An AI-based auxiliary method and system for cervical cell image analysis

Abstract

This invention relates to the field of medical image processing technology, specifically to an AI-based auxiliary method and system for cervical cell image analysis. The method includes: acquiring multimodal data, including cervical cell image data, clinical data, and patient genetic data; preprocessing the acquired cervical cell image data; constructing a generative adversarial network (GAN) to enhance and extract features from the cervical cell images; fusing the extracted features with clinical data and patient genetic data; and training a cervical cell classification model using the fused feature data; inputting real cervical cell images into the trained cervical cell classification model; and outputting the analysis results regarding whether the cervical cells are abnormal and the type and degree of abnormality. This invention improves the efficiency of cervical cell image analysis and reduces the workload of doctors by combining multimodal data acquisition with image enhancement and classification methods based on deep learning technology.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to an AI-based method and system for assisting in the analysis of cervical cell images. Background Technology

[0002] Cervical cancer is one of the most common malignant tumors among women worldwide, seriously threatening their health and lives. Early and accurate diagnosis is crucial for improving the survival rate of cervical cancer patients. Currently, cervical cell image analysis is one of the important means of cervical cancer screening, which involves observing and analyzing cervical cell smear images to determine whether there are abnormalities in the cells. However, traditional cervical cell image analysis methods have many limitations.

[0003] On the one hand, manual image interpretation relies heavily on the experience and expertise of pathologists, is highly subjective, and involves a large workload, making it prone to misdiagnosis and missed diagnosis. On the other hand, while some existing automated analysis methods based on machine learning have improved diagnostic efficiency to some extent, they rely solely on single cell image data and struggle to comprehensively capture the characteristic information of cells, thus limiting diagnostic accuracy. Furthermore, due to the complexity and diversity of cervical cell images, the cell image appearance varies significantly among different patients, resulting in insufficient generalization ability of the models. Summary of the Invention

[0004] To address the aforementioned technical shortcomings, this invention provides an AI-based cervical cell image analysis assistance method and system, thereby improving the accuracy and reliability of cervical cell image analysis.

[0005] This invention is achieved through the following technical solution:

[0006] An AI-based cervical cell image analysis-assisted method is provided, the method comprising the following steps:

[0007] Step S10: Perform multimodal data acquisition, including cervical cell image data, clinical data and patient genetic data, and preprocess the acquired cervical cell image data;

[0008] Step S20: Construct a generative adversarial network suitable for cervical cell images. Generate cervical cell images through the generator in the generative adversarial network. Input the cervical cell images generated by the generator and the real cervical cell images collected together into a pre-trained convolutional neural network for feature extraction.

[0009] Step S30: The extracted features are fused with the collected clinical data and patient gene data. The fused feature data is divided into a training set and a test set. The training set is used to train the cervical cell classification model, and the test set is used to evaluate the cervical cell classification model. The model is optimized based on the evaluation results to obtain the optimized cervical cell classification model.

[0010] Step S40: The real cervical cell images to be tested are subjected to the data acquisition and preprocessing in step S10 and fused. The data is then input into the optimized cervical cell classification model. The model outputs whether the cervical cells are abnormal and the analysis results of the abnormality type and degree.

[0011] Preferably, step S10 involves multimodal data acquisition, including cervical cell image data, clinical data, and patient genetic data, and preprocessing of the acquired cervical cell image data, including the following steps:

[0012] Cervical cell image data acquisition: Cervical cell images were acquired using a high-resolution microscope to ensure clear images that accurately display the morphology and structural characteristics of the cervical cells. Relevant parameters during image acquisition, including magnification and exposure time, were also recorded.

[0013] Clinical information collection: Collect clinical information from patients who provide cervical cells, including age, medical history, and symptoms. This information helps to understand the patient's condition from different perspectives and provides more comprehensive background information for subsequent analysis.

[0014] Genetic data collection: Collect relevant genetic data from patients who provide cervical cells, including gene expression levels closely related to the development of cervical cancer. Genetic data can reflect the intrinsic molecular characteristics of cells, and when combined with cell image data, it can reveal abnormalities in cells more deeply.

[0015] Cervical cell image preprocessing: The acquired cervical cell images are preprocessed, including image denoising, contrast enhancement, and normalization, to improve image quality and facilitate subsequent analysis. Clinical information and genetic data are also standardized, and feature vectors of clinical information and genetic data are extracted so that they can be effectively fused with the cervical cell images.

[0016] Preferably, in step S20, a generative adversarial network suitable for cervical cell images is constructed, including a generator and a discriminator, wherein the step of constructing the generator includes:

[0017] The input layer integrates cervical cell features: In addition to inputting traditional random noise vectors, the generator's input layer also incorporates multimodal data closely related to the biological characteristics of cervical cells. For example, it extracts cell density information (number of cells per square millimeter) and statistical features of cell size (average diameter, size variance) from cervical cell smear samples. These features are encoded as vectors and concatenated with random noise as input to the generator, making the generator's image generation process more closely match the actual feature distribution of cervical cells.

[0018] Deconvolutional layer with attention mechanism: Introducing attention mechanism into deconvolutional layer. In cervical cell images, different regions such as cell nucleus and cytoplasm have different importance. Through attention mechanism, the generator focuses more on the image generation of key regions.

