Artificial Intelligence-Assisted Pathological Diagnosis System for Pancreatic Cancer Based on Deep Learning

Abstract

The present invention provides an artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning, which relates to the field of medical diagnosis technology. By acquiring the original pancreatic cancer pathological images and performing preprocessing operations, corresponding target pancreatic cancer pathological images are generated; the target pancreatic cancer pathological images are classified to generate training pancreatic cancer pathological images and validation pancreatic cancer pathological images; a deep learning algorithm is used to construct and train a pancreatic cancer pathological diagnosis model based on the training pancreatic cancer pathological images, and the model parameters of the pancreatic cancer pathological diagnosis model are adjusted in real time; the diagnostic performance of the pancreatic cancer pathological diagnosis model is evaluated based on the validation pancreatic cancer pathological images, and the pancreatic cancer pathological diagnosis model is optimized according to the evaluation results; the current pancreatic cancer pathological image is acquired, preprocessed and then input into the pancreatic cancer pathological diagnosis model, and the current diagnostic result is output and displayed to the physician side, so that the pathological images can be automatically analyzed, the accurate diagnosis of pancreatic cancer can be realized, and the diagnostic efficiency and accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical diagnosis, and particularly to an artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning. Background Art

[0002] Pancreatic cancer is a highly malignant digestive system tumor, which has characteristics such as atypical early symptoms, strong invasiveness, and easy metastasis and recurrence after surgery. At present, the diagnosis of pancreatic cancer mainly relies on imaging examinations and pathological diagnoses, but the diagnostic accuracy of these methods is extremely susceptible to various factors, such as the size, location, and morphology of the tumor. In recent years, the application of artificial intelligence in the medical field has developed rapidly, especially machine learning and deep learning algorithms have been widely used in aspects such as image recognition, risk prediction, and diagnostic decision-making of pancreatic cancer.

[0003] Traditional pathological diagnosis methods for pancreatic cancer are extremely susceptible to factors such as image quality and doctor experience, which may lead to limited diagnostic accuracy. For example, the specificity of endoscopic ultrasound in differentiating malignant tumors from benign masses is not ideal, with a range of only 50% - 60%. And existing diagnostic systems often lack multi-center databases and have a small data sample size. At the same time, they cannot automatically complete the analysis and diagnosis of pathological images, requiring a long diagnostic time, which may lead to insufficient model training, affecting the efficiency and accuracy of pancreatic cancer diagnosis and unable to provide auxiliary decision-making support for doctors.

[0004] Therefore, it is necessary to provide an artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning to solve the above technical problems. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides an artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning to solve the problems that traditional pathological diagnosis methods for pancreatic cancer are extremely susceptible to factors such as image quality and doctor experience, resulting in limited diagnostic accuracy, and existing diagnostic systems lack multi-center databases and have a small data sample size. At the same time, they cannot automatically complete the analysis and diagnosis of pathological images, requiring a long diagnostic time, leading to insufficient model training, affecting the efficiency and accuracy of pancreatic cancer diagnosis, and unable to provide auxiliary decision-making support for doctors.

[0006] The artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning provided by the present invention includes:

[0007] An acquisition module, configured to obtain an original pancreatic cancer pathological image and perform preprocessing operations on the original pancreatic cancer pathological image to generate a corresponding target pancreatic cancer pathological image;

[0008] A classification module for classifying the target pancreatic cancer pathological image to generate a training pancreatic cancer pathological image and a validation pancreatic cancer pathological image;

[0009] A construction module for constructing and training a pancreatic cancer pathological diagnosis model based on the training pancreatic cancer pathological image by using a deep learning algorithm, and adjusting the model parameters of the pancreatic cancer pathological diagnosis model in real time;

[0010] An evaluation module for evaluating the diagnostic performance of the pancreatic cancer pathological diagnosis model based on the validation pancreatic cancer pathological image, and optimizing the pancreatic cancer pathological diagnosis model according to the evaluation results;

[0011] A result module for obtaining the current pancreatic cancer pathological image, preprocessing it and inputting it into the pancreatic cancer pathological diagnosis model, outputting the current diagnosis result and displaying it to the physician side through a visualization interface.

