectrum during the movement of the positioning mating component connected to the bed frame toward the multiple preset positioning positions corresponding to the bed frame, based on the audio signal and the preset spectrum.

[0249] In the embodiments disclosed herein and other possible embodiments, a Fourier transform is performed on the audio signal corresponding to the positioning and mating component of the footrest connected to the bed frame during its movement to multiple preset positioning positions corresponding to the bed frame to obtain an audio spectrum; and a preset spectrum corresponding to the connection between the positioning body and the positioning and mating component (emitting a "click" sound) is extracted from the audio spectrum.

[0250] In the embodiments disclosed herein and other possible embodiments, a preset audio signal corresponding to the connection between the positioning body and the positioning component (emitting a "click" sound) is extracted from the audio signal corresponding to the positioning and mating component of the foot pedal connected to the bed frame during the movement of the positioning and mating component to multiple preset positioning positions corresponding to the bed frame; the preset audio signal is subjected to Fourier transform to obtain a preset spectrum.

[0251] In this embodiment of the disclosure, the foot pedal and the bed frame are positioned at the preset positioning positions corresponding to the preset positioning signals according to the plurality of positioning trigger signals and preset positioning signals. The method further includes: using a limiting mechanism to limit and fix the preset positioning position after positioning.

[0252] In this embodiment of the present disclosure, when a second force that generates the plurality of positioning trigger signals is applied to the positioning and mating component, the limiting mechanism is controlled to disengage from the bed frame; when the second force that generates the plurality of positioning trigger signals is removed from the positioning and mating component, the limiting mechanism is controlled to limit the bed frame.

[0253] For example, multiple positioning bodies, including: a first positioning body and a second positioning body; when disengaging from the first or second positioning body corresponding to the multiple positioning bodies, firstly, a force is applied to the outer handle surface of the handle corresponding to the drive mechanism (a force along the direction of movement of the positioning body corresponding to the preset positioning position), thereby releasing the connection of the first or second positioning body corresponding to the multiple positioning bodies (the limiting head of the limiting mechanism disengages from the limiting groove set in the first or second positioning body corresponding to the multiple positioning bodies), and reconnects with the positioning and mating components of the foot pedal according to the preset positioning position of the first or second positioning body corresponding to the multiple positioning bodies.

[0254] In the embodiments disclosed herein and other possible embodiments, the bed includes: a bed frame and a footrest connected to the bed frame; the bed frame is provided with a positioning body corresponding to a preset positioning position; the positioning body corresponding to the preset positioning position is connected to and disconnected from a positioning engagement component of the footrest; the positioning body and the positioning engagement component are used to adjust the relative position between the bed frame and the footrest. This addresses at least one technical problem in the current process of handling or assembly where it is difficult to quickly release positioning or quickly assemble positioning.

[0255] In the embodiments disclosed herein and other possible embodiments, when a force is applied to the positioning and mating component, the applied force acts on the positioning and mating component, and the positioning and mating component drives the foot pedal to move along the position of the positioning body corresponding to the preset position to be positioned. When the positioning and mating component moves to the position of the positioning body, the positioning body connects with the positioning and mating component and makes a "click" sound, thereby indicating that the relative position between the bed frame and the foot pedal has been adjusted to the preset relative position.

[0256] In the embodiments disclosed herein and other possible embodiments, the positioning and mating assembly includes: a sliding body disposed on the bed frame and a positioning and mating component disposed on the sliding body; the sliding body is slidably connected to the bed frame and is used to drive the positioning and mating component of the sliding body to engage and connect with the positioning body.

[0257] In the embodiments disclosed herein and other possible embodiments, the sliding body includes: a telescopic shaft, a compression spring connected to the telescopic shaft, and a positioning and mating component connected to the compression spring.

