Cell morphology feature extraction and predictive analysis method for high-throughput cervical TCT image
By using a UNet-based cervical cell segmentation network and morphological feature extraction method, the problems of low manual efficiency and insufficient automated throughput in cervical cancer screening are solved, achieving efficient and automated cervical cancer screening and providing interpretable quantitative data support.
Patent Information
- Application Number
- CN202511351388.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-22
- Publication Date
- 2026-01-09
AI Technical Summary
Existing cervical cancer screening methods rely on manual microscopic examination, which is inefficient, time-consuming, and highly influenced by subjective experience. Traditional image processing algorithms struggle to handle cell overlap and noise issues, and automated systems have low throughput, failing to meet the needs of large-scale screening.
Image segmentation was performed using a UNet-based cervical cell segmentation network, combined with shape and texture feature extraction and calculation. Initial screening was conducted using the Bethesda cervical cytology system standard. The segmentation network was used to obtain the segmentation results of the cell nucleus and cytoplasm, and a morphological feature extraction and calculation module was constructed to achieve efficient and automated analysis.
It has improved the automation level of cervical cancer screening, significantly increased screening efficiency, provided medically interpretable quantitative data support, and promoted the development of intelligent assisted diagnosis.
Smart Images
Figure CN121304545A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to a cell morphological feature extraction and prediction analysis method, belonging to the field of medical image analysis and computer vision technology. BACKGROUND
[0002] Cervical cancer is one of the most common cancers in women, with high incidence and mortality, ranking fourth in the world. The occurrence of cervical cancer is a long-term and slow evolution process, and accurate cervical cancer screening is of great significance for developing diagnosis and treatment programs and prognosis. The existing cervical cancer screening methods are mainly cytological examination and HPV detection. With the development of modern cytological technology, thin-layer cytological examination (TCT) is widely used in domestic early screening of cervical cancer. Cervical TCT examination collects epithelial cells that fall off from the cervix, prepares and stains, and reads under a microscope, and makes a diagnosis according to the Bethesda system (TBS) standard. However, manual microscopy is inefficient, and traditional TCT detection relies on pathologists to manually observe cell morphology under a microscope, and single sample analysis takes as long as 10-15 minutes, and is greatly influenced by subjective experience. The shortage of professionals, high training cost, and long-term high-intensity reading of doctors are prone to missed detection and misdiagnosis.
[0003] With the progress of digital whole slide image scanning instruments, accurate and efficient computer-aided cervical cancer screening becomes possible. Whole slide image (WSI) is obtained by scanning the slide with an optical microscope. The cytological WSI usually has thousands of cells that have little structural correlation with each other, and even a single isolated pathological cell may be considered as an abnormal sample. With the help of computer-aided screening system, pathologists can avoid browsing billions of pixels and significantly improve screening efficiency. Existing cell morphological analysis methods, cervical cell automatic analysis system mostly use traditional image processing algorithms (such as threshold segmentation, edge detection), which are difficult to handle problems such as cell overlap, uneven staining, and background noise. At the same time, the existing automatic system has the limitation of low throughput, and the processing time of the sample is longer, which cannot meet the demand of large-scale screening. A perfect cervical cell feature extraction and calculation analysis method can provide detailed cytological features with clinical significance, thereby further supporting fine-grained cervical cell identification.
[0004] Therefore, it is urgent to propose a high-throughput cervical TCT image cell morphological feature extraction and calculation analysis method to solve the above technical problems. SUMMARY
[0005] The present application is to solve the problem that the cell analysis technology of cervical TCT image does not fully consider the complex morphological characteristics of cells, the feature extraction accuracy is low, and the artificial dependence is high, and further proposes a high-throughput cervical TCT image cell morphological feature extraction and prediction analysis method.
[0006] The technical solution adopted by the present invention to solve the above problems is as follows: The steps of the present invention include: Step 1: Select cellular morphological features for cervical TCT image-assisted diagnosis; Step 2: Obtain the segmentation results of the cell nucleus and cytoplasm based on the segmentation network; Step 3: Extract and calculate cell morphological features based on the cervical TCT image segmentation results; Step 4: Analysis and visualization of cell morphology characteristic measurement results.
[0007] Furthermore, step 1 specifically includes: Step 101: Perform initial screening of cervical cell morphological characteristics and define core morphological characteristic indicators of squamous intraepithelial lesions. Step 102: Based on clinical pathological standards, cervical cell morphological characteristics are classified into two main categories: shape characteristics and texture characteristics.