[0019] The output layer simulates cervical cell texture: A special combination of activation functions is used in the output layer to make the generated cervical cell image closer to the texture of the real cervical cell image. The Tanh function is used to limit the output value to the range of -1 to 1, and a texture enhancement layer is added. This texture enhancement layer extracts the texture features of the image by performing a convolution operation on the output image and fuses it with the original output image to enhance the texture details of the generated image and simulate the delicate texture features in the cervical cell image.

[0020] The generator loss function combines cervical cell feature matching: The generator loss function includes traditional cross-entropy loss and cell feature matching loss. It calculates the differences between the generated cervical cell image and the real cervical image in key cell features, such as the size, shape, and staining intensity of the cell nucleus. The mean squared error loss can be used to measure the differences in these features. The weighted sum of the mean squared error loss and the cross-entropy loss is used as the total loss of the generator. This makes the image generated by the generator closer to the real image in terms of key cell features.

[0021] Preferably, in step S20, a generative adversarial network suitable for cervical cell images is constructed, including a generator and a discriminator, wherein the step of constructing the discriminator includes:

[0022] Multi-scale feature extraction input layer: The discriminator's input includes the cervical cell image itself and feature maps of the cervical cell image at different scales. The features of the cervical cell image exhibit different characteristics at different scales. Through multi-scale feature extraction, the discriminator can gain a more comprehensive understanding of the image's features. By using convolutional kernels of different sizes to process the input cervical cell image in parallel, cervical cell feature maps at different scales are obtained. These cervical cell feature maps are then concatenated together as the input to the discriminator.

[0023] Cervical cell morphology-based feature pooling layer: A feature pooling layer based on cervical cell morphology is added between convolutional layers. Cervical cells have unique morphological characteristics, such as round or oval shapes. This pooling layer performs pooling operations on the feature map based on the morphological information of cervical cells, preserving important features related to cervical cell morphology.

[0024] The output layer incorporates clinical knowledge constraints: the discriminator's output judges the authenticity of images based on their own features and is constrained by clinical knowledge. By introducing a clinical knowledge module into the discriminator's output layer, which contains information on the ratio of normal to abnormal cervical cells, the discriminator's accuracy is improved by adjusting the discrimination results based on this ratio information.

[0025] The discriminator loss function incorporates cell distribution information: The discriminator's loss function incorporates the distribution information of cervical cells in the image to distinguish between real and fake cervical cell images. In real cervical cell images, the distribution of cells follows certain patterns, such as the distance between cells and the degree of clustering. The difference in cell distribution between the generated image and the real image can be calculated, and the similarity of cervical cell distribution can be measured using Wasserstein distance, which is then incorporated into the discriminator's loss function.

[0026] Preferably, in step S30, the extracted features are fused with the collected clinical data and patient gene data using a splicing fusion method, and the steps include:

[0027] Concatenation and fusion: The feature vectors of cervical cell images extracted by the pre-trained convolutional neural network, the pre-processed clinical information feature vectors and gene data feature vectors are concatenated in sequence to form a new fused feature vector;

[0028] Normalization: The fused feature vector obtained after splicing is normalized to balance the contribution of different modalities and ensure that each feature has equal importance in subsequent classification or analysis tasks.

[0029] Preferably, the step of training the cervical cell classification model in step S30 includes:

[0030] Data partitioning: The fused feature data is divided into training and test sets in a 7:3 ratio;

[0031] Cervical cell classification model construction and training: The cervical cell classification model is built based on support vector machines and random forests. Kernel function parameters and model parameters are adjusted to make the model suitable for cervical cell classification calculations. Based on the complexity of cervical cell image features, radial basis functions (RBFs) are chosen as the kernel function. RBFs can handle non-linearly separable data and are effective in capturing the complex relationships between cervical cell features. The penalty factor C and kernel coefficient γ are initialized and dynamically adjusted using a grid search method. The model performance under different parameter combinations is evaluated on the test set to select the optimal parameter combination. The penalty factor C controls the balance between training error and model complexity of the cervical cell classification model. A moderate initial value is set. Initial values, such as C=1; kernel coefficient γ affects the calculation result of radial basis function, and can be initialized to γ=0.1; random forest includes decision trees. Choose an appropriate number of decision trees based on the actual data, such as 100 trees. Too few decision trees may lead to underfitting of the model, while too many will increase the computational cost and may lead to overfitting. Based on the model training results, use a random search method to dynamically adjust the number of decision trees and other random forest parameters, such as the feature subset size. When splitting at the node of each decision tree, randomly select a certain number of features as candidate features from the fused feature data. For cervical cell image fusion features, an appropriate feature subset size can be selected based on the number of features and experience, such as selecting the square root of the total number of features as the feature subset size.

[0032] Cervical cell classification model evaluation and optimization: The performance of the cervical cell classification model is comprehensively evaluated using accuracy, recall and F1 score. The three evaluation indicators are weighted differently and the cervical cell classification model evaluation score is obtained through weighted calculation. When the evaluation score is greater than the set qualified value, the optimal classification model is obtained. After determining the model version, it is saved and put into actual use.