[0258] In the embodiments disclosed herein and other possible embodiments, when a force is applied to the positioning and mating assembly, the force along the positioning body corresponding to the preset positioning position drives the foot pedal to move along the position of the positioning body corresponding to the preset positioning position. The force acting on the telescopic shaft of the positioning and mating assembly causes the telescopic shaft of the positioning and mating assembly to retract. When the telescopic shaft retracts, it compresses the compression spring connected to the telescopic shaft. The compressed compression spring acts on the positioning and mating component to move. When the telescopic shaft of the positioning and mating assembly moves to the position of the positioning body, the positioning body connects with the telescopic shaft of the positioning and mating assembly and emits a "click" sound, thereby indicating that the relative position between the bed frame and the foot pedal has been adjusted to the preset relative position.

[0259] In the embodiments disclosed herein and other possible embodiments, the sliding body further includes: a bushing fitted onto one end of the telescopic shaft; a compression spring connected to one end of the telescopic shaft and located within the bushing; and a positioning fitting connected to the compression spring and capable of ejecting the bushing.

[0260] In the embodiments disclosed herein and other possible embodiments, when a force is applied to the positioning and mating assembly, the force along the positioning body corresponding to the preset positioning position drives the foot pedal to move along the position of the positioning body corresponding to the preset positioning position. The force acting on the telescopic shaft of the positioning and mating assembly causes the telescopic shaft of the positioning and mating assembly to retract. When the telescopic shaft retracts, it compresses the compression spring connected to the telescopic shaft. The compressed compression spring acts on the positioning and mating component connected to the compression spring and can eject the bushing, thus ejecting the bushing. When the telescopic shaft of the positioning and mating assembly moves to the position of the positioning body, the positioning body connects with the telescopic shaft of the positioning and mating assembly and emits a "click" sound, thereby indicating that the relative position between the bed frame and the foot pedal has been adjusted to the preset relative position.

[0261] In embodiments disclosed herein and other possible embodiments, the positioning and mating component further includes: a driving mechanism connected to the sliding body; the driving mechanism is used to drive the sliding body to slide on the bed frame.

[0262] In the embodiments disclosed herein and other possible embodiments, the force applied by the drive mechanism to drive the sliding body to slide on the bed frame causes the foot pedal to move along the position of the positioning body corresponding to the preset positioning position when a force is applied to the positioning and fitting assembly. The force acting on the telescopic shaft of the positioning and fitting assembly causes the telescopic shaft to retract. When the telescopic shaft retracts, it compresses the spring connected to the telescopic shaft. The compressed spring acts on the positioning fitting component connected to the spring and capable of ejecting the bushing, causing it to pop out of the bushing. When the telescopic shaft of the positioning and fitting assembly moves to the position of the positioning body, the positioning body connects to the telescopic shaft of the positioning and fitting assembly, producing a "click" sound, thus indicating that the relative position between the bed frame and the foot pedal has been adjusted to the preset relative position. When the force on the other end of the telescopic shaft of the sliding body is withdrawn, the telescopic shaft changes from a retracted state to a natural state, the spring compression ceases, and the positioning fitting component retracts into the bushing.

[0263] In embodiments disclosed herein and other possible embodiments, the positioning and mating assembly further includes: a housing disposed on the outside of the sliding body, wherein the other end of the telescopic shaft of the sliding body passes through the housing and is connected to the drive mechanism.

[0264] In embodiments of this disclosure and other possible embodiments, the other end of the telescopic shaft of the sliding body is connected to a drive mechanism through the body housing disposed on its outer side, which is configured as a handle.

[0265] In the embodiments disclosed herein and other possible embodiments, the sliding body is driven by the handle corresponding to the drive mechanism to slide on the bed frame. When a force is applied to the positioning and mating assembly, the force along the positioning body corresponding to the preset positioning position drives the foot pedal to move along the position of the positioning body corresponding to the preset positioning position. The force acting on the other end of the telescopic shaft of the sliding body causes the telescopic shaft of the positioning and mating assembly to retract. When the telescopic shaft retracts, it compresses the compression spring connected to the telescopic shaft. The compressed compression spring acts on the positioning and mating component connected to the compression spring and ejects the bushing. When the telescopic shaft of the positioning and mating assembly moves to the position of the positioning body, the positioning body connects with the telescopic shaft of the positioning and mating assembly and makes a "click" sound, thereby indicating that the relative position between the bed frame and the foot pedal has been adjusted to the preset relative position. When the force on the other end of the telescopic shaft of the sliding body is withdrawn, the telescopic shaft changes from the retracted state to the open state, the compression spring is no longer compressed, and the positioning and mating component retracts into the bushing.

