Subregion-based cervical cancer colposcope image cancerous tissue discrimination system
By using a subregion-based colposcopy image discrimination system for cervical cancer, K-means clustering and deep learning techniques are employed to address the lack of sensitivity and specificity in existing technologies for early, small lesions. This achieves efficient and accurate diagnosis of cervical cancer tissue and reduces reliance on physician experience.
Patent Information
- Application Number
- CN202510778051.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-11
- Publication Date
- 2025-11-11
AI Technical Summary
Current colposcopy-based diagnostic methods for cervical cancer ignore the spatial heterogeneity of lesions and subtle differences between subregions in images at the level of a single image or the overall ROI, resulting in insufficient sensitivity and specificity for early, small lesions and a high dependence on physician experience.
A sub-region-based colposcopy image discrimination system for cervical cancer was adopted. Through data preprocessing, sub-region processing, discrimination classification and model training modules, K-means clustering, VGG16 convolutional neural network, LASSO algorithm and logistic regression model were used, combined with SMOTE oversampling technology to achieve fine-grained analysis and classification of cervical cancer pathological images.
It improves the sensitivity and specificity for early, minor lesions, reduces reliance on physician experience, enhances diagnostic accuracy and stability, is suitable for multi-source data environments, and is easy to integrate and expand clinically.
Smart Images
Figure CN120931553A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of image processing and deep learning, and in particular to a subregion-based system for identifying cancerous tissue in colposcopy images of cervical cancer. Background Technology
[0002] Clinically, colposcopy is an important tool for identifying and intervening in the early stages of cervical precancerous lesions. However, its diagnostic accuracy is highly dependent on the physician's professional experience, is subjective, and has a sensitivity of only about 48%. Currently known colposcopy-assisted diagnostic methods for cervical cancer extract features only from a single image or the overall ROI (Region of Interest), ignoring the spatial heterogeneity of lesions in the image and the subtle differences between subregions, resulting in insufficient sensitivity and specificity for early, small lesions. Summary of the Invention
[0003] The purpose of this invention is to address the shortcomings of existing technologies and, in order to fully utilize the advantages of subregion features, to propose a subregion-based system for identifying cancerous tissue in colposcopy images of cervical cancer.
[0004] The discrimination system proposed in this invention includes the following modules:
[0005] Data preprocessing module: This module is used to annotate the acquired cervical cancer colposcopy images, delineate lesion areas, and obtain mask files. It preprocesses the image data to obtain a cervical cancer pathology image dataset (including images and masks).
[0006] Sub-region processing module: This module applies a mask to the preprocessed image, retaining only the region of interest (ROI). Then, it uses the K-means clustering algorithm to divide the ROI into multiple sub-regions, thus splitting the preprocessed image into multiple sub-regions.
[0007] Discrimination and classification module: The cervical cancer pathology image discrimination network is used to discriminate and classify each input sub-region image.
[0008] Model training module: The sub-region images obtained by the sub-region processing module are used to train the cervical cancer pathology image discrimination network of the discrimination and classification module, outputting the classification accuracy of each sub-region, and selecting the sub-region with the highest accuracy, i.e. the best performance.
[0009] The output module processes the cervical cancer pathology image to be identified through the data preprocessing module and the sub-region processing module to obtain the sub-region with the best performance. The trained cervical cancer pathology image discrimination network is then used to make a discrimination, and the final discrimination result of cancerous tissue is output.
[0010] In one possible implementation, the data preprocessing module employs a preprocessing method including image resampling to a specified interval; after resampling, the image undergoes denoising (Gaussian smoothing). Then, the image is mean-centered. Finally, the image is Z-score normalized.
[0011] In one possible implementation, the cervical cancer pathology image discrimination network of the discrimination and classification module is constructed based on a VGG16 convolutional neural network, the LASSO algorithm, and a logistic regression model. The discrimination and classification operation is as follows:
[0012] First, features of each subregion are extracted using a VGG16 convolutional neural network with the last softmax layer removed. These features are then checked for missing values, extreme values, and Z-score normalization. Next, the LASSO algorithm is used to filter out important features related to cancerous tissue identification. Finally, a logistic regression model is used for classification to obtain the classification results for each subregion image.
[0013] In one possible implementation, during the training process of the model training module, the SMOTE (Synthetic Minority Over-sampling Technique) algorithm is used to oversample the images in the cervical cancer pathology image dataset to balance the class distribution. The loss function used is the cross-entropy loss function.