[0008] Furthermore, step 2 specifically includes: Step 201: Load the dataset, randomly divide it into training, validation and test sets according to the proportion, and perform data augmentation on the training data; Step 202: Build and train a UNet-based cervical cell segmentation network. The basic architecture consists of two parts: an encoder and a decoder, forming a U-shaped symmetrical structure. Tune the parameters and optimize the model, and evaluate the model performance. Step 203: Obtain the segmentation results of cell nuclei and cytoplasm in the test set, and visualize the mask.
[0009] Furthermore, step 3 specifically includes: Step 301: Load the data and the segmentation results obtained in step 2, and read the cervical TCT cell image and its corresponding cell nucleus and cytoplasm mask; Step 302: Construct a cell morphology feature extraction and calculation module, including a shape feature calculation module and a texture feature calculation module, and integrate a batch processing module; Step 303: Input the data from step 301 into the module of step 302 in batches, and output and record the morphological characteristic measurement results of the cell nucleus and cytoplasm in tabular form.
[0010] Furthermore, step 4 specifically includes: Step 401: Based on the TBS standard and using the size of the middle layer cell nucleus as a reference baseline, calculate the multiple of the cell nucleus relative to the baseline in batches, and perform an overall analysis of the shape feature measurement results in Step 3. Step 402: Visualize the texture feature calculation results in Step 3, draw GLCM histograms of cell nuclei and cytoplasm respectively, and analyze the degree of lesion.
[0011] The beneficial effects of this invention are: 1. This invention integrates medical prior knowledge with a data-driven research method, which is different from traditional manual or purely data-driven methods. It ensures that the features have both medical interpretability and statistical significance through initial screening using clinical standards. 2. This invention can efficiently and automatically extract and calculate morphological features from cervical TCT images, significantly improving the automation level of cervical cancer screening and providing technical support for large-scale cervical cancer screening. 3. Based on the effectively extracted cell morphological features, this invention provides more interpretable quantitative data for subsequent cervical cell research through the joint analysis of shape and texture features. 4. This invention relates to the fields of image processing technology and medical auxiliary diagnosis, and has a positive promoting effect on intelligent auxiliary diagnosis. Attached Figure Description
[0012] Figure 1 This is a flowchart of a method for extracting and analyzing cellular morphological features from high-throughput cervical TCT images. Figure 2 This is a schematic diagram of the cellular morphological features of selected cervical TCT images; Figure 3 This is a schematic diagram of obtaining segmentation results based on a segmentation network; Figure 4 This is a schematic diagram illustrating the extraction and calculation of cell morphological features based on segmentation results; Figure 5 This is a schematic diagram illustrating the analysis and visualization of the results of cell morphology characteristic measurement. Figure 5 (a) is a schematic diagram of the GLCM histogram of cervical cytoplasm; Figure 5 (b) is a schematic diagram of the GLCM histogram of cervical cell nuclei. Detailed Implementation
[0013] Specific implementation method one: as follows Figures 1 to 5 As shown, a method for extracting and analyzing cellular morphological features from high-throughput cervical TCT images includes the following steps: Step 1: Select cellular morphological features for cervical TCT image-assisted diagnosis; Step 2: Obtain the segmentation results of the cell nucleus and cytoplasm based on the segmentation network; Step 3: Extract and calculate cell morphological features based on the cervical TCT image segmentation results; Step 4: Analysis and visualization of cell morphology characteristic measurement results.