[0266] In embodiments disclosed herein and other possible embodiments, the positioning mechanism further includes a first slider disposed between the bed frame and the main body housing.

[0267] In the embodiments disclosed herein and other possible embodiments, a first slider disposed between the bed frame and the main body housing is provided with a second slider on the opposite side of the bed frame.

[0268] In this embodiment and other possible embodiments, a first slider is provided on one side of the bed frame or the sliding body of the bed frame, and a second slider is provided on the other side of the bed frame. The first and second sliders are used to clamp the bed frame vertically, thereby cooperating with the sliding body to fix the entire positioning mechanism vertically. The first and second sliders are configured as self-lubricating sliders with low hardness (less than a preset hardness), such as plastic blocks, to facilitate movement and prevent scratching the bed frame.

[0269] In the embodiments disclosed herein and other possible embodiments, the positioning body corresponding to the preset position to be positioned is configured as a groove, and the positioning mating component or the positioning mating part corresponding to the positioning mating component is configured as a protrusion that matches the groove.

[0270] In the embodiments disclosed herein and other possible embodiments, the limiting mechanism is further included: the limiting mechanism disposed on the bed frame; the limiting mechanism includes: a limiting head and a rotatable limiting base connected to the limiting head.

[0271] In the embodiments disclosed herein and other possible embodiments, a first slider groove is provided on one side of the bed frame, and the first slider slides within the first slider groove along the sliding direction of the positioning and mating assembly.

[0272] In the embodiments disclosed herein and other possible embodiments, the limiting mechanism is provided on both sides of the bed frame.

[0273] In the embodiments disclosed herein and other possible embodiments, the positioning body corresponding to the preset positioning position is configured as multiple positioning bodies corresponding to different preset positioning positions. For example, multiple positioning bodies include: a first positioning body and a second positioning body; when detached from the first positioning body or the second positioning body, a force is first applied to the outer handle surface of the handle corresponding to the drive mechanism (a force along the direction of movement of the positioning body corresponding to the preset positioning position), releasing the limiting mechanism from the limiting head of the limiting mechanism connected to the first positioning body or the second positioning body, which is inserted into the limiting groove of the first positioning body and the second positioning body corresponding to the bed frame, and reconnecting with the positioning and mating component of the foot pedal according to the preset positioning position of the first positioning body or the second positioning body.

[0274] In the embodiments disclosed herein and other possible embodiments, when force is applied to the handle corresponding to the drive mechanism on the outer handle surface, the limiting head of the limiting mechanism disengages from the limiting groove provided in the bed frame. Simultaneously, when force is applied to the handle corresponding to the drive mechanism on the outer handle surface, the force along the positioning body corresponding to the preset positioning position drives the foot pedal to move along the position of the positioning body corresponding to the preset positioning position. The force acting on the other end of the telescopic shaft of the sliding body (the force applied to the outer handle surface acts on the other end of the telescopic shaft of the sliding body) causes the telescopic shaft to retract, and the compression spring connected to the telescopic shaft to retract under force, forcing the positioning fitting connected to the compression spring and capable of popping out of the bushing to pop out of the bushing so that the positioning fitting can connect to the positioning body corresponding to the preset positioning position (at this time, a "click" sound is emitted). At this time, no force is applied to the handle corresponding to the drive mechanism on the outer handle surface, and the direction of the limiting head is adjusted by using the rotatable limiting base connected to the limiting head. The limiting head of the limiting mechanism is inserted into the limiting groove provided on the bed frame to realize the positioning limitation of the bed frame and the foot pedal, thereby realizing the adjustment of the relative position between the bed frame and the foot pedal.