[0012] Preferably, the original pancreatic cancer pathological image includes images of pancreatic cancer pathological sections of benign and malignant, as well as images of pancreatic cancer pathological sections in the early, middle and late stages.

[0013] Preferably, the preprocessing operations include:

[0014] Cleaning the obtained original pancreatic cancer pathological image, that is, removing the blurred and duplicate original pancreatic cancer pathological images to generate a corresponding first pancreatic cancer pathological image;

[0015] Removing the noise in the first pancreatic cancer pathological image based on the Gaussian filtering algorithm to generate a corresponding second pancreatic cancer pathological image;

[0016] Enhancing the contrast of the second pancreatic cancer pathological image by using the histogram equalization method to generate a corresponding third pancreatic cancer pathological image;

[0017] Performing normalization processing on the third pancreatic cancer pathological image to generate the corresponding target pancreatic cancer pathological image.

[0018] Preferably, the construction module for constructing and training a pancreatic cancer pathological diagnosis model based on the training pancreatic cancer pathological image by using a deep learning algorithm and adjusting the model parameters of the pancreatic cancer pathological diagnosis model in real time specifically includes:

[0019] Using a deep learning algorithm to input the training pancreatic cancer pathological image into the input layer of the pancreatic cancer pathological diagnosis model, and the size of the training pancreatic cancer pathological image is , where represents the height of the training pancreatic cancer pathological image; represents the width of the training pancreatic cancer pathological image; Indicates the number of channels for training pancreatic cancer pathological images;

[0020] The pancreatic cancer pathological diagnosis model includes multiple convolutional layers. Each convolutional layer includes a preset number of convolutional kernels. By adjusting the size, number, and stride of the convolutional kernels, introducing corresponding bias terms and activation functions, local features of different scales and complexities in the training pancreatic cancer pathological images are extracted, and the corresponding first pancreatic cancer pathological feature is obtained by summing all the local features. The mathematical formula of the first pancreatic cancer pathological feature is:

[0021]

[0022] In the formula, represents the pixel value of the first pancreatic cancer pathological feature at position ; represents the pixel value of the training pancreatic cancer pathological image at position ; represents the weight at position in the convolutional kernel; represents the activation function of the convolutional layer; represents the bias term of the convolutional layer.

[0023] Preferably, the dimension of the first pancreatic cancer pathological feature is reduced through the pooling layer of the pancreatic cancer pathological diagnosis model, and the important feature information in the first pancreatic cancer pathological feature is retained, that is, the second pancreatic cancer pathological feature. The mathematical formula of the second pancreatic cancer pathological feature is:

[0024]

[0025] In the formula, represents the pixel value of the second pancreatic cancer pathological feature at position ; represents the pixel value of the first pancreatic cancer pathological feature at position ; represents the pooling window; represents taking the maximum value among all positions within the pooling window;

[0026] The first pancreatic cancer pathological feature and the second pancreatic cancer pathological feature are flattened to obtain the corresponding input pancreatic cancer pathological feature .

[0027] Preferably, the input pancreatic cancer pathological feature is subjected to feature fusion and dimensionality reduction processing through the fully connected layer of the pancreatic cancer pathological diagnosis model to generate the corresponding output feature. The mathematical formula of the output feature is:

[0028]

[0029] In the formula, represents the output feature of the fully connected layer; represents the transpose of the weight of the fully connected layer; represents the input pancreatic cancer pathological features; represents the bias term of the fully connected layer;

[0030] Based on the output layer of the pancreatic cancer pathological diagnosis model, a softmax classifier is used to map the output feature to the probability distribution of the pancreatic cancer diagnosis result, and the corresponding pancreatic cancer diagnosis category and diagnosis probability are obtained:

[0031]

[0032] In the formula, represents the pancreatic cancer diagnosis category of the diagnosis probability; represents the pancreatic cancer diagnosis category corresponding output feature; represents the exponential sum of the output features corresponding to all pancreatic cancer diagnosis categories.

[0033] Preferably, the loss function corresponding to the pancreatic cancer pathological diagnosis model is determined to measure the difference between the pancreatic cancer diagnosis result and the actual pathological result. The mathematical formula of the loss function is:

[0034]

[0035] In the formula, represents the loss function; represents the number of input training pancreatic cancer pathological images; represents the loss value corresponding to a single training pancreatic cancer pathological image; represents the actual pathological result corresponding to the training pancreatic cancer pathological image; represents the pancreatic cancer diagnosis result corresponding to the training pancreatic cancer pathological image.