[0033] Preferably, the model output in step S40, including whether the cervical cells are abnormal and the analysis results of the type and degree of abnormality, includes:

[0034] Determining whether cervical cells are abnormal: The cervical cell classification model determines whether cervical cells are normal or abnormal based on the input real cervical cell images. For example, the cervical cell classification model comprehensively analyzes the features of the input cervical cell images and the fused clinical and genetic data features. When its feature pattern highly matches the feature pattern of normal cervical cells in the training set, the model outputs that the cervical cells are normal; when it highly matches the feature pattern of abnormal cervical cells, the model outputs that the cervical cells are abnormal.

[0035] Analysis of abnormal cervical cell results: When the model outputs abnormal cervical cells, the abnormal cervical cell types are further subdivided, and the degree of abnormal cervical cells is assessed through a series of quantitative indicators, including cell nucleus size, degree of irregularity of cell nucleus shape, chromatin density and distribution uniformity, and specific values ​​of nucleocytoplasmic ratio. Based on the above quantitative indicators, the model classifies the degree of abnormal cervical cells into different levels.

[0036] Furthermore, to achieve the above objectives, this invention also proposes an AI-based cervical cell image analysis assistance system, which includes:

[0037] Cervical cell multimodal data acquisition module: used to acquire multimodal data, including cervical cell image data, clinical data and patient genetic data, and to preprocess the acquired cervical cell image data;

[0038] Cervical cell image generation and feature extraction module: This module is used to construct a generative adversarial network suitable for cervical cell images. The generator in the generative adversarial network generates cervical cell images, and the cervical cell images generated by the generator and the real cervical cell images collected are input together into a pre-trained convolutional neural network for feature extraction.

[0039] Cervical cell feature fusion and cervical cell classification model training module: This module is used to fuse extracted features with collected clinical data and patient gene data. The fused feature data is divided into a training set and a test set. The training set is used to train the cervical cell classification model, and the test set is used to evaluate the cervical cell classification model. Based on the evaluation results, the model is optimized to obtain the optimized cervical cell classification model.

[0040] Cervical cell abnormality judgment module: It is used to collect and preprocess the real cervical cell images to be tested in step S10, and then fuse them. The data is input into the optimized cervical cell classification model, and the model outputs whether the cervical cells are abnormal and the analysis results of the abnormality type and degree.

[0041] Furthermore, to achieve the above objectives, the present invention also proposes an AI-based cervical cell image analysis auxiliary device, the device comprising: a memory, a processor, and programs such as an AI-based cervical cell image analysis auxiliary algorithm stored in the memory and executable on the processor, wherein the AI-based cervical cell image analysis auxiliary algorithm and other programs are steps for implementing the AI-based cervical cell image analysis auxiliary method described above.

[0042] In addition, to achieve the above objectives, the present invention also provides a computer program product, which includes programs such as an AI-based cervical cell image analysis auxiliary algorithm. When the AI-based cervical cell image analysis auxiliary algorithm and other programs are executed by a processor, they implement an AI-based cervical cell image analysis auxiliary method as described above.

[0043] The advantages and effects of this invention are:

[0044] This invention proposes an AI-based cervical cell image analysis assistance method and system. By fully utilizing the complementary information from cervical cell images, clinical information, and genetic data through multimodal data acquisition, it can gain a more comprehensive and in-depth understanding of the state of cervical cells, improve the diagnostic accuracy of cervical cell abnormalities, and reduce the occurrence of misdiagnosis and missed diagnosis. In addition, by combining generative adversarial networks to enhance the features of cervical cell image data, the cervical cell image dataset is expanded, the data diversity is increased, and the model can learn richer cell features, improving the model's generalization ability and better handling complex cell images from different patients. At the same time, it provides doctors with complete, objective, and accurate analysis results, assisting them in diagnosing cervical cell images, reducing their workload, improving diagnostic efficiency, and providing strong support for the early diagnosis and treatment of cervical cancer. Attached Figure Description

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 This is a flowchart of an AI-based cervical cell image analysis assistance method according to the present invention.

[0047] Figure 2 This is a schematic diagram of the structure of an AI-based cervical cell image analysis auxiliary system according to the present invention.

[0048] Figure 3 This is a schematic block diagram of an AI-based cervical cell image analysis auxiliary electronic device according to the present invention. Detailed Implementation

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] like Figure 1 As shown, in one embodiment of the present invention, an AI-based cervical cell image analysis-assisted method includes the following steps:

[0051] Step S10: Perform multimodal data acquisition, including cervical cell image data, clinical data and patient genetic data, and preprocess the acquired cervical cell image data.

[0052] Specifically, step S10 involves multimodal data acquisition, including cervical cell image data, clinical data, and patient genetic data, and preprocessing of the acquired cervical cell image data, including the following steps:

[0053] Cervical cell image data acquisition: Cervical cell images are acquired using a high-resolution microscope to ensure clarity and accurate display of morphology and structural features of cervical cells. Relevant parameters during image acquisition are also recorded, including magnification and exposure time. For example, 1000 cervical cell smear samples were collected from a hospital, and images were acquired using a 400x resolution microscope to ensure clear display of cell morphology.