[0275] In the embodiments disclosed herein and other possible embodiments, the limiting head includes: a limiting head body and a plurality of limiting holes (at least one limiting hole) disposed on the limiting head body. Multiple limiting screws are used to fix the head in place by passing through the plurality of limiting holes and into limiting grooves provided on the bed frame.

[0276] In the embodiments disclosed herein and other possible embodiments, when multiple limiting screws are used to fix the bed frame by passing through the multiple limiting holes and the limiting grooves provided thereon, when the connection between the limiting mechanism and the first positioning body or the second positioning body is released, it is first necessary to remove the fixed connection between the multiple limiting screws and the limiting grooves provided thereon.

[0277] In the embodiments disclosed herein and other possible embodiments, a second slider is provided on the other side of the first slider groove on the bed frame, and the second slider slides within the second slider groove along the sliding direction of the positioning and mating assembly.

[0278] In the embodiments disclosed herein and other possible embodiments, the second slider includes: a second slider body and a positioning body connector disposed on the second slider body; the positioning body connector includes: a positioning body connecting body and a plurality of connecting holes (at least one connecting hole) disposed on the positioning body. A plurality of limiting screws are used to pass through the plurality of limiting holes and connect to the limiting grooves provided on the bed frame to fix one side of the bed frame in a fixed connection.

[0279] The above technical solution solves at least one of the technical problems of the current foot pedal or bed frame of foot-operated or digital X-ray cameras (bedside X-ray machines) being difficult to quickly release and reposition during the handling or assembly process.

[0280] In this embodiment and other possible embodiments, by pressing down on the first handle, a force is applied to the housing of the first handle along the direction of movement of the positioning body corresponding to the preset positioning position. The housing of the first handle is pressed against one side surface of the bed frame, which releases the connection between the limiting mechanism and the positioning body. The limiting head of the limiting mechanism disengages from the limiting groove corresponding to the bed frame. While pressing down on the first handle, the inside of the palm presses against one end of the telescopic shaft. The telescopic shaft retracts, and the compression spring connected to the telescopic shaft contracts under force, forcing the positioning fitting connected to the compression spring and capable of popping out of the bushing to pop out of the bushing, so that the positioning fitting is preset to the positioning position. Connect the corresponding first positioning body (a "click" sound is heard at this time); if the first positioning body corresponding to the preset positioning position is the specified preset positioning position, then release the force applied to the housing of the first handle along the direction of movement of the positioning body corresponding to the preset positioning position and the force between the inside of the palm and one end of the telescopic shaft. At this time, the telescopic shaft returns from retracted to its natural state, and the compression spring connected to the telescopic shaft also returns from retracted to its natural state. The positioning fitting part connected to the compression spring and capable of popping out the bushing retracts to the inside of the bushing. The limiting head of the limiting mechanism connecting the limiting mechanism and the positioning body is inserted into the limiting groove corresponding to the bed frame to achieve quick assembly and positioning. If the first positioning body corresponding to the preset positioning position is not the specified preset positioning position, then only the force between the inside of the palm and one end of the telescopic shaft is released. The positioning fitting part connected to the compression spring and capable of popping out the bushing retracts to the inside of the bushing. Under the force applied to the housing of the first handle along the direction of movement of the positioning body corresponding to the preset positioning position, it continues to move in the direction of movement of the positioning body corresponding to the next preset positioning position. After the positioning fitting part disengages from the first positioning body, the force between the inside of the palm and one end of the telescopic shaft continues until the preset positioning position is reached, achieving rapid assembly and positioning. If rapid release of positioning is achieved, only the force is applied to the housing of the first handle (no force is applied to one end of the telescopic shaft), and the limiting head of the limiting mechanism connecting the limiting mechanism and the positioning body disengages from the limiting groove corresponding to the bed frame.