[0036] Preferably, based on the backpropagation algorithm, the gradient of the loss function with respect to the model parameters of the pancreatic cancer pathological diagnosis model is calculated:

[0037]

[0038] In the formula, represents the loss function; represents the weight of the fully connected layer; represents the pancreatic cancer diagnosis result corresponding to the training pancreatic cancer pathological image; represents the input pancreatic cancer pathological features; represents the loss function with respect to the weight of the fully connected layer Gradient of Denote the loss function For the diagnosis result of pancreatic cancer Gradient of Denote the input pathological features of pancreatic cancer For the weights of the fully connected layer Gradient of

[0039] Preferably, update the model parameters of the pancreatic cancer pathological diagnosis model based on the gradient descent method to minimize the loss function:

[0040]

[0041] In the formula, Denote the Weights of the fully connected layer after the Denote the Weights of the fully connected layer after the Denote the learning rate, which is used to control the update step of the model parameters of the pancreatic cancer pathological diagnosis model; Denote the loss function; Denote the weights of the fully connected layer; Denote the loss function For the weights of the fully connected layer At the Gradient at the

[0042] Preferably, the evaluation module is used to evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model based on the verified pancreatic cancer pathological images, and optimize the pancreatic cancer pathological diagnosis model according to the evaluation results, specifically including:

[0043] Input the verified pancreatic cancer pathological images into the pancreatic cancer pathological diagnosis model to determine the confusion matrix corresponding to the verification result. The elements of the confusion matrix include true positives, false positives, true negatives, and false negatives. The true positives are the correct diagnosis of pancreatic cancer pathological results; the false positives are the incorrect diagnosis of pancreatic cancer pathological results; the true negatives are the correct diagnosis of non-pancreatic cancer pathological results; the false negatives are the incorrect diagnosis of non-pancreatic cancer pathological results;

[0044] Based on the confusion matrix, calculate the diagnostic indicators corresponding to the pancreatic cancer pathological diagnosis model, where the diagnostic indicators include accuracy, precision, recall, F1 score, and false positive rate;

[0045] The formula for calculating the accuracy is:

[0046]

[0047] In the formula, Represents the accuracy rate corresponding to the pathological diagnosis model of pancreatic cancer; Represents a true positive, that is, correctly diagnosing the pathological result of pancreatic cancer; Represents a true negative, that is, correctly diagnosing the pathological result of non-pancreatic cancer; Represents a false positive, that is, incorrectly diagnosing the pathological result of pancreatic cancer; Represents a false negative, that is, incorrectly diagnosing the pathological result of non-pancreatic cancer;

[0048] The formula for calculating the precision rate is:

[0049]

[0050] In the formula, Represents the precision rate corresponding to the pathological diagnosis model of pancreatic cancer; Represents a true positive, that is, correctly diagnosing the pathological result of pancreatic cancer; Represents a false positive, that is, incorrectly diagnosing the pathological result of pancreatic cancer;

[0051] The formula for calculating the recall rate is:

[0052]

[0053] In the formula, Represents the recall rate corresponding to the pathological diagnosis model of pancreatic cancer; Represents a true positive, that is, correctly diagnosing the pathological result of pancreatic cancer; Represents a false negative, that is, incorrectly diagnosing the pathological result of non-pancreatic cancer;

[0054] The formula for calculating the F1 score is:

[0055]

[0056] In the formula, Represents the F1 score corresponding to the pathological diagnosis model of pancreatic cancer; Represents the precision rate corresponding to the pathological diagnosis model of pancreatic cancer; Represents the recall rate corresponding to the pathological diagnosis model of pancreatic cancer;

[0057] The formula for calculating the false positive rate is:

[0058]

[0059] In the formula, Represents the false positive rate corresponding to the pathological diagnosis model of pancreatic cancer; Represents a false positive, that is, incorrectly diagnosing the pathological result of pancreatic cancer; Represents a true negative, that is, correctly diagnosing the pathological result of non-pancreatic cancer;

[0060] Analyze the diagnostic indicators, weigh the precision and recall rate, evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model, and optimize the pancreatic cancer pathological diagnosis model according to the evaluation results.