[0014] The beneficial effects of this invention are as follows:
[0015] 1. Subregional fine-grained analysis enhances the ability to detect early, minute lesions.
[0016] By subdividing the ROI region in colposcopic images into multiple sub-regions, it is possible to fully explore the subtle differences between different tissue types, significantly improving the sensitivity and specificity for early and small lesions.
[0017] 2. Multi-stage feature screening enhances the accuracy of discrimination.
[0018] By combining a three-level discrimination process of deep feature extraction (based on VGG16), LASSO feature selection, and logistic regression classification, we can not only retain the powerful representation ability of deep networks for image semantic information, but also eliminate redundant features through L1 regularization, thus achieving accurate and efficient organization discrimination.
[0019] 3. SMOTE oversampling to balance class distribution.
[0020] Introducing SMOTE technology during the model training phase to perform synthetic oversampling on minority class (cancer tissue) samples significantly alleviates the data imbalance problem, improves the model's detection rate of cancerous areas, and reduces the risk of missed diagnoses.
[0021] 4. Standardized preprocessing enhances the model's generalization ability.
[0022] This invention systematically completes resampling, denoising, mean decentering, and Z-score standardization in the data preprocessing module, unifying the image input scale and distribution, reducing interference caused by differences in equipment and shooting conditions, and improving the applicability of the model to multi-source data.
[0023] 5. Modular design, easy to integrate and expand in clinical settings.
[0024] The overall system is divided into five modules: data preprocessing, sub-region processing, discrimination and classification, model training, and result output. This facilitates subsequent function upgrades and algorithm replacements, as well as integration with existing colposcope equipment and hospital information systems, enabling rapid deployment and promotion.
[0025] 6. Reduce reliance on doctors' experience and minimize subjective errors.
[0026] Through automated and quantified image processing and discrimination processes, the reliance on doctors' subjective experience in manual diagnosis is effectively reduced, enabling stable and reliable auxiliary diagnostic support in low- and middle-income areas or in environments with limited experience, thereby reducing the misdiagnosis rate. Attached Figure Description
[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.
[0028] Figure 1 This is a schematic diagram of the system structure and operation process of the present invention. Detailed Implementation
[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of this application.
[0030] Computer-aided diagnosis of medical images generally involves dataset acquisition, preprocessing, feature extraction, feature selection, and diagnostic classification. Different medical image computer diagnostic systems can vary significantly in their processes, such as the algorithms used for feature extraction and feature selection. Therefore, this invention proposes a sub-region-based discrimination system for cancerous tissue in colposcopy images of cervical cancer. By extracting features from sub-regions of the image, it effectively improves the accuracy of cervical cancer lesion classification, which has great potential for clinical applications.
[0031] This application provides a method for identifying cancerous tissue in cervical cancer colposcopy images based on subregions, including the following steps:
[0032] Step 1: Data preprocessing.
[0033] Doctors annotate the acquired colposcopy images of cervical cancer, delineate the lesion areas, and obtain mask files. The image data undergoes preprocessing, including resampling, denoising (Gaussian smoothing), mean decentering, and Z-score normalization, to obtain a cervical cancer pathology image dataset (including images and masks).
[0034] Step 2: Sub-region processing.
[0035] A mask is applied to the preprocessed image to retain only the region of interest (ROI). Then, the K-means clustering algorithm is used to divide the ROI regions, dividing the preprocessed image into multiple sub-regions. Each sub-region corresponds to one image, meaning that each patient corresponds to multiple sub-region images.
[0036] Step 3: Classify the data.
[0037] A cervical cancer pathology image discrimination network based on a VGG16 convolutional neural network, the LASSO algorithm, and a logistic regression model was used to classify each input sub-region image. First, the VGG16 convolutional neural network with the last softmax layer removed extracted features for each sub-region. These features were then processed (including missing value checking, extreme value checking, and Z-score normalization). Next, the LASSO algorithm was used to filter out important features related to cancerous tissue discrimination. Finally, a logistic regression model was used for classification to obtain the classification results for each sub-region image.
[0038] Step 4: Model training.
[0039] The sub-region images obtained by the sub-region processing module are used to train the cervical cancer pathology image discrimination network of the discrimination and classification module. The classification accuracy of each sub-region is output, and the sub-region with the highest accuracy, i.e. the best performance, is selected.
[0040] Step 5: Output the results.
[0041] The cervical cancer pathological images to be identified are processed sequentially through the data preprocessing module and the sub-region processing module to obtain the sub-region with the best performance. The cervical cancer pathological image discrimination network of the trained discrimination and classification module is used to make a discrimination and output the final discrimination result of cancerous tissue.