[0054] Clinical information collection: Collect clinical information from patients who provide cervical cells, including age, medical history, and symptoms. This information helps to understand the patient's condition from different perspectives and provides more comprehensive background information for subsequent analysis. For example, record the patient's age, whether they have a gynecological history, and whether they are infected with human papillomavirus (HPV).

[0055] Genetic data collection: Collect relevant genetic data from patients who provide cervical cells, including the expression levels of genes closely related to the development of cervical cancer. Genetic data can reflect the intrinsic molecular characteristics of cells. Combined with cell image data, it can reveal abnormalities in cells more deeply. For example, through gene detection technology, the expression level data of 10 key genes related to cervical cancer in patients can be obtained.

[0056] Cervical cell image preprocessing: The acquired cervical cell images are preprocessed, including image denoising, contrast enhancement, and normalization, to improve image quality and facilitate subsequent analysis. Clinical information and genetic data are standardized, and feature vectors of clinical information and genetic data are extracted to enable effective fusion with the cervical cell images. For example, Gaussian denoising is performed on the cell images, histogram equalization is used to enhance contrast, and the images are normalized to the [0,1] interval. Clinical information is encoded, and genetic data is standardized so that its mean is 0 and its standard deviation is 1.

[0057] Step S20: Construct a generative adversarial network suitable for cervical cell images. Generate cervical cell images through the generator in the generative adversarial network. Input the cervical cell images generated by the generator and the real cervical cell images collected together into a pre-trained convolutional neural network for feature extraction.

[0058] Specifically, in step S20, a generative adversarial network suitable for cervical cell images is constructed, including a generator and a discriminator. The step of constructing the generator includes:

[0059] The input layer integrates cervical cell features: In addition to inputting traditional random noise vectors, the generator's input layer also incorporates multimodal data closely related to the biological characteristics of cervical cells. For example, it extracts cell density information (number of cells per square millimeter) and statistical features of cell size (average diameter, size variance) from cervical cell smear samples. These features are encoded as vectors and concatenated with random noise as input to the generator, making the generator's image generation process more closely match the actual feature distribution of cervical cells.

[0060] Introducing attention mechanisms in deconvolutional layers: By introducing attention mechanisms into deconvolutional layers, different regions such as the cell nucleus and cytoplasm in cervical cell images have different importance. The attention mechanism allows the generator to focus more on the image generation of key regions. For example, after a certain deconvolution operation, the feature map is weighted by calculating the attention weight at each position, so that the features of key regions such as the cell nucleus are generated more accurately. Convolution-based attention modules, such as channel attention modules (SE-Block), can be used to weight the channel dimension of the feature map and highlight important feature channels.

[0061] The output layer simulates cervical cell texture: A special combination of activation functions is used in the output layer to make the generated cervical cell image closer to the texture of the real cervical cell image. The Tanh function is used to limit the output value to the range of -1 to 1, and a texture enhancement layer is added. This texture enhancement layer extracts the texture features of the image by performing a convolution operation on the output image and fuses it with the original output image to enhance the texture details of the generated image and simulate the delicate texture features in the cervical cell image.

[0062] The generator loss function combines cervical cell feature matching: The generator loss function includes traditional cross-entropy loss and cell feature matching loss. It calculates the differences between the generated cervical cell image and the real cervical image in key cell features, such as the size, shape, and staining intensity of the cell nucleus. The mean squared error loss can be used to measure the differences in these features. The weighted sum of the mean squared error loss and the cross-entropy loss is used as the total loss of the generator. This makes the image generated by the generator closer to the real image in terms of key cell features.

[0063] Specifically, in step S20, a generative adversarial network suitable for cervical cell images is constructed, including a generator and a discriminator, wherein the step of constructing the discriminator includes:

[0064] Multi-scale feature extraction input layer: The discriminator's input includes the cervical cell image itself and feature maps of the cervical cell image at different scales. The features of the cervical cell image exhibit different characteristics at different scales. For example, at a small scale, the microscopic structure of the cell, such as the distribution of chromatin, can be captured; at a large scale, the overall morphology and distribution of the cell can be observed. Multi-scale feature extraction allows the discriminator to gain a more comprehensive understanding of the image's features. By using convolutional kernels of different sizes to process the input cervical cell image in parallel, cervical cell feature maps at different scales are obtained. These cervical cell feature maps are then concatenated as the input to the discriminator.

[0065] Cervical cell morphology-based feature pooling layer: A cervical cell morphology-based feature pooling layer is added between convolutional layers. Cervical cells have unique morphological characteristics, such as round or oval shapes. This pooling layer performs pooling operations on the feature map based on the morphological information of cervical cells, preserving important features related to cervical cell morphology. For example, shape-based pooling kernels can be used to pool regions that conform to cell morphology, enhancing the discriminator's sensitivity to cell morphology features.

[0066] The output layer incorporates clinical knowledge constraints: the discriminator's output judges the authenticity of images based on their inherent features, while also incorporating clinical knowledge constraints. For example, in real cervical cell images, the ratio of normal to abnormal cervical cells remains relatively stable within a certain range. By introducing a clinical knowledge module into the discriminator's output layer, which contains information on the ratio of normal to abnormal cervical cells, the discriminator's accuracy is improved by adjusting the discrimination results based on this information.