[0061] Compared with the related technologies, the artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning provided by the present invention has the following beneficial effects:

[0062] Through the acquisition module, classification module, construction module, evaluation module and result module, the present invention can obtain the original pancreatic cancer pathological images, perform preprocessing operations on the original pancreatic cancer pathological images to generate corresponding target pancreatic cancer pathological images; classify the target pancreatic cancer pathological images to generate training pancreatic cancer pathological images and verification pancreatic cancer pathological images; use deep learning algorithms to construct and train a pancreatic cancer pathological diagnosis model based on the training pancreatic cancer pathological images, and adjust the model parameters of the pancreatic cancer pathological diagnosis model in real time; evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model based on the verification pancreatic cancer pathological images, and optimize the pancreatic cancer pathological diagnosis model according to the evaluation results; obtain the current pancreatic cancer pathological image, input it into the pancreatic cancer pathological diagnosis model after preprocessing, output the current diagnosis result and display it to the physician side through a visualization interface. The present invention can automatically extract and analyze the feature information in the pathological images through deep learning algorithms, realize the accurate diagnosis of pancreatic cancer, and improve the diagnostic accuracy; the system of the present invention can automatically complete the analysis and diagnosis of pathological images, greatly shorten the diagnosis time, and improve the diagnostic efficiency; at the same time, it can provide auxiliary decision-making support for doctors, help doctors more accurately judge the condition, formulate reasonable treatment plans, and thus can promote the development of medical intelligence and improve the level and quality of medical services. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is the system block diagram of the artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning of the present invention;

[0064] Figure 2 is the schematic diagram of preprocessing the original pancreatic cancer pathological image of the present invention;

[0065] Figure 3 is the schematic diagram of displaying the current diagnosis result to the physician side of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] The present invention will be further described below with reference to the drawings and embodiments. Embodiment

[0067] As Figure 1 shown, an artificial intelligence-assisted pathological diagnosis system for pancreatic cancer based on deep learning, the system includes:

[0068] The acquisition module is used to obtain the original pancreatic cancer pathological image and perform preprocessing operations on the original pancreatic cancer pathological image to generate the corresponding target pancreatic cancer pathological image;

[0069] The classification module is used to classify the target pancreatic cancer pathological image to generate training pancreatic cancer pathological images and validation pancreatic cancer pathological images;

[0070] The construction module is used to construct and train a pancreatic cancer pathological diagnosis model based on the training pancreatic cancer pathological images by using deep learning algorithms, and adjust the model parameters of the pancreatic cancer pathological diagnosis model in real time;

[0071] The evaluation module is used to evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model based on the validation pancreatic cancer pathological images, and optimize the pancreatic cancer pathological diagnosis model according to the evaluation results;

[0072] The result module is used to obtain the current pancreatic cancer pathological image, input it to the pancreatic cancer pathological diagnosis model after preprocessing, output the current diagnosis result and display it to the physician side through a visualization interface.

[0073] Among them, the acquisition module uses high-precision image acquisition equipment, such as a high-resolution microscope combined with a digital imaging system, which can ensure the high definition and detail integrity of the acquired images. Subsequently, preprocessing operations can be performed on the acquired original pancreatic cancer pathological images to improve the image quality and reduce interference factors, and finally generate the corresponding target pancreatic cancer pathological images.

[0074] The classification module can further divide the preprocessed target pancreatic cancer pathological images into training pancreatic cancer pathological images and validation pancreatic cancer pathological images. The classification process is based on factors such as the pathological features, stages, and grades of the images, so as to ensure that the training pancreatic cancer pathological images and validation pancreatic cancer pathological images are representative in terms of pathological types, complexity, etc. Specifically, the subsequent training pancreatic cancer pathological images are used to construct the pancreatic cancer pathological diagnosis model, while the validation pancreatic cancer pathological images are used to evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model. The two are independent of each other, which helps to improve the generalization ability of the pancreatic cancer pathological diagnosis model.

[0075] The construction module uses deep learning algorithms to construct a pancreatic cancer pathological diagnosis model based on the training pancreatic cancer pathological images. Through iterative training, the pancreatic cancer pathological diagnosis model can gradually learn and identify pathological features in the images, such as cell morphology, mitotic figures, etc. At the same time, the construction module can adjust the model parameters in real time, such as the learning rate, weight decay, etc., to optimize the performance of the pancreatic cancer pathological diagnosis model and ensure its high precision in complex and changeable pathological images.