[0281] In the embodiments disclosed herein and other possible embodiments, a first handle and a second handle are respectively provided at one end of the first bed frame and one side of the second bed frame; the first handle and the second handle are respectively provided with shaft holes corresponding to the telescopic shafts of the two sliding bodies, and one end of the telescopic shaft of the two sliding bodies passes through the corresponding shaft holes to form a protrusion mechanism on the outer handle surface of the first handle and the second handle. By applying force to the protrusion mechanism, the telescopic shaft retracts, the compression spring connected to the telescopic shaft retracts under force, and forces the positioning fitting member connected to the compression spring and capable of popping out of the bushing to pop out of the bushing, so that the positioning fitting member can connect to the positioning body corresponding to the preset positioning position.

[0282] Meanwhile, in the embodiments disclosed herein and other possible embodiments, the first bed frame and the second bed frame are respectively configured with the positioning and mating components. The positioning and mating components include: a housing corresponding to the first bed frame and a housing corresponding to the second bed frame disposed on the outer side of the sliding body.

[0283] In the embodiments disclosed herein and other possible embodiments, the first handle and the second handle are respectively connected to one side of the housing corresponding to the first bed frame and the housing corresponding to the second bed frame, such as by welding. The housing corresponding to the first bed frame and the housing corresponding to the second bed frame are respectively slidably connected to the other side of the first bed frame and the other side of the second bed frame. The housing corresponding to the first bed frame and the housing corresponding to the second bed frame are respectively connected to the ends of the first bed frame and the second bed frame.

[0284] In the embodiments disclosed herein and other possible embodiments, by simultaneously pressing down on the first handle and the second handle, a force is applied to the housings of the first handle and the second handle in the direction of movement of the positioning body corresponding to the preset positioning position. The housings of the first handle and the second handle are respectively pressed against one side surface of the first bed frame and the second bed frame, thereby releasing the limiting head of the limiting mechanism of the connection limiting mechanism between the first bed frame and the positioning body from the limiting groove corresponding to the first bed frame and the second bed frame. While pressing down on the first handle and the second handle, the inner sides of the two palms respectively press against one end of the two telescopic shafts, causing the two telescopic shafts to retract. The compression springs connected to the two telescopic shafts are compressed, forcing the bushings connected to the compression springs to pop out. The positioning fitting pops out from the outside of the bushing so that the positioning fitting can connect to the first positioning body corresponding to the preset positioning position (a "click" sound is made at this time); if the first positioning body corresponding to the preset positioning position is the specified preset positioning position, the force applied to the housing of the first handle and the housing of the second handle along the direction of movement of the positioning body corresponding to the preset positioning position and the force between the inner sides of the two palms and one end of the telescopic shaft are released. At this time, the two telescopic shafts return from retracted to their natural state, and the compression spring connected to the telescopic shaft also returns from retracted to its natural state. The positioning fitting connected to the compression spring and capable of popping out from the bushing retracts to the inside of the bushing. The limiting head of the limiting mechanism connected to the positioning body is inserted into the limiting groove corresponding to the first bed frame and the second bed frame to achieve rapid assembly and positioning. If the first positioning body corresponding to the preset positioning position is not the specified preset positioning position, then only the force between the inside of the palm and one end of the telescopic shaft is released. The positioning fitting part connected to the compression spring and capable of popping out the bushing retracts to the inside of the bushing. Under the force applied to the housing of the first handle along the direction of movement of the positioning body corresponding to the preset positioning position, it continues to move in the direction of movement of the positioning body corresponding to the next preset positioning position. After the positioning fitting part disengages from the first positioning body, the force between the inside of the palm and one end of the telescopic shaft continues until the preset positioning position is reached, thus achieving rapid assembly and positioning. If rapid release of positioning is achieved, only the force is applied to the housing of the first handle and the housing of the second handle (no force is applied to one end of the telescopic shaft). The limiting heads of the two limiting mechanisms and the limiting mechanisms connecting the positioning body disengage from the limiting slots corresponding to the first bed frame and the second bed frame, respectively.

[0285] This disclosure also proposes a medical image processing method, including: applying the positioning control method described above to adjust the relative position between the bed frame and the footrest; and acquiring X-ray medical images of preset locations corresponding to the relative position between the bed frame and the footrest.