[0014] In some embodiments: such as Figure 2As shown, step 1 specifically includes: Step 101: Perform initial screening of cervical cell morphology using the Bethesda Reporting System (TBS) for cervical cytology. Squamous cells located in different positions of the cervical epithelium have different characteristics, and various abnormal cervical cells exhibit even greater morphological diversity, such as... Figure 2 As shown, the characteristic changes in cancerous cells are mainly concentrated in the nucleus and cytoplasm, such as enlarged nuclei, increased nucleus-to-cytoplasm ratio (N / C), and irregular nuclear membrane contours. Compared to normal cervical cells, the nuclei of cervical cancerous cells are more heterogeneous, with more complex boundaries and disordered texture, and the cytoplasm of cervical cancerous cells is more irregular. Cytopathologists typically use the nuclear size of the middle layer cells as a baseline reference for judging abnormalities in other cells. Step 102: Based on the above features, classification can be divided into two main categories: shape features (area, perimeter, roundness, rectangularity, aspect ratio, and nucleocytoplasmic ratio of the cell nucleus and cytoplasm) and texture features (contrast, uniformity, energy, and correlation based on the gray-level co-occurrence matrix (GLCM). For texture features, high contrast means a coarser texture, high uniformity indicates a smoother image, higher energy values indicate a more regular texture, and high correlation indicates stronger correlation between pixels. Therefore, cancerous cells will have higher contrast, lower uniformity, and lower energy. In some embodiments: such as Figure 3 As shown, step 2 specifically includes: Step 201: Use the image after image preprocessing as the training dataset for the segmentation network and create corresponding label data; use several methods such as randomly adjusting brightness, saturation, and contrast to perform data augmentation on the cervical cell image and normalize the pixel value range; Step 202: Build and train a UNet-based cervical cell segmentation network. The basic architecture consists of two parts: an encoder and a decoder, forming a U-shaped symmetrical structure. Tune the parameters and optimize the model, and evaluate the model performance. Step 203: Obtain the segmentation results of cell nuclei and cytoplasm in the test set, and visualize the mask as follows. Figure 3 As shown; In some embodiments: such as Figure 4 As shown, step 3 specifically includes: Step 301: Load the data and the segmentation results obtained in step 2, and read the cervical TCT cell image and its corresponding cell nucleus and cytoplasm mask; Step 302: Based on the OpenCV library, construct a cell morphology feature extraction and calculation module, including a shape feature calculation module and a texture feature calculation module, and integrate a batch processing module; Specifically, for the cell morphology feature extraction and measurement module, shape feature calculation module and texture feature calculation module are calculated separately, and some indicators are defined as follows: Circularity is a metric that measures how similar a particle's projection is to a circle. It can be expressed as the formula: (1), In formula (1), S 0 represents the area of the object. L Indicates the perimeter of an object; C When the value is 1, the shape is a circle; C The smaller the size, the more irregular the shape, and the greater the difference from a circle; Rectangularity is a parameter that reflects the degree to which an object fills its bounding rectangle, indicating how similar an object is to a rectangle; it is a rectangle fitting factor. Its formula is: (2), In formula (2), S 0 represents the area of the object, while S MER It is the area of its smallest bounding rectangle; R It reflects the degree to which an object fills its smallest bounding rectangle; for rectangular objects... R For a circular object, the maximum value is 1. R The value is 0, for slender, curved objects. R The value decreases; the value of the rectangle fitting factor is 0-1; The aspect ratio is the ratio of the major axis to the minor axis of the smallest bounding rectangle, and its calculation formula is as follows: (3), In formula (3), b It is the major axis of the smallest bounding rectangle. a It is the minor axis of its smallest bounding rectangle. For ease of comparison, the aspect ratio is uniformly defined to be greater than 1. The nucleus-to-cytoplasm ratio (N / C ratio) is a measurement used in cell biology, representing the ratio of the size of the cell nucleus to the size of the cytoplasm. Normally, the N / C ratio compares the volume of the nucleus to the cytoplasm. However, due to the unique, flattened morphology of cervical squamous cells, they are laid flat in the field of view in TCT slides. Therefore, in cervical cell TCT images, cells can be considered two-dimensional, and their N / C ratio can be approximated using the area ratio. The formula is as follows: (4), In formula (4),S nuc It is the area of the cell nucleus. S cyt This refers to the cytoplasmic area of the cell. In pathological analysis, the nucleocytic ratio is an important indicator because cancer cells often have a large nucleocytic ratio. Step 303: Input the data from Step 301 into the module of Step 302 in batches, output and record the morphological feature calculation results of the cell nucleus and cytoplasm in tabular form; apply the module to the test set. An example of the cell morphological feature calculation results is shown below. Figure 4 As shown; In some embodiments: such as Figure 5 As shown, step 4 specifically includes: Step 401: Based on the TBS standard and using the size of the middle layer cell nucleus as a reference baseline, calculate the multiple of the cell nucleus relative to the baseline in batches, and perform an overall analysis of the shape feature measurement results in Step 3. The baseline nuclear fold was obtained by comparing the nucleocytoplasmic ratio of the test cells with the baseline nuclear index; Table 1 shows an example of the analysis of the shape feature measurement results, and only a portion of the results are shown in this table. Table 1. Example of Shape Feature Calculation Results Analysis
[0015] Using the baseline nuclear fold as an indicator, a fold increase of 1-1.5 indicates an inflammatory response; a fold increase of 2.5-3 indicates a tendency toward low-grade squamous intraepithelial lesion (ASC-US); a fold increase of more than 3 indicates a preliminary low-grade squamous intraepithelial lesion (LSIL); as emphasized in bold in Table 1, the baseline nuclear fold increase is 3 or more and requires further analysis. Furthermore, the changes in nuclear roundness and nuclear length-to-width ratio are more pronounced in high-grade squamous intraepithelial lesions (HSIL) because not only is the nucleoplasm ratio significantly increased, but the nuclear membrane outline is also very irregular and often has obvious indentation or nuclear grooves. Therefore, if the baseline nuclear fold is greater than 5 times, and the nuclear roundness is significantly lower and the nuclear length-to-width ratio is significantly higher, it can be preliminarily analyzed as an HSIL lesion. For example, in Table 1, nuclear number 146 has significantly lower nuclear roundness and significantly higher nuclear length-to-width ratio, greatly increasing the likelihood of it being an HSIL lesion. Step 402: Visualize the texture feature calculation results from Step 3 by plotting GLCM histograms for the cell nucleus and cytoplasm, as shown below. Figure 5 As shown; Histograms can be used to analyze the overall data distribution; the X-axis of the histogram represents the value of a specific GLCM texture feature, and the Y-axis represents the frequency of occurrence of that feature value. Figure 5 The following is an example analysis.