[0076] The evaluation module can use the verified pancreatic cancer pathological images to evaluate the diagnostic performance of the trained pancreatic cancer pathological diagnosis model. The evaluation metrics include accuracy, recall, F1-score, etc., so as to comprehensively measure the ability of the pancreatic cancer pathological diagnosis model in identifying pancreatic cancer pathological features. According to the evaluation results, the evaluation module can perform corresponding model optimizations, such as adjusting the network structure, increasing data augmentation strategies, etc., so as to further improve the diagnostic accuracy and robustness of the pancreatic cancer pathological diagnosis model.

[0077] The result module can obtain the current pancreatic cancer pathological image to be diagnosed. After passing through the same preprocessing process, it is input into the optimized pancreatic cancer pathological diagnosis model, and the current diagnosis results are output, including information such as pathological type, stage, and grade. This information is clearly displayed to the physician side through the visualization interface, so as to assist the physician side to make more accurate and rapid diagnostic decisions, improve the efficiency of pancreatic cancer diagnosis, and reduce the misdiagnosis rate caused by human factors.

[0078] In the specific implementation process, the original pancreatic cancer pathological images include images of pathological sections of benign and malignant pancreatic cancers, as well as images of pathological sections of pancreatic cancers in the early, middle, and late stages.

[0079] It can be understood that the original pancreatic cancer pathological images include histopathological section images of benign and malignant pancreatic cancers, as well as pathological sections at different stages of disease development, namely pathological section images of early pancreatic cancer, middle pancreatic cancer, and late pancreatic cancer. Specifically, early pancreatic cancer refers to the tumor being confined within the pancreas without obvious spread; middle pancreatic cancer refers to the tumor having invaded surrounding tissues or having local lymph node metastasis; late pancreatic cancer refers to the tumor having distant metastases, such as to organs such as the liver and lungs.

[0080] The preprocessing operations include:

[0081] Clean the obtained original pancreatic cancer pathological images, that is, remove the blurred and duplicate original pancreatic cancer pathological images, and generate the corresponding first pancreatic cancer pathological image;

[0082] Remove the noise in the first pancreatic cancer pathological image based on the Gaussian filtering algorithm, and generate the corresponding second pancreatic cancer pathological image;

[0083] Adopt the histogram equalization method to enhance the contrast of the second pancreatic cancer pathological image, and generate the corresponding third pancreatic cancer pathological image;

[0084] Perform normalization processing on the third pancreatic cancer pathological image to generate the corresponding target pancreatic cancer pathological image.

[0085] In practical applications, such as Figure 2As shown, firstly, an image cleaning operation can be performed on the acquired original pancreatic cancer pathology images to remove blurred and repeated images caused by poor shooting conditions or technical factors, thereby ensuring the purity of the data set and generating a clear first pancreatic cancer pathology image set.

[0086] Subsequently, the Gaussian filtering algorithm can be used to remove noise from the first pancreatic cancer pathology image. As a linear smoothing filtering technology, Gaussian filtering can effectively reduce or eliminate random noise in the image, such as Gaussian noise, while retaining the edge details of the image as much as possible to generate a smoother second pancreatic cancer pathology image.

[0087] In order to further improve the visualization effect of the image, the histogram equalization method is used to enhance the contrast of the second pancreatic cancer pathology image. This method adjusts the grayscale histogram distribution of the image to make the grayscale value in the image more evenly distributed, thereby significantly improving the contrast of the image and generating a third pancreatic cancer pathology image with enhanced contrast.

[0088] Finally, the third pancreatic cancer pathology image can be normalized, that is, the pixel values ​​of the image are mapped into a standardized range to eliminate the brightness differences between different images caused by factors such as lighting and shooting equipment, and generate a standardized target pancreatic cancer pathology image, laying the foundation for subsequent image segmentation, feature extraction and pathology analysis.