[0286] The execution entity of the multi-level intelligent analysis method for cervical images can be an image processing device. For example, the multi-level intelligent analysis method for cervical images can be executed by a terminal device, server, or other processing device. The terminal device can be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, wearable device, etc. In some possible implementations, the multi-level intelligent analysis method for cervical images can be implemented by a processor calling computer-readable instructions stored in memory.

[0287] Those skilled in the art will understand that, in the above-described multi-level intelligent analysis method for cervical images in specific embodiments, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0288] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a detection unit for performing cervical cancer target detection on cervical images using a preset cervical cancer target detection model; a judgment unit for determining the presence of cervical cancer if the preset cervical cancer target detection model detects cervical cancer lesions in the cervical images; and for extracting quantitative features of the cervical cancer lesions corresponding to the cervical images if the preset cervical cancer target detection model does not detect cervical cancer lesions in the cervical images; and a screening unit for performing cervical cancer screening and early warning using a preset early cervical cancer screening model and the quantitative features of the cervical cancer lesions.

[0289] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the aforementioned multi-level intelligent analysis method for cervical images.

[0290] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a computer-readable storage medium storing a computer program / instructions and a bit stream thereon, wherein the computer program / instructions, when executed by a processor, implement the aforementioned multi-level intelligent analysis method for cervical images to generate the bit stream.

[0291] According to one aspect of this disclosure, a multi-level intelligent analysis system for cervical images is provided, comprising: a computer program product configured with a computer program / instruction, which, when executed by a processor, implements the aforementioned multi-level intelligent analysis method for cervical images.

[0292] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to execute the cervical image multi-level intelligent analysis method described in the above method embodiments. The specific implementation can be referred to the description of the cervical image multi-level intelligent analysis method embodiments above, and will not be repeated here for the sake of brevity.

[0293] This disclosure also proposes a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the aforementioned multi-level intelligent analysis method for cervical images. The computer-readable storage medium may be a non-volatile computer-readable storage medium.

[0294] This disclosure also proposes an electronic device, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured for the aforementioned multi-level intelligent analysis method for cervical images. The electronic device can be provided as a terminal, a server, or other form of device.

[0295] This disclosure also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the above-described multi-level intelligent analysis method for cervical images.

[0296] Figure 2 This is a block diagram illustrating an electronic device 800 according to an exemplary embodiment. For example, the electronic device 800 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, or other terminal.

[0297] Reference Figure 2 The electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0298] Processing component 802 typically controls the overall operation of electronic device 800, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the methods described above. Furthermore, processing component 802 may include one or more modules to facilitate interaction between processing component 802 and other components. For example, processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.

[0299] Memory 804 is configured to store various types of data to support the operation of electronic device 800. Examples of such data include instructions for any application or method operating on electronic device 800, contact data, phonebook data, messages, pictures, videos, etc. Memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0300] Power supply component 806 provides power to various components of electronic device 800. Power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 800.

[0301] Multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 808 includes a front-facing camera and / or a rear-facing camera. When the electronic device 800 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0302] Audio component 810 is configured to output and / or input audio signals. For example, audio component 810 includes a microphone (MIC) configured to receive external audio signals when electronic device 800 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 804 or transmitted via communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

[0303] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0304] Sensor assembly 814 includes one or more sensors for providing state assessments of various aspects of electronic device 800. For example, sensor assembly 814 can detect the on / off state of electronic device 800, the relative positioning of components such as the display and keypad of electronic device 800, changes in position of electronic device 800 or a component of electronic device 800, the presence or absence of user contact with electronic device 800, orientation or acceleration / deceleration of electronic device 800, and temperature changes of electronic device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 814 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0305] Communication component 816 is configured to facilitate wired or wireless communication between electronic device 800 and other devices. Electronic device 800 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 816 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 816 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0306] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0307] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 804 including computer program instructions that can be executed by a processor 820 of an electronic device 800 to perform the above-described method.