[0016] The contrast histogram of the cytoplasm and nucleus shows overall low contrast, shifted to the left, indicating that most cells are normal. A few instances of high contrast indicate coarse texture, possibly indicating diseased cells. The uniformity histogram of the nucleus shifts to the right, indicating relatively uniform gray values, suggesting a more regular tissue structure, mostly normal cells. A few instances of low uniformity indicate uneven tissue structure, possibly indicating diseased cells. The energy of the nucleus is mostly high, shifted to the right, indicating a relatively consistent tissue structure, mostly normal cells. A few instances of low energy indicate disordered texture, possibly indicating diseased cells. The correlation histogram is similar.
[0017] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.
Claims
1. A method for extracting and predicting cellular morphological features from high-throughput cervical TCT images, characterized in that, The specific steps include: Step 1: Select cellular morphological features for cervical TCT image-assisted diagnosis; Step 2: Obtain the segmentation results of the cell nucleus and cytoplasm based on the segmentation network; Step 3: Extract and calculate cell morphological features based on the cervical TCT image segmentation results; Step 4: Analysis and visualization of cell morphology characteristic measurement results.
2. The method for extracting and predicting cellular morphological features from high-throughput cervical TCT images according to claim 1, characterized in that, Step 1 specifically includes: Step 101: Perform initial screening of cervical cell morphological characteristics and define core morphological characteristic indicators of squamous intraepithelial lesions. Step 102: Based on clinical pathological standards, cervical cell morphological characteristics are classified into two main categories: shape characteristics and texture characteristics.
3. The method for extracting and predicting cellular morphological features from high-throughput cervical TCT images according to claim 1, characterized in that, Step 2 specifically includes: Step 201: Load the dataset, randomly divide it into training, validation and test sets according to the proportion, and perform data augmentation on the training data; Step 202: Build and train a UNet-based cervical cell segmentation network. The basic architecture consists of two parts: an encoder and a decoder, forming a U-shaped symmetrical structure. Tune the parameters and optimize the model, and evaluate the model performance. Step 203: Obtain the segmentation results of cell nuclei and cytoplasm in the test set, and visualize the mask.
4. The method for extracting and predicting cellular morphological features from high-throughput cervical TCT images according to claim 1, characterized in that, Step 3 specifically includes: Step 301: Load the data and the segmentation results obtained in step 2, and read the cervical TCT cell image and its corresponding cell nucleus and cytoplasm mask; Step 302: Construct a cell morphology feature extraction and calculation module, including a shape feature calculation module and a texture feature calculation module, and integrate a batch processing module; Step 303: Input the data from step 301 into the module of step 302 in batches, and output and record the morphological characteristic measurement results of the cell nucleus and cytoplasm in tabular form.
5. The method for extracting and predicting cellular morphological features from high-throughput cervical TCT images according to claim 1, characterized in that, Step 4 specifically includes: Step 401: Based on the TBS standard and using the size of the middle layer cell nucleus as a reference baseline, calculate the multiple of the cell nucleus relative to the baseline in batches, and perform an overall analysis of the shape feature measurement results in Step 3. Step 402: Visualize the texture feature calculation results in Step 3, draw GLCM histograms of cell nuclei and cytoplasm respectively, and analyze the degree of lesion.