[0042] In one possible implementation, the data preprocessing process includes image resampling, sampling to a specified interval. Assume the original interval is... The target spacing is Original image size is The new image size The calculation formula is:
[0043]
[0044] After resampling, the image will undergo denoising (Gaussian smoothing). The formula is as follows:
[0045]
[0046] in, The image after resampling. For the RGB channels of the target pixel, For relative shift within the nucleus, The radius is the core radius.
[0047] Then, the image is centered using the mean. Using the RGB channel mean values (123.68, 116.779, 103.939) from ImageNet pre-training, the following operation is performed:
[0048]
[0049] in, The RGB channel mean values used for ImageNet pre-training.
[0050] Finally, the image is normalized using Z-Score, as shown in the following formula:
[0051]
[0052] in, The original image intensity, The mean, The standard deviation is denoted as .
[0053] In one possible implementation, during the sub-region processing, a pre-annotated mask file is used to preserve the Region of Interest (ROI) in the image. The pixel values of the ROI are then used as input to a K-means algorithm to divide the ROI into K sub-regions. The formula is as follows:
[0054]
[0055] in, For the first Cluster centers, For the first The centroid of each cluster center.
[0056] In one possible implementation, the LASSO algorithm is used for feature selection during the discriminative classification process to improve the model's generalization ability and avoid overfitting. The principle of LASSO regression is to minimize the sum of squared residuals by adding an L1 regularization term, causing some regression coefficients to shrink to zero, thus achieving feature selection. The objective function is as follows:
[0057]
[0058] in, The regularization parameter controls the sparsity of feature selection. is the regression coefficient.
[0059] In one possible implementation, during model training, the SMOTE (Synthetic Minority Over-sampling Technique) algorithm is used to oversample images in the cervical cancer pathology image dataset. The aim is to address the class imbalance problem and prevent the model from being insufficiently able to distinguish the minority class (cancer tissue).
[0060] The SMOTE algorithm first identifies the minority class samples in the dataset. Then, for each minority class sample, it finds its K nearest neighbors. A new sample is generated by randomly selecting one of the K nearest neighbors and connecting it along a line in the feature space. The specific formula is as follows:
[0061]
[0062] in, For the original minority class samples, Its nearest neighbor samples It is a random number between [0,1].
[0063] In the model training module, the classification model used is Logistic Regression. The model formula is as follows:
[0064]
[0065] in, This represents the probability that the sample belongs to cancerous tissue. , ,..., For input features, , ,..., These are the model parameters.
[0066] The loss function used is the cross-entropy loss function, as shown in the following formula:
[0067]
[0068] in, It is the sample size. It is the first The true label (0 or 1) of each sample. It is the first The predicted probability of a sample.
[0069] refer to Figure 1 As shown in the embodiments of this application, a system for identifying cancerous tissue in cervical cancer colposcopy images based on subregions is also provided, comprising the following modules:
[0070] Data Preprocessing Module: Acquire cervical acetic acid images (cervical cancer pathology images) using a high-resolution cervical endoscopy device. Ensure the device has good image quality, capable of capturing subtle tissue changes. Save acquired images in standard PNG format to avoid information loss due to compression. Each image should include necessary metadata, such as acquisition time and patient medical record number. Acquire at least 1000 high-quality images to ensure the diversity and representativeness of the dataset.
[0071] Sub-region processing module: The region of interest in the image is further divided into 3 sub-regions using the K-means algorithm to distinguish different types of tissue (such as normal tissue, mild lesions, and severe lesions).
[0072] The discriminant classification module uses a VGG16 convolutional neural network, the LASSO algorithm, and a logistic regression model to construct a discriminant network for cervical cancer pathology images. The final softmax layer of the VGG16 is removed, resulting in a 1000-dimensional feature vector as the model's output. These features are processed, including missing value checking, extreme value checking, and Z-score normalization. The vector is then fed into the LASSO algorithm to filter out features useful for classification, and finally, it is input into the logistic regression model to output the discrimination result.
[0073] Model training module: The dataset is input into the cervical cancer pathology image discrimination network for training, and the classification accuracy for each subregion is output. SMOTE oversampling is applied to the dataset to balance class classification.
[0074] The results output module selects the sub-region with the best classification accuracy and uses a pre-trained cervical cancer pathology image discrimination network to discriminate the specified sub-regions in the cervical cancer pathology images that have passed through the data preprocessing and sub-region processing modules. Based on the model's output, it determines whether cancerous lesions exist in the image.
