Algorithm based on cervical tissue squamous epithelium lesion diagnosis

Abstract

The invention particularly relates to an algorithm based on cervical tissue squamous epithelium lesion diagnosis. The algorithm comprises the following steps of tissue segmentation, region cutting and a classification algorithm. According to the method, multi-step fine processing is carried out on a cervical tissue slice image, for example, in the tissue segmentation stage, gray transformation and Gaussian filtering are adopted to enhance image features, and a cross entropy loss function is used to optimize model parameters; flexible segmentation is carried out according to form contour information during region cutting; in the classification algorithm, image preprocessing and data enhancement are carried out, and a multi-layer neural network is utilized to extract and analyze features, so that squamous epithelium lesion tissues can be identified more accurately, the lesion degree is judged comprehensively, and the diagnosis accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and particularly to an algorithm for diagnosing squamous epithelial lesions of cervical tissue. Background Art

[0002] Cervical cancer is one of the most common malignant tumors in women. Early screening and diagnosis are crucial for reducing the incidence and mortality rates. Traditional histopathological examinations for diagnosing cervical squamous epithelial lesions rely on manual judgment under a microscope, which has problems such as strong subjectivity and low diagnostic efficiency and cannot meet the needs of large-scale screening.

[0003] In recent years, with the progress of computer technology and artificial intelligence, the use of image processing and machine learning technologies for automated histopathological diagnosis has become a research hotspot. Existing studies have attempted to use tissue segmentation, cell classification, and deep learning algorithms, but still face challenges such as low diagnostic accuracy and efficiency for squamous epithelial lesions.

[0004] Therefore, an algorithm for diagnosing squamous epithelial lesions of cervical tissue is proposed to address the above-mentioned problems. Summary of the Invention

[0005] The object of the present invention is to provide an algorithm for diagnosing squamous epithelial lesions of cervical tissue to solve the above problems.

[0006] To achieve the above object, the present invention adopts the following technical solutions: An algorithm for diagnosing squamous epithelial lesions of cervical tissue includes the following parts: Tissue segmentation: Collect cervical tissue section images to form a training data set, and input the training data set into a model for training. After the model training is completed, input the cervical tissue section image to be diagnosed into the trained model; Region cutting: After determining the cutting method and parameters, cut the segmented squamous epithelial tissue region, and record the cut region as a small region. After numbering it, store it in a specified database; Classification algorithm: Extract the data of the small region image from the database, process the image, and input the processed data into a model for training. After the model training is completed, input the small region image to be diagnosed into the trained model; the model outputs the lesion classification result, and based on the classification result, comprehensively judge the lesion degree of the cervical tissue.

[0007] Preferably, in the tissue segmentation, it specifically includes the following parts: Collect cervical tissue section images of a preset number of patients, which should include normal tissues and squamous epithelial lesion tissues of different degrees; annotate the collected images to mark the squamous epithelial tissue regions, forming a training data set; at the same time, reserve a preset number of images as the validation set and the test set for evaluating the performance of the model; Input the prepared training data set into the model for training. After the model training is completed and reaches the performance index, input the cervical tissue section image to be diagnosed into the trained model. The model outputs the image of the segmented squamous epithelial tissue region, and post-process the segmentation result.

[0008] Preferably, when the model is training, use the cross-entropy loss function to optimize the model parameters; Calculate the gradient of the loss function with respect to the model parameters through backpropagation, and continuously adjust the parameters to make the value of the loss function reach the preset range, so as to improve the segmentation accuracy of the model for the squamous epithelial tissue region; Before segmentation, perform gray-scale transformation on the input image, enhance the contrast of the tissues in the image to highlight the edges and texture information of the squamous epithelial tissue. At the same time, use Gaussian filtering to denoise the image to remove the noise points in the image.