[0067] The discriminator's loss function incorporates cell distribution information: It combines the distribution information of cervical cells in the image to distinguish between real and fake cervical cell images. In real cervical cell images, cell distribution follows certain patterns, such as the distance between cells and their clustering degree. By calculating the difference in cell distribution between the generated image and the real image, the similarity of cervical cell distribution is measured using Wasserstein distance and incorporated into the discriminator's loss function. This allows the discriminator to better determine the authenticity of the image.

[0068] Step S30: The extracted features are fused with the collected clinical data and patient gene data. The fused feature data is divided into a training set and a test set. The training set is used to train the cervical cell classification model, and the test set is used to evaluate the cervical cell classification model. The model is then optimized based on the evaluation results to obtain the optimized cervical cell classification model.

[0069] Specifically, in step S30, the extracted features are fused with the collected clinical data and patient gene data using a splicing fusion method. The steps include:

[0070] Concatenation and fusion: The feature vector of the cervical cell image extracted by the pre-trained convolutional neural network, the pre-processed clinical information feature vector, and the gene data feature vector are concatenated in sequence to form a new fused feature vector. For example, if the length of the cervical cell image feature vector is 512, the length of the clinical information feature vector is 10, and the length of the gene data feature vector is 20, then the length of the concatenated feature vector is 542.

[0071] Normalization: The fused feature vector obtained after splicing is normalized to balance the contribution of different modalities and ensure that each feature has equal importance in subsequent classification or analysis tasks.

[0072] Specifically, the steps in step S30 for training the cervical cell classification model include:

[0073] Data partitioning: The fused feature data is divided into training and test sets in a 7:3 ratio;

[0074] Cervical cell classification model construction and training: The cervical cell classification model is built based on support vector machines and random forests. Kernel function parameters and model parameters are adjusted to make the model suitable for cervical cell classification calculations. Based on the complexity of cervical cell image features, radial basis functions (RBFs) are chosen as the kernel function. RBFs can handle non-linearly separable data and are effective in capturing the complex relationships between cervical cell features. The penalty factor C and kernel coefficient γ are initialized and dynamically adjusted using a grid search method. The model performance under different parameter combinations is evaluated on the test set to select the optimal parameter combination. The penalty factor C controls the balance between training error and model complexity of the cervical cell classification model. A moderate initial value is set. Initial values, such as C=1; kernel coefficient γ affects the calculation result of radial basis function, and can be initialized to γ=0.1; random forest includes decision trees. Choose an appropriate number of decision trees based on the actual data, such as 100 trees. Too few decision trees may lead to underfitting of the model, while too many will increase the computational cost and may lead to overfitting. Based on the model training results, use a random search method to dynamically adjust the number of decision trees and other random forest parameters, such as the feature subset size. When splitting at the node of each decision tree, randomly select a certain number of features as candidate features from the fused feature data. For cervical cell image fusion features, an appropriate feature subset size can be selected based on the number of features and experience, such as selecting the square root of the total number of features as the feature subset size.

[0075] Cervical cell classification model evaluation and optimization: The performance of the cervical cell classification model is comprehensively evaluated using accuracy, recall, and F1 score. The three evaluation indicators are assigned different weights. For example, for cervical cell classification tasks, recall (especially the recall of abnormal cells) is more important because missing abnormal cells may lead to serious consequences. Therefore, the weight of recall is set at 50%, accuracy at 25%, and F1 score at 25%. The cervical cell classification model evaluation score is obtained through weighted calculation. When the evaluation score is greater than the set qualified value, the optimal classification model is obtained. After determining the model version, it is saved and put into practical use.

[0076] Step S40: The real cervical cell images to be tested are subjected to the data acquisition and preprocessing in step S10 and fused. The data is then input into the optimized cervical cell classification model. The model outputs whether the cervical cells are abnormal and the analysis results of the abnormality type and degree.

[0077] Specifically, in step S40, the model outputs the results of whether cervical cells are abnormal and the analysis of the type and degree of abnormality, including:

[0078] Determining whether cervical cells are abnormal: The cervical cell classification model determines whether cervical cells are normal or abnormal based on the input real cervical cell images. For example, the cervical cell classification model comprehensively analyzes the features of the input cervical cell images and the fused clinical and genetic data features. When its feature pattern highly matches the feature pattern of normal cervical cells in the training set, the model outputs that the cervical cells are normal; when it highly matches the feature pattern of abnormal cervical cells, the model outputs that the cervical cells are abnormal.