[0075] In one possible implementation, in the sub-region processing module, the pixel values of the ROI region are used as input data for the K-means algorithm, and the settings are... (In clinical practice, adjustments can be made based on the doctor's experience.) Corresponding to the three sub-regions, cluster centers are repeatedly assigned and updated until convergence. The output consists of images and mask files for the three sub-regions.
[0076] In one possible implementation, the loss function of the LASSO algorithm in the discriminative classification module is as follows:
[0077]
[0078] in, for Regularization term, This is the regularization coefficient or penalty coefficient. The LASSO algorithm in this module... Regularization coefficient The value was chosen to be 5 to prevent overfitting.
[0079] In one possible implementation, the discriminant classification module incorporates elements from the logistic regression model. The regularization term and loss function are as follows:
[0080]
[0081] in, It refers to the number of samples. It is the first The true label (0 or 1) of each sample. It is the predicted probability of logistic regression. yes Regularization term, This is the regularization coefficient or penalty coefficient. In the logistic regression model of this module... Regularization coefficient The value is selected as 1000.
[0082] In one possible implementation, the number of nearest neighbors in the model training module of the SMOTE algorithm The value is set to 5. After the data is processed by the sub-region processing module, it is processed by the SMOTE algorithm and then input into the cervical cancer pathology image discrimination network for training. The classification accuracy of each sub-region is output and the best sub-region is selected.
[0083] This application also provides an electronic device, which includes a processor and a memory;
[0084] The memory is used to store computer programs;
[0085] When the processor executes a program stored in the memory, it implements any of the methods described in this application.
[0086] In one possible implementation, the electronic device of this application embodiment further includes a communication interface and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.
[0087] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not indicate that there is only one bus or one type of bus.
[0088] The communication interface is used for communication between the aforementioned electronic devices and other devices.
[0089] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0090] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0091] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements any of the methods described in this application.
[0092] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the methods described in this application.
[0093] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).
[0094] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0095] The various embodiments in this specification are described in a related manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other.
[0096] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.
Claims
1. A system for identifying cancerous tissue in cervical cancer colposcopy images based on subregions, characterized in that, Includes the following modules: Data preprocessing module: used to annotate the acquired cervical cancer colposcopy images, delineate lesion areas, and obtain mask files; preprocess the image data to obtain a cervical cancer pathology image dataset; Sub-region processing module: This module applies a mask to the preprocessed image, retaining only the region of interest; then, it uses the K-means clustering algorithm to divide the region of interest into multiple sub-regions. The discrimination and classification module uses a cervical cancer pathology image discrimination network to classify each input sub-region image. Model training module: The sub-region images obtained by the sub-region processing module are used to train the cervical cancer pathology image discrimination network of the discrimination and classification module, outputting the classification accuracy of each sub-region, and selecting the sub-region with the highest accuracy, i.e. the best performance. The output module processes the cervical cancer pathology image to be identified through the data preprocessing module and the sub-region processing module to obtain the sub-region with the best performance. The trained cervical cancer pathology image discrimination network is then used to make a discrimination, and the final discrimination result of cancerous tissue is output.
2. The system for identifying cancerous tissue in cervical cancer colposcopy images based on subregions, as described in claim 1, is characterized in that... The data preprocessing module employs a preprocessing method that includes image resampling to a specified interval; after resampling, the image is denoised; then, the image is mean-centered; and finally, the image is Z-score normalized.
3. The system for identifying cancerous tissue in cervical cancer colposcopy images based on subregions, as described in claim 1, is characterized in that... The cervical cancer pathology image discrimination network in the discrimination and classification module is built based on a VGG16 convolutional neural network, the LASSO algorithm, and a logistic regression model. The discrimination and classification operations are as follows: First, features of each subregion are extracted using a VGG16 convolutional neural network with the last softmax layer removed. These features are then checked for missing values, extreme values, and Z-score normalization. Next, the LASSO algorithm is used to filter out important features related to cancerous tissue identification. Finally, a logistic regression model is used for classification to obtain the classification results for each subregion image.
4. The system for identifying cancerous tissue in cervical cancer colposcopy images based on subregions, as described in claim 1, is characterized in that... During the training process of the model training module, the SMOTE algorithm is used to oversample the images in the cervical cancer pathology image dataset to balance the class distribution; the cross-entropy loss function is used as the loss function.