[0009] Preferably, in the region cutting, it specifically includes the following parts: Flexibly segment according to the morphology and contour information of the squamous epithelial tissue region. First, it is necessary to obtain the contour of the tissue region and use an edge detection algorithm; this algorithm includes: Gaussian filtering for noise reduction, using a Gaussian filter to smooth the image to reduce the interference of noise in the image on edge detection; Calculate the gradient magnitude and direction, and use the operator to calculate the gradient of the image in and directions; Non-maximum suppression. In the calculated gradient magnitude image, check each pixel point. Only when the gradient magnitude of the pixel point is the local maximum in its gradient direction, keep the point, otherwise set its magnitude to 0; Double-threshold detection and edge connection. Preset the lowest threshold and the highest threshold. Mark the pixel points with gradient magnitude greater than the highest threshold as strong edge points, mark the pixel points with gradient magnitude between the lowest threshold and the highest threshold as weak edge points, and mark the pixel points with gradient magnitude less than the lowest threshold as non-edge points; After obtaining the contour information, according to the geometric characteristics of the contour, divide the tissue region into small block regions with relatively regular shapes and similar sizes according to certain rules.

[0010] Preferably, the calculation of the gradient magnitude and direction specifically includes the following parts: Calculate operators in and directions of the template and mark them as and ; Convolve the image with and respectively to obtain and , and then calculate the gradient magnitude and direction according to the formula.

[0011] Preferably, the classification algorithm specifically includes the following parts: Read data from the stored small region images, perform preprocessing operations on the images, and at the same time perform random rotation, flipping, scaling, and color adjustment operations on the training data; Input the preprocessed data into the model for training; during the training process, adjust the parameters of the model continuously according to the set cross-entropy loss function and optimizer; After the model training is completed, input the small region images to be diagnosed into the trained model. The model outputs the lesion classification results, and according to the classification results, count the number and proportion of small regions with different lesion degrees in each cervical tissue section, so as to comprehensively judge the lesion degree of the cervical tissue.

[0012] Preferably, during the training by the model, perform data augmentation operations on the input images; Use the cross-entropy loss function to measure the difference between the model prediction results and the true labels, and select optimizer; Input the cut and preprocessed small region images into the trained model. The model extracts and analyzes the image features through convolutional layers, pooling layers, and fully connected layers, and finally outputs the probability values of each small region belonging to CIN1, CIN2, or CIN3, and determines the final lesion classification result of the small region according to the magnitude of the probability values.

[0013] Preferably, after obtaining the comprehensive judgment of the lesion degree of the cervical tissue according to the classification results, present the diagnosis results in an intuitive form through a visualization interface.

[0014] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are: 1. The present invention performs multi-step fine processing on cervical tissue section images. For example, during the tissue segmentation stage, gray-scale transformation and Gaussian filtering are used to enhance image features, and the cross-entropy loss function is used to optimize model parameters. During region cutting, flexible segmentation is performed based on morphological contour information. In the classification algorithm, image preprocessing, data augmentation are carried out, and multi-layer neural networks are used to extract and analyze features, so as to more accurately identify squamous epithelial lesion tissues, comprehensively judge the degree of lesions, and thus improve the diagnostic accuracy.

[0015] 2. The present invention uses a visualization interface to label different lesion degree regions with different colors or patterns following medical image annotation specifications, and at the same time intuitively presents the proportion of lesion degrees with bar charts or pie charts, enabling doctors to quickly and accurately understand the cervical tissue lesion situation, providing strong support for clinical diagnostic decisions, and improving the diagnostic efficiency and decision-making scientificity. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In the following description of exemplary embodiments in conjunction with the drawings, more details, features and advantages of the present application are disclosed. In the drawings: Figure 1 is a flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The following will describe several embodiments of the present application in more detail with reference to the drawings so that those skilled in the art can implement the present application. The present application can be embodied in many different forms and purposes and should not be limited to the embodiments described herein. These embodiments are provided to make the present application comprehensive and complete, and fully convey the scope of the present application to those skilled in the art. The embodiments do not limit the present application.

[0018] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the relevant art and / or the context of this specification, and will not be interpreted in an idealized or overly formal sense unless clearly defined herein.