[0079] Analysis of abnormal cervical cell results: When the model outputs abnormal cervical cells, the abnormal cervical cell types are further subdivided, and the degree of abnormal cervical cells is assessed through a series of quantitative indicators, including cell nucleus size, degree of irregularity of cell nucleus shape, chromatin density and distribution uniformity, and specific values ​​of nucleocytoplasmic ratio, etc. For example, regarding cell nucleus size, the model calculates its deviation from the average normal cell nucleus size and outputs it as a percentage, such as a cell nucleus size exceeding the normal average by 50%. For irregularity in shape, it quantifies the degree by calculating the ratio of the cell nucleus's perimeter to the perimeter of a circle of the same area; the greater the deviation of this ratio from 1, the more irregular the shape. Chromatin density can be quantified by statistically analyzing the grayscale values ​​of chromatin regions in an image; the higher the degree of abnormality, the higher the chromatin density value and the more uneven the distribution. Based on these quantitative indicators, the model classifies the degree of cervical cell abnormality into different levels, such as mild, moderate, and severe abnormalities. Mild abnormalities are characterized by only slight deviations in cell nucleus size and shape from the normal range, with no significant changes in chromatin. Moderate abnormalities are characterized by more significant deviations in various indicators, with some degree of change in cell morphology. Severe abnormalities are characterized by severe changes in the cell nucleus, chromatin, and overall cell morphology, with a significant difference from normal cells; in this case, the model outputs a severe level of abnormality. This severity classification helps doctors understand the extent of cervical cell abnormalities more clearly, thus enabling them to develop appropriate treatment plans.

[0080] In addition, such as Figure 2 As shown, in one embodiment of the present invention, an AI-based cervical cell image analysis assistance system is proposed. The AI-based cervical cell image analysis assistance system includes:

[0081] Cervical cell multimodal data acquisition module: used to acquire multimodal data, including cervical cell image data, clinical data and patient genetic data, and to preprocess the acquired cervical cell image data;

[0082] Cervical cell image generation and feature extraction module: This module is used to construct a generative adversarial network suitable for cervical cell images. The generator in the generative adversarial network generates cervical cell images, and the cervical cell images generated by the generator and the real cervical cell images collected are input together into a pre-trained convolutional neural network for feature extraction.

[0083] Cervical cell feature fusion and cervical cell classification model training module: This module is used to fuse extracted features with collected clinical data and patient gene data. The fused feature data is divided into a training set and a test set. The training set is used to train the cervical cell classification model, and the test set is used to evaluate the cervical cell classification model. Based on the evaluation results, the model is optimized to obtain the optimized cervical cell classification model.

[0084] Cervical cell abnormality judgment module: It is used to collect and preprocess the real cervical cell images to be tested in step S10, and then fuse them. The data is input into the optimized cervical cell classification model, and the model outputs whether the cervical cells are abnormal and the analysis results of the abnormality type and degree.

[0085] This application provides an AI-based cervical cell image analysis assistance system, employing an AI-based cervical cell image analysis assistance method as described in the above embodiments. This system addresses the technical problems of low efficiency and low accuracy in traditional cervical cell analysis methods. Compared to existing technologies, the beneficial effects of the AI-based cervical cell image analysis assistance system provided in this application are the same as those of the AI-based cervical cell image analysis assistance method provided in the above embodiments. Furthermore, other technical features of the AI-based cervical cell image analysis assistance system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0086] This application provides an AI-based cervical cell image analysis auxiliary device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform an AI-based cervical cell image analysis auxiliary method as described in Embodiment 1 above.

[0087] like Figure 3As shown in the illustration, in one embodiment of the present invention, a structural schematic diagram of an AI-based cervical cell image analysis auxiliary device suitable for implementing the embodiments of this application is presented. An AI-based cervical cell image analysis auxiliary device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), etc., as well as fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The AI-based cervical cell image analysis aid shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments in this application.

[0088] Figure 3 The AI-based cervical cell image analysis auxiliary device shown may include a processing system 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage system 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the AI-based cervical cell image analysis auxiliary device. The processing system 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input systems 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output systems 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage systems 1003 including, for example, magnetic tapes, hard disks, etc.; and communication systems 1009. Communication system 1009 allows an AI-based cervical cell image analysis aid to wirelessly or wiredly communicate with other devices to exchange data. Although the figure shows an AI-based cervical cell image analysis aid with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0089] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication system, or installed from storage system 1003, or installed from ROM 1002. When the computer program is executed by processing system 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0090] This application provides an AI-based cervical cell image analysis auxiliary device, employing an AI-based cervical cell image analysis auxiliary method as described in the above embodiments. This method addresses the technical problems of low efficiency and low accuracy in traditional cervical cell analysis methods. Compared to existing technologies, the beneficial effects of the AI-based cervical cell image analysis auxiliary device provided in this application are the same as those of the AI-based cervical cell image analysis auxiliary method provided in the above embodiments. Furthermore, other technical features of this AI-based cervical cell image analysis auxiliary device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0091] The various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0092] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the AI-based cervical cell image analysis assistance method described above.

[0093] The computer program product provided in this application can solve the technical problems of low efficiency and low accuracy in traditional cervical cell analysis methods. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the AI-based cervical cell image analysis auxiliary method provided in the above embodiments, and will not be repeated here.