[0089] The construction module is used to adopt a deep learning algorithm to construct and train a pancreatic cancer pathology diagnosis model based on the training pancreatic cancer pathology images, and adjust the model parameters of the pancreatic cancer pathology diagnosis model in real time, specifically including:

[0090] Using a deep learning algorithm, the training pancreatic cancer pathology image is input into the input layer of the pancreatic cancer pathology diagnosis model, and the size of the training pancreatic cancer pathology image is ,in, represents the height of the training pancreatic cancer pathology image; represents the width of the training pancreatic cancer pathology image; represents the number of channels of the training pancreatic cancer pathology image;

[0091] The pancreatic cancer pathology diagnosis model includes multiple convolutional layers, each of which includes a preset number of convolutional kernels. By adjusting the size, number and step size of the convolutional kernels, introducing corresponding bias terms and activation functions, extracting local features of different scales and complexities in the training pancreatic cancer pathology image, and summing all the local features to obtain the corresponding first pancreatic cancer pathology feature. The mathematical formula of the first pancreatic cancer pathology feature is:

[0092]

[0093] In the formula, represents the pixel value of the first pancreatic cancer pathological feature at position ; represents the pixel value of the training pancreatic cancer pathological image at position ; represents the weight at position in the convolutional kernel; represents the activation function of the convolutional layer; represents the bias term of the convolutional layer.

[0094] The dimension of the first pancreatic cancer pathological feature is reduced through the pooling layer of the pancreatic cancer pathological diagnosis model, and the important feature information in the first pancreatic cancer pathological feature is retained, that is, the second pancreatic cancer pathological feature. The mathematical formula of the second pancreatic cancer pathological feature is:

[0095]

[0096] In the formula, represents the pixel value of the second pancreatic cancer pathological feature at position ; represents the pixel value of the first pancreatic cancer pathological feature at position ; represents the pooling window; represents taking the maximum value among all positions within the pooling window;

[0097] The first pancreatic cancer pathological feature and the second pancreatic cancer pathological feature are flattened to obtain the corresponding input pancreatic cancer pathological feature .

[0098] The input pancreatic cancer pathological feature is subjected to feature fusion and dimensionality reduction processing through the fully connected layer of the pancreatic cancer pathological diagnosis model to generate the corresponding output feature. The mathematical formula of the output feature is:

[0099]

[0100] In the formula, represents the output feature of the fully connected layer; represents the transpose of the weight of the fully connected layer; represents the input pancreatic cancer pathological feature; represents the bias term of the fully connected layer;

[0101] Based on the output layer of the pancreatic cancer pathological diagnosis model, the softmax classifier is used to map the output feature to the probability distribution of the pancreatic cancer diagnosis result, and the corresponding pancreatic cancer diagnosis category and diagnosis probability are obtained:

[0102]

[0103] In the formula, represents the diagnosis probability of pancreatic cancer diagnosis category ; represents the output feature corresponding to the pancreatic cancer diagnosis category ; represents the exponential sum of the output features corresponding to all pancreatic cancer diagnosis categories.

[0104] It can be understood that based on the deep learning framework, a convolutional neural network (CNN) architecture is selected as the basis and appropriately modified according to the characteristics of pancreatic cancer pathological images. Then, multiple convolutional layers can be designed, and each convolutional layer contains a certain number of convolutional kernels. By adjusting the size, number, and stride of the convolutional kernels, local features in the image can be extracted, and features of different scales and complexities can be captured.

[0105] Furthermore, a pooling layer can be added after the convolutional layer to reduce the dimension of the feature map, reduce the computational amount, and retain important feature information at the same time. After flattening the outputs of the convolutional layer and the pooling layer, corresponding feature fusion and dimensionality reduction can be performed on the output features through the fully connected layer, and finally connected to the output layer. The output layer usually uses the softmax function to map the features to the probability distribution of the diagnosis results and output the diagnosis results of benign, malignant, and different stages of pancreatic cancer.

[0106] Determine the loss function corresponding to the pancreatic cancer pathological diagnosis model to measure the difference between the pancreatic cancer diagnosis result and the actual pathological result. The mathematical formula of the loss function is:

[0107]

[0108] In the formula, represents the loss function; represents the number of input training pancreatic cancer pathological images; represents the loss value corresponding to a single training pancreatic cancer pathological image; represents the actual pathological result corresponding to the training pancreatic cancer pathological image; represents the pancreatic cancer diagnosis result corresponding to the training pancreatic cancer pathological image.