[0019] Please refer to Figure 1 as shown, the present invention provides a technical solution: An algorithm for diagnosing squamous epithelial lesions of cervical tissue, comprising the following parts: Tissue segmentation: Collect cervical tissue section images to form a training data set, and input the training data set into the model for training. After the model training is completed, input the cervical tissue section image to be diagnosed into the trained model; In tissue segmentation, it specifically includes the following parts: Collect cervical tissue section images of a preset number of patients, and the images should contain normal tissues and squamous epithelial lesion tissues of different degrees; annotate the collected images to mark the squamous epithelial tissue areas, forming a training data set; at the same time, reserve a preset number of images as the validation set and the test set for evaluating the performance of the model; Input the prepared training data set into the model for training. After the model training is completed and reaches the performance index, input the cervical tissue section image to be diagnosed into the trained model. The model outputs the image of the segmented squamous epithelial tissue area, and post-process the segmentation result; such as removing noise areas with too small area and filling holes in the segmented area to obtain the final accurate segmentation result of the squamous epithelial tissue; When the model is training, use the cross-entropy loss function to optimize the model parameters. The formula is where is the number of samples, is the number of classes. In the diagnosis of squamous epithelial lesions of cervical tissues, can be regarded as two categories: squamous epithelial tissue and other tissues, is the true label that the sample belongs to the class , usually in the unique coding form of 0 or 1; is the probability that the model predicts the sample belongs to the class ; Calculate the gradient of the loss function with respect to the model parameters through backpropagation, and continuously adjust the parameters to make the value of the loss function reach the preset range, so as to improve the segmentation accuracy of the model for the squamous epithelial tissue area; Before segmentation, perform gray-scale transformation on the input image, and enhance the contrast of the tissues in the image to highlight the edges and texture information of the squamous epithelial tissue. At the same time, use Gaussian filtering to perform noise reduction processing on the image to remove the noise points in the image; Specifically include: mark the input image as , and the gray-scale transformation formula can adopt linear transformation , where is the image pixel coordinate, a and b are constants, and by adjusting the values of a and b, enhance the contrast of the tissues in the image to highlight the edges and texture information of the squamous epithelial tissue for easy model recognition; The calculation formula of Gaussian filtering is , where is the mean value, usually taken as 0, is the standard deviation; Region cutting: After determining the cutting method and parameters, cut the segmented squamous epithelial tissue region, and record the cut region as a small region. After numbering it, store it in the specified database; During region cutting, it specifically includes the following parts: Flexibly segment according to the morphology and contour information of the squamous epithelial tissue region. First, the contour of the tissue region needs to be obtained, and an edge detection algorithm is used; this algorithm includes: Gaussian filtering for noise reduction, using a Gaussian filter to smooth the image to reduce the interference of noise in the image on edge detection; Calculate the gradient magnitude and direction, using operator in and directions to calculate the gradient of the image; Non-maximum suppression: In the calculated gradient magnitude image, check each pixel point. Only when the gradient magnitude of the pixel point is a local maximum in its gradient direction, keep the point, otherwise set its magnitude to 0; Double-threshold detection and edge connection: Preset the lowest threshold and the highest threshold. Mark the pixel points with gradient magnitude greater than the highest threshold as strong edge points, mark the pixel points with gradient magnitude between the lowest threshold and the highest threshold as weak edge points, and mark the pixel points with gradient magnitude less than the lowest threshold as non-edge points; After obtaining the contour information, according to the geometric features of the contour, such as the center, area, perimeter, etc. of the contour, divide the tissue region into small regions with relatively regular shapes and similar sizes according to certain rules; For example, it can be divided by using a grid division method based on the geometric center of the contour, or by performing adaptive polygon division according to the convex hull information of the contour; Calculate the gradient magnitude and direction, which specifically includes the following parts: Calculate operator in and directions and mark as and ; , ; Convolve the image with and respectively to obtain and , and then calculate the gradient magnitude and direction according to the formula; The calculation formula is: , to obtain the gradient magnitude; , to obtain the direction; Classification algorithm: Extract data of small area images from the database, process the images, and input the processed data into the model for training. After the model training is completed, input the small area image to be diagnosed into the trained model; the model outputs the lesion classification result, and based on the classification result, comprehensively judge the lesion degree of the cervical tissue.