[0094] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An AI-based cervical cell image analysis-assisted method, characterized in that, The method includes the following steps: Step S10: Perform multimodal data acquisition, including cervical cell image data, clinical data and patient genetic data, and preprocess the acquired cervical cell image data; Step S20: Construct a generative adversarial network suitable for cervical cell images. Generate cervical cell images through the generator in the generative adversarial network. Input the cervical cell images generated by the generator and the collected cervical cell images into a pre-trained convolutional neural network for feature extraction. Step S30: The extracted features are fused with the collected clinical data and patient gene data. The fused feature data is divided into a training set and a test set. The training set is used to train the cervical cell classification model, and the test set is used to evaluate the cervical cell classification model. The model is optimized based on the evaluation results to obtain the optimized cervical cell classification model. Step S40: The real cervical cell images to be tested are subjected to the data acquisition and preprocessing in step S10 and fused. The data is then input into the optimized cervical cell classification model. The model outputs whether the cervical cells are abnormal and the analysis results of the abnormality type and degree. In step S20, a generative adversarial network suitable for cervical cell images is constructed, including a generator and a discriminator. The step of constructing the discriminator includes: Multi-scale feature extraction input layer: The input of the discriminator includes the cervical cell image itself received by the discriminator and the feature maps of the cervical cell image at different scales. The features of the cervical cell image have different manifestations at different scales. By using convolution kernels of different sizes to process the input cervical cell images in parallel, cervical cell image feature maps at different scales are obtained. These cervical cell image feature maps are stitched together as the input of the discriminator. Cervical cell morphology-based feature pooling layer: A cervical cell morphology-based feature pooling layer is added between the convolutional layers. This pooling layer performs pooling operations on the feature map based on the morphological information of cervical cells, retaining features related to cervical cell morphology. The output layer incorporates clinical knowledge constraints: the discriminator's output judges the authenticity of images based on their own features and is constrained by clinical knowledge. This is achieved by introducing a clinical knowledge module into the discriminator's output layer, which contains information on the ratio of normal to abnormal cervical cells. The discrimination results are then adjusted based on this ratio information. The discriminator loss function incorporates cell distribution information: The discriminator's loss function incorporates the distribution information of cervical cells in the image to distinguish between real and fake cervical cell images. In real cervical cell images, the distribution of cells follows a certain pattern. By calculating the difference in cell distribution between the generated image and the real image, the Wasserstein distance is used to measure the similarity of cervical cell distribution and is incorporated into the discriminator's loss function.

2. The AI-based cervical cell image analysis-assisted method according to claim 1, characterized in that, In step S10, multimodal data acquisition is performed, including cervical cell image data, clinical data, and patient genetic data. The acquired cervical cell image data is then preprocessed, including the following steps: Cervical cell image data acquisition: Cervical cell images are acquired using a microscope, and relevant parameters during the acquisition of cervical cell images are recorded, including magnification and exposure time; Clinical information collection: Collect clinical information from patients who provide cervical cells, including age, medical history, and symptoms; Genetic data collection: Collect relevant genetic data from patients who provide cervical cells, including gene expression levels closely related to the development of cervical cancer; Cervical cell image preprocessing: The acquired cervical cell images are preprocessed, including image denoising, contrast enhancement, and normalization. Clinical information and genetic data are standardized, and feature vectors of clinical information and genetic data are extracted and effectively fused with the cervical cell images.

3. The AI-based cervical cell image analysis-assisted method according to claim 1, characterized in that, In step S20, a generative adversarial network suitable for cervical cell images is constructed, including a generator and a discriminator. The step of constructing the generator includes: The input layer integrates cervical cell features: while inputting traditional random noise vectors, the generator's input layer also incorporates multimodal data closely related to the biological characteristics of cervical cells, including extracting cell density information and statistical features of cell size from cervical cells. These features are encoded into vectors and concatenated with random noise as the input to the generator. Deconvolutional layer with attention mechanism: Introducing attention mechanism into deconvolutional layer. In cervical cell images, the nucleus and cytoplasm regions have different importance. Through attention mechanism, the generator focuses more on the image generation of key regions. The output layer simulates cervical cell texture: A special combination of activation functions is used in the output layer to make the generated cervical cell image closer to the texture of the real cervical cell image. The Tanh function is used to limit the output value to the range of -1 to 1, and a texture enhancement layer is added. This texture enhancement layer extracts the texture features of the image by performing a convolution operation on the output image and fuses it with the generated image of the output layer to enhance the texture details of the generated image and simulate the texture features in the cervical cell image. The generator loss function combines cervical cell feature matching: The generator loss function includes traditional cross-entropy loss and cell feature matching loss. It calculates the difference between the generated cervical cell image and the real cervical image in key cell features, uses mean squared error loss to measure the difference of these features, and sums them with cross-entropy loss as the total loss of the generator.

4. The AI-based cervical cell image analysis assistance method according to claim 1, characterized in that, In step S30, the extracted features are fused with the collected clinical data and patient gene data using a splicing fusion method. The steps include: Concatenation and fusion: The feature vectors of cervical cell images extracted by the pre-trained convolutional neural network, the pre-processed clinical information feature vectors and gene data feature vectors are concatenated in sequence to form a new fused feature vector; Normalization: The fused feature vector obtained after splicing is normalized.