[0109] Calculate the gradient of the loss function with respect to the model parameters of the pancreatic cancer pathological diagnosis model based on the backpropagation algorithm:

[0110]

[0111] In the formula, represents the loss function; represents the weight of the fully connected layer; Represents the pancreatic cancer diagnosis result corresponding to the training pancreatic cancer pathological image; Represents the input pancreatic cancer pathological features; Represents the loss function The gradient of the weights of the fully connected layer ; Represents the loss function The gradient of the pancreatic cancer diagnosis result ; Represents the input pancreatic cancer pathological features The gradient of the weights of the fully connected layer ;

[0112] Update the model parameters of the pancreatic cancer pathological diagnosis model based on the gradient descent method to minimize the loss function:

[0113]

[0114] In the formula, Represents the weights of the fully connected layer after the -th iteration; Represents the weights of the fully connected layer after the -th iteration; Represents the learning rate, which is used to control the update step of the model parameters of the pancreatic cancer pathological diagnosis model; Represents the loss function; Represents the weights of the fully connected layer; Represents the loss function The gradient of the weights of the fully connected layer At the -th iteration.

[0115] In practical applications, continuously adjust the model parameters of the pancreatic cancer pathological diagnosis model through the backpropagation algorithm to minimize the difference between the output of the pancreatic cancer pathological diagnosis model and the true label.

[0116] The evaluation module is used to evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model based on the validation pancreatic cancer pathological images, and optimize the pancreatic cancer pathological diagnosis model according to the evaluation results, specifically including:

[0117] Input the validation pancreatic cancer pathological images into the pancreatic cancer pathological diagnosis model, determine the confusion matrix corresponding to the validation results, and the elements of the confusion matrix include true positives, false positives, true negatives, and false negatives. The true positives are the correct diagnosis of pancreatic cancer pathological results; the false positives are the incorrect diagnosis of pancreatic cancer pathological results; the true negatives are the correct diagnosis of non-pancreatic cancer pathological results; the false negatives are the incorrect diagnosis of non-pancreatic cancer pathological results;

[0118] Based on the confusion matrix, the diagnostic indicators corresponding to the pancreatic cancer pathological diagnosis model are calculated, where the diagnostic indicators include accuracy, precision, recall, F1 score, and false positive rate;

[0119] The formula for calculating the accuracy is:

[0120]

[0121] In the formula, represents the accuracy corresponding to the pancreatic cancer pathological diagnosis model; represents the true positive, that is, the correct diagnosis of the pancreatic cancer pathological result; represents the true negative, that is, the correct diagnosis of the non-pancreatic cancer pathological result; represents the false positive, that is, the incorrect diagnosis of the pancreatic cancer pathological result; represents the false negative, that is, the incorrect diagnosis of the non-pancreatic cancer pathological result;

[0122] The formula for calculating the precision is:

[0123]

[0124] In the formula, represents the precision corresponding to the pancreatic cancer pathological diagnosis model; represents the true positive, that is, the correct diagnosis of the pancreatic cancer pathological result; represents the false positive, that is, the incorrect diagnosis of the pancreatic cancer pathological result;

[0125] The formula for calculating the recall is:

[0126]

[0127] In the formula, represents the recall corresponding to the pancreatic cancer pathological diagnosis model; represents the true positive, that is, the correct diagnosis of the pancreatic cancer pathological result; represents the false negative, that is, the incorrect diagnosis of the non-pancreatic cancer pathological result;

[0128] The formula for calculating the F1 score is:

[0129]

[0130] In the formula, represents the F1 score corresponding to the pancreatic cancer pathological diagnosis model; represents the precision corresponding to the pancreatic cancer pathological diagnosis model; represents the recall corresponding to the pancreatic cancer pathological diagnosis model;

[0131] The formula for calculating the false positive rate is:

[0132]

[0133] In the formula, represents the false positive rate corresponding to the pancreatic cancer pathological diagnosis model; represents false positive examples, that is, the misdiagnosed pancreatic cancer pathological results; represents true negative examples, that is, the correctly diagnosed non-pancreatic cancer pathological results;

[0134] Analyze the diagnostic indicators, weigh the precision and recall rate, evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model, and optimize the pancreatic cancer pathological diagnosis model according to the evaluation results.