[0020] In the classification algorithm, it specifically includes the following parts: Read data from the stored small area images, perform preprocessing operations on the images, such as normalization processing, map the pixel values of the images to the range of [0, 1] to accelerate the training speed of the model and improve the stability of the model; at the same time, perform random rotation, flipping, scaling and color adjustment operations on the training data; Input the preprocessed data into the model for training; during the training process, according to the set cross-entropy loss function and optimizer, continuously adjust the parameters of the model; When the model training is completed, input the small area image to be diagnosed into the trained model. The model outputs the lesion classification result, and based on the classification result, count the number and proportion of small area images with different lesion degrees in each cervical tissue section, so as to comprehensively judge the lesion degree of the cervical tissue; Classify each small area to obtain the number of CIN1, CIN2, and CIN3, and divide the number of CIN1, CIN2, and CIN3 by the total number of cuts of the cervical tissue section respectively, so as to obtain the proportions of CIN1, CIN2, and CIN3 respectively; Preset the proportion thresholds of CIN1, CIN2, and CIN3, and compare the proportions of CIN1, CIN2, and CIN3 with the proportion thresholds in turn. If the proportion of CIN1 is greater than the proportion threshold, it is determined that there is a CIN1 lesion. If the proportion of CIN2 is greater than the proportion threshold, it is determined that there is a CIN2 lesion. If the proportion of CIN3 is greater than the proportion threshold, it is determined that there is a CIN3 lesion.

[0021] Through model training, during the training process, perform data augmentation operations on the input images; Such as random rotation, the rotation operation can be achieved through affine transformation, and the affine transformation matrix is: , horizontal and vertical flipping operations, horizontal flipping is to transform the coordinates of the image pixels, ; where is the image width, is the original pixel coordinate, are the flipped pixels coordinates; Vertical flipping transforms the coordinates of the image pixels, ; where is the image height, are the coordinates of the original pixel , are the flipped pixels coordinates; The scaling operation is also achieved through affine transformation. The scaling matrix is: ; where and are and scaling factors in the and directions; Color adjustment includes adjusting parameters such as brightness, contrast, saturation, etc.; When the optimizer updates the model parameters, the following formulas are used to calculate the first-order moment estimate and the second-order moment estimate : ; ; where is the gradient at the current moment, and are the decay coefficients, usually , , and are the first-order moment estimate and the second-order moment estimate at the previous moment; The model parameters are updated through the following formula : ; ; ; where is the learning rate, is a preset constant, is the current training step; Input the cut and preprocessed small regional images into the trained In the model, the model extracts and analyzes image features through convolutional layers, pooling layers, and fully connected layers, and finally outputs the probability values that each small block area belongs to CIN1, CIN2, or CIN3. The final lesion classification result of the small block area is determined according to the magnitude of the probability values.

[0022] After obtaining the lesion degree of the cervical tissue through comprehensive judgment based on the classification results, the diagnosis results are presented in an intuitive form through a visualization interface; the visualization interface has the following characteristics: Precisely label the areas with different lesion degrees in the cervical tissue section image with different colors or patterns. The labeling method follows the medical image labeling standard to ensure that doctors can clearly identify the boundaries and scopes of normal tissues and areas with different lesion degrees such as CIN1, CIN2, and CIN3; Use a bar chart or a pie chart to present the proportions of small block areas with different lesion degrees in the entire cervical tissue section; for the bar chart, the lesion type is the horizontal axis and the proportion value is the vertical axis, and the height of the bars intuitively reflects the proportion differences of each lesion degree; the pie chart uses different fan-shaped areas to represent different lesion degrees, and each fan-shaped area is accompanied by a percentage label to clearly show the proportion of each type of lesion in the whole. Through the above visualization methods, it is convenient for doctors to quickly and accurately understand the lesion situation of the cervical tissue and provide strong support for clinical diagnosis and decision-making.

[0023] The above formulas are all obtained through software simulation by collecting a large amount of data and selecting a formula close to the true value. The influence weight factors and specific coefficient values in the formula are set by those skilled in the art according to the actual situation and can be adjusted and modified subsequently.

[0024] The above description of the embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An algorithm for diagnosing squamous epithelial lesions of cervical tissue, characterized in that, It includes the following parts: Tissue segmentation: Collect cervical tissue section images to form a training dataset, and input the training dataset into the model for training. After the model training is completed, input the cervical tissue section image to be diagnosed into the trained model; Region cutting: After determining the cutting method and parameters, cut the segmented squamous epithelial tissue region, and record the cut region as a small block region. After numbering it, store it in the specified database; Classification algorithm: Extract data of small regional images from the database, process the images, and input the processed data into the model for training. After the model training is completed, input the small regional images to be diagnosed into the trained model; The model outputs the lesion classification result, and based on the classification result, comprehensively judge the lesion degree of the cervical tissue.