[0308] Figure 3 This is a block diagram illustrating an electronic device 1900 according to an exemplary embodiment. For example, the electronic device 1900 may be provided as a server. (Refer to...) Figure 3The electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0309] Electronic device 1900 may also include a power supply component 1926 configured to perform power management of electronic device 1900, a wired or wireless network interface 1950 configured to connect electronic device 1900 to a network, and an input / output (I / O) interface 1958. Electronic device 1900 can operate on an operating system stored in memory 1932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0310] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of an electronic device 1900 to perform the above-described method.

[0311] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0312] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0313] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0314] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A multi-level intelligent analysis method for cervical images, characterized in that, include: A pre-defined cervical cancer target detection model was used to detect cervical cancer targets in cervical images. If the preset cervical cancer target detection model detects cervical cancer lesions in the cervical image, then cervical cancer is present. If the preset cervical cancer target detection model does not detect cervical cancer lesions in the cervical image, then the cervical features corresponding to the cervical image are extracted. A pre-set early cervical cancer screening model and the aforementioned cervical characteristics are used to conduct cervical cancer screening and early warning.

2. The multi-level intelligent analysis method for cervical images according to claim 1, characterized in that, Determining the preset cervical cancer target detection model includes: training a preset deep learning target detection network using a preset number of cervical cancer training images and their corresponding cervical cancer labeled training images to obtain a preset feature target detection model constructed based on the deep learning target detection network; and / or, determining a preset early cervical cancer screening model includes: training a preset deep learning classification network or a preset machine learning classifier using cervical features corresponding to a preset number of cervical cancer training images and cervical cancer pathology result labels corresponding to the cervical cancer features to obtain a preset early cervical cancer screening model; and / or, before training the preset deep learning classification network or preset machine learning classifier, includes: encoding the corresponding classification features corresponding to the cervical features to obtain quantitative features of cervical cancer lesion encoding; and standardizing or normalizing the corresponding values ​​corresponding to the cervical features to obtain quantitative features of cervical cancer lesion processing.

3. The multi-level intelligent analysis method for cervical images according to any one of claims 1 or 2, characterized in that, If cervical cancer is present, the method further includes: segmenting the cervical cancer lesions in the cervical image to obtain a cervical cancer lesion image; extracting quantitative features from the cervical cancer lesions in the cervical cancer lesion image to obtain quantitative features of the cervical cancer lesions; and using the quantitative features of the cervical cancer lesions to perform cervical cancer grading or cervical cancer metastasis prediction.

4. The multi-level intelligent analysis method for cervical images according to claim 3, characterized in that, The segmentation of cervical cancer lesions in the cervical image includes: constructing a binary hybrid location mask training image for each set of cervical cancer training images in the cervical cancer training image set, and training images of the cervix and the corresponding cervical cancer lesions within the cervix; training the preset deep learning segmentation network using the binary hybrid location mask training images, the edge mask training images of the cervical cancer lesions, and the corresponding cervical cancer training images to obtain a cervical cancer lesion segmentation model; and segmenting the cervical cancer lesions in the cervical image to be segmented based on the cervical cancer lesion segmentation model.

5. The multi-level intelligent analysis method for cervical images according to any one of claims 3 or 4, characterized in that, The quantitative feature configuration of the cervical cancer lesion is whether the cervical cancer lesion exhibits lobulation; wherein, determining whether the cervical cancer lesion exhibits lobulation includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical image to obtain a cervical cancer lesion mask erosion image; subtracting the cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain a cervical cancer lesion mask-processed image; judging the morphology of the cervical cancer lesion in the cervical cancer lesion mask-processed image; if the cervical cancer lesion morphology is irregular, determining whether the cervical cancer lesion exhibits lobulation based on the cervical cancer lesion mask-processed image.