[0135] Furthermore, the optimized pancreatic cancer pathological diagnosis model can be integrated into the pancreatic cancer pathological diagnosis system to ensure the stable operation of the system and the processing of actual pathological images. As Figure 3 shown, based on the terminal device, by designing a user visualization interface, the physician side can input the pathological section to be diagnosed and view the diagnostic results of the pancreatic cancer pathological diagnosis system in real time.

[0136] It should be noted that the physician side refers to the terminal corresponding to the doctor. The pancreatic cancer pathological diagnosis system can not only provide accurate diagnostic results, but also provide diagnostic basis and possible misdiagnosis analysis, providing comprehensive auxiliary decision-making support for the physician side.

[0137] Through the introduction of the above embodiments, the present invention can obtain the original pancreatic cancer pathological images through the acquisition module, classification module, construction module, evaluation module and result module, and perform preprocessing operations on the original pancreatic cancer pathological images to generate corresponding target pancreatic cancer pathological images; classify the target pancreatic cancer pathological images to generate training pancreatic cancer pathological images and verification pancreatic cancer pathological images; adopt a deep learning algorithm to construct and train a pancreatic cancer pathological diagnosis model based on the training pancreatic cancer pathological images, and adjust the model parameters of the pancreatic cancer pathological diagnosis model in real time; evaluate the diagnostic performance of the pancreatic cancer pathological diagnosis model based on the verification pancreatic cancer pathological images, and optimize the pancreatic cancer pathological diagnosis model according to the evaluation results; obtain the current pancreatic cancer pathological image, input it into the pancreatic cancer pathological diagnosis model after preprocessing, output the current diagnostic result and display it to the physician side through the visualization interface. Through the deep learning algorithm, the present invention can automatically extract and analyze the feature information in the pathological images, realize the accurate diagnosis of pancreatic cancer, and improve the diagnostic accuracy; the system of the present invention can automatically complete the analysis and diagnosis of pathological images, greatly shortening the diagnostic time and improving the diagnostic efficiency; at the same time, the present invention can provide auxiliary decision-making support for doctors, helping doctors to more accurately judge the condition and formulate reasonable treatment plans, thereby promoting the development of medical intelligence and improving the level and quality of medical services.

[0138] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0139] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and this program can be stored in a computer-readable storage medium. The storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disk memories, tape memories, or any other medium that can be used to carry or store data and is computer-readable.

[0140] It should also be noted that the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

Claims

1. The deep learning-based artificial intelligence-assisted pathological diagnosis system for pancreatic cancer is characterized by: The system comprises: An acquisition module is used to acquire original pancreatic cancer pathology images, which include images of benign and malignant pancreatic cancer pathology sections, and images of early, mid and late pancreatic cancer pathology sections, and perform preprocessing operations on the original pancreatic cancer pathology images to generate corresponding target pancreatic cancer pathology images; A classification module, used for classifying the target pancreatic cancer pathology image to generate a training pancreatic cancer pathology image and a verification pancreatic cancer pathology image; A construction module is used to construct and train a pancreatic cancer pathology diagnosis model based on the training pancreatic cancer pathology image using a deep learning algorithm, wherein the pancreatic cancer pathology diagnosis model gradually learns and recognizes pathological features including cell morphology and nuclear division images in the image through iterative training, and adjusts model parameters including learning rate and weight attenuation of the pancreatic cancer pathology diagnosis model in real time; An evaluation module, used to evaluate the diagnostic performance of the pancreatic cancer pathology diagnosis model based on the verified pancreatic cancer pathology image, wherein the evaluation results include accuracy, recall rate, and F1 score, and to adjust the network structure and add data enhancement strategies according to the evaluation results to optimize the pancreatic cancer pathology diagnosis model; A result module is used to obtain the current pancreatic cancer pathology image, input it into the pancreatic cancer pathology diagnosis model after preprocessing, output the current diagnosis result and display it to the physician through a visual interface; When training the pancreatic cancer pathology diagnosis model, local features of different scales and complexities are extracted from the training pancreatic cancer pathology image, and all the local features are summed to obtain the corresponding first pancreatic cancer pathology feature. The dimension of the first pancreatic cancer pathological feature is reduced by the pooling layer of the pancreatic cancer pathological diagnosis model