2. The algorithm for diagnosing squamous epithelial lesions of cervical tissue according to claim 1, wherein In the tissue segmentation, it specifically includes the following parts: Collect cervical tissue section images of a preset number of patients. The images should include normal tissues and squamous epithelial lesion tissues of different degrees; annotate the collected images to mark the squamous epithelial tissue regions to form a training data set; at the same time, reserve a preset number of images as the validation set and the test set for evaluating the performance of the model; Input the prepared training data set into the model for training. After the model training is completed and the performance metrics are met, input the cervical tissue section images to be diagnosed into the trained model. The model outputs the image of the segmented squamous epithelial tissue region, and post-process the segmentation result.

3. The algorithm for diagnosing squamous epithelial lesions of cervical tissue according to claim 2, wherein The model uses the cross-entropy loss function to optimize the model parameters during training; calculates the gradient of the loss function with respect to the model parameters through backpropagation, and continuously adjusts the parameters to make the value of the loss function reach the preset range, thereby improving the segmentation accuracy of the model for the squamous epithelial tissue region; Before segmentation, perform gray-scale transformation on the input image, enhance the contrast of the tissues in the image to highlight the edges and texture information of the squamous epithelial tissue. At the same time, use Gaussian filtering to perform noise reduction processing on the image to remove the noise points in the image.

4. The algorithm for diagnosing squamous epithelial lesions of cervical tissue according to claim 1, wherein In the region cutting, it specifically includes the following parts: Flexibly segment according to the morphological and contour information of the squamous epithelial tissue region. First, obtain the contour of the tissue region and use an edge detection algorithm; this algorithm includes: Gaussian filtering for noise reduction, using a Gaussian filter to smooth the image to reduce the interference of the noise in the image on edge detection; Calculate the gradient magnitude and direction, using operator to calculate the gradient of the image in the and directions; Non-maximum suppression, in the calculated gradient magnitude image, check each pixel point. Only when the gradient magnitude of this pixel point is a local maximum in its gradient direction, keep this point, otherwise set its magnitude to 0; Double-threshold detection and edge connection, preset the lowest threshold and the highest threshold, mark the pixel points with gradient magnitude greater than the highest threshold as strong edge points, mark the pixel points with gradient magnitude between the lowest threshold and the highest threshold as weak edge points, and mark the pixel points with gradient magnitude less than the lowest threshold as non-edge points; After obtaining the contour information, according to the geometric characteristics of the contour, divide the tissue region into small block regions with relatively regular shapes and similar sizes according to certain rules.

5. An algorithm for diagnosing squamous epithelial lesions of cervical tissue according to claim 4, characterized in that, The calculation of the gradient magnitude and direction specifically includes the following parts: Calculation The operator and templates in the and directions and mark them as Convolve the image with and respectively to obtain and , and then calculate the gradient magnitude and direction according to the formula.

6. The algorithm for diagnosing squamous epithelial lesions of cervical tissue according to claim 1, wherein, In the classification algorithm, it specifically includes the following parts: Read data from the stored small block region images, perform preprocessing operations on the images, and at the same time perform random rotation, flipping, scaling, and color adjustment operations on the training data; Input the preprocessed data into the model for training; During the training process, according to the set cross-entropy loss function and optimizer, continuously adjust the parameters of the model; When the model training is completed, input the small block region image to be diagnosed into the trained model. The model outputs the lesion classification result. According to the classification result, count the number and proportion of small block regions with different lesion degrees in each cervical tissue section, so as to comprehensively judge the lesion degree of the cervical tissue.

7. An algorithm for diagnosing squamous epithelial lesions of cervical tissue according to claim 6, characterized in that, The training is performed through a model, and during the training process, data augmentation operations are performed on the input images; The cross-entropy loss function is used to measure the difference between the model's prediction results and the true labels, and optimizer is selected; Input the cut and preprocessed small area images into the trained model. The model extracts and analyzes the image features through convolutional layers, pooling layers and fully connected layers, and finally outputs the probability values that each small area belongs to CIN1, CIN2 or CIN3. Determine the final lesion classification result of the small area according to the size of the probability value.

8. An algorithm for diagnosing squamous epithelial lesions of cervical tissue according to claim 7, characterized in that, After obtaining the comprehensive judgment of the lesion degree of the cervical tissue based on the classification result, present the diagnosis result in an intuitive form through a visualization interface.

Images