6. The multi-level intelligent analysis method for cervical images according to any one of claims 3-5, characterized in that, The quantitative feature configuration of the cervical cancer lesion is whether the boundary of the cervical cancer lesion is clear; wherein, determining whether the boundary of the cervical cancer lesion is clear includes: performing an erosion operation on the cervical cancer lesion mask image corresponding to the cervical image to obtain a cervical cancer lesion mask erosion image; subtracting the cervical cancer lesion mask erosion image from the cervical cancer lesion mask image to obtain a cervical cancer lesion mask-processed image; mapping the cervical cancer lesion position coordinates in the cervical cancer lesion mask-processed image to the cervical image; calculating multiple gradients corresponding to the cervical cancer lesion position coordinates in the cervical image respectively; if the gradient magnitude values ​​corresponding to the multiple cervical cancer lesions are greater than or equal to a preset gradient magnitude value, then the boundary of the cervical cancer lesion corresponding to the preset gradient magnitude value is determined to be clear; otherwise, the boundary of the cervical cancer lesion is determined to be unclear.

7. The multi-level intelligent analysis method for cervical images according to any one of claims 3-6, characterized in that, The quantitative characteristics of cervical cancer lesions are configured as one or more of the following: the average gray value of the cervical cancer lesion, the major axis of the largest cross-sectional area of ​​the cervical cancer lesion, the minor axis of the largest cross-sectional area of ​​the cervical cancer lesion, the ratio of the major axis to the minor axis of the largest cross-sectional area of ​​the cervical cancer lesion, the location of the cervical cancer lesion on the cervix, the morphology of the cervical cancer lesion, the presence of a spike sign, the presence of a vacuolar sign, and the proportion of solid cervical cancer lesions to the total number of cervical cancer lesions.

8. The multi-level intelligent analysis method for cervical images according to any one of claims 3-7, characterized in that, The method of grading cervical cancer using the quantitative features of cervical cancer lesions includes: determining the grading model corresponding to each preset classifier in the preset classifier pool corresponding to the cervical image training set, and the sub-grading model corresponding to each preset classifier in the sub-training set of the cervical image training set; determining the sub-grading probability value set on the sub-validation set using the sub-grading model and the sub-validation set of the cervical image training set; determining the grading probability value set on the validation set using the grading model and the cervical image validation set corresponding to the cervical image training set; and optimizing the grading probability value set using the fusion weights corresponding to the sub-grading probability value set to obtain the corresponding fusion probability set for cervical cancer grading, thereby completing the cervical cancer grading process.

9. The multi-level intelligent analysis method for cervical images according to any one of claims 3-7, characterized in that, The method for predicting cervical cancer metastasis using the quantitative features of cervical cancer lesions includes: determining the prediction model corresponding to each preset predictor in the preset predictor pool corresponding to the cervical image training set, and the sub-prediction model corresponding to each preset predictor in the sub-training set of the cervical image training set; determining the sub-prediction probability value set on the sub-validation set using the sub-prediction model and the sub-validation set of the cervical image training set; determining the prediction probability value set on the validation set using the prediction model and the cervical image validation set corresponding to the cervical image training set; and optimizing the prediction probability value set using the fusion weights corresponding to the sub-prediction probability value set to obtain the corresponding fusion probability set for cervical cancer prediction, thereby completing the cervical cancer metastasis prediction.

10. A multi-level intelligent analysis system for cervical images, characterized in that, include: The detection unit is used to perform cervical cancer target detection on cervical images using a preset cervical cancer target detection model. The judgment unit is used to determine whether cervical cancer exists if the preset cervical cancer target detection model detects cervical cancer lesions in the cervical image. If the preset cervical cancer target detection model does not detect cervical cancer lesions in the cervical image, then the cervical features corresponding to the cervical image are extracted. The screening unit is used to perform cervical cancer screening and early warning using a preset early cervical cancer screening model and the cervical characteristics. or, The method comprises: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the multi-level intelligent analysis method for cervical images according to any one of claims 1-9; or, it comprises: a computer-readable storage medium having a computer program / instructions and a bit stream stored thereon, wherein the computer program / instructions, when executed by the processor, implement the multi-level intelligent analysis method for cervical images according to any one of claims 1-9 to generate the bit stream; or, it comprises: a computer program product having a computer program / instructions configured thereon, wherein the computer program / instructions, when executed by the processor, implement the multi-level intelligent analysis method for cervical images according to any one of claims 1-9.

Images