5. The AI-based cervical cell image analysis assistance method according to claim 1, characterized in that, The step of training the cervical cell classification model in step S30 includes: Data partitioning: The fused feature data is divided into training and test sets in a 7:3 ratio; Cervical cell classification model construction and training: The cervical cell classification model is built based on support vector machine and random forest. The kernel function parameters and model parameters are adjusted to make the model suitable for cervical cell classification calculation. Radial basis function is selected as the kernel function. The penalty factor C and kernel coefficient γ are initialized. The penalty factor C and kernel coefficient γ are dynamically adjusted using grid search. The model performance under different parameter combinations is evaluated on the test set to select the optimal parameter combination. The penalty factor C controls the balance between training error and model complexity of the cervical cell classification model. An initial value is set first. The kernel coefficient γ affects the calculation result of the radial basis function and is initialized to γ ​​= 0.

1. The random forest includes decision trees. An appropriate number of decision trees is selected according to the actual data. The number of decision trees and other random forest parameters are dynamically adjusted using random search based on the model training results. When splitting at the node of each decision tree, a certain number of features are randomly selected as candidate features from the fused feature data. Cervical cell classification model evaluation and optimization: The performance of the cervical cell classification model is comprehensively evaluated using accuracy, recall and F1 score. The three evaluation indicators are weighted differently and the evaluation score of the cervical cell classification model is obtained by weighted calculation. When the evaluation score is greater than the set qualified value, the optimal classification model is obtained.

6. The AI-based cervical cell image analysis-assisted method according to claim 1, characterized in that, In step S40, the model outputs the results of whether the cervical cells are abnormal and the analysis of the type and degree of abnormality, including: Determination of cervical cell abnormality: The cervical cell classification model determines whether cervical cells are normal or abnormal based on the input real cervical cell image. Analysis of abnormal cervical cell results: When the model outputs abnormal cervical cells, the abnormal cervical cell types are further subdivided, and the degree of abnormal cervical cells is assessed through a series of quantitative indicators, including the specific values ​​of cell nucleus size, degree of irregularity of cell nucleus shape, chromatin density and distribution uniformity, and nucleocytoplasmic ratio. Based on the above quantitative indicators, the model classifies the degree of abnormal cervical cells into different levels.

7. An AI-based cervical cell image analysis assistance system, characterized in that, The AI-based cervical cell image analysis assistance system includes: Cervical cell multimodal data acquisition module: used to acquire multimodal data, including cervical cell image data, clinical data and patient genetic data, and to preprocess the acquired cervical cell image data; Cervical cell image generation and feature extraction module: This module is used to construct a generative adversarial network suitable for cervical cell images. The generator in the generative adversarial network generates cervical cell images, and the cervical cell images generated by the generator and the real cervical cell images collected are input together into a pre-trained convolutional neural network for feature extraction. Cervical cell feature fusion and cervical cell classification model training module: This module is used to fuse extracted features with collected clinical data and patient gene data. The fused feature data is divided into a training set and a test set. The training set is used to train the cervical cell classification model, and the test set is used to evaluate the cervical cell classification model. Based on the evaluation results, the model is optimized to obtain the optimized cervical cell classification model. Cervical cell abnormality judgment module: It is used to collect and preprocess the data of the real cervical cell image to be tested in step S10 and fuse it, input it into the optimized cervical cell classification model, and the model outputs whether the cervical cells are abnormal and the analysis results of the abnormality type and degree. The steps for constructing the discriminator in the cervical cell image generation and feature extraction module include: Multi-scale feature extraction input layer: The input of the discriminator includes the cervical cell image itself received by the discriminator and the feature maps of the cervical cell image at different scales. The features of the cervical cell image have different manifestations at different scales. By using convolution kernels of different sizes to process the input cervical cell images in parallel, cervical cell image feature maps at different scales are obtained. These cervical cell image feature maps are stitched together as the input of the discriminator. Cervical cell morphology-based feature pooling layer: A cervical cell morphology-based feature pooling layer is added between the convolutional layers. This pooling layer performs pooling operations on the feature map based on the morphological information of cervical cells, retaining features related to cervical cell morphology. The output layer incorporates clinical knowledge constraints: the discriminator's output judges the authenticity of images based on their own features and is constrained by clinical knowledge. This is achieved by introducing a clinical knowledge module into the discriminator's output layer, which contains information on the ratio of normal to abnormal cervical cells. The discrimination results are then adjusted based on this ratio information. The discriminator loss function incorporates cell distribution information: The discriminator's loss function incorporates the distribution information of cervical cells in the image to distinguish between real and fake cervical cell images. In real cervical cell images, the distribution of cells follows a certain pattern. By calculating the difference in cell distribution between the generated image and the real image, the Wasserstein distance is used to measure the similarity of cervical cell distribution and is incorporated into the discriminator's loss function.

8. An AI-based cervical cell image analysis auxiliary device, characterized in that, The AI-based cervical cell image analysis auxiliary device includes: The present invention includes a memory, a processor, and an AI-based cervical cell image analysis assistance program stored in the memory and executable on the processor, wherein the AI-based cervical cell image analysis assistance program, when executed by the processor, implements an AI-based cervical cell image analysis assistance method as described in any one of claims 1 to 6.

9. A computer program product, characterized in that, The computer program product includes an AI-based cervical cell image analysis assistance program, which, when executed by a processor, implements an AI-based cervical cell image analysis assistance method as described in any one of claims 1 to 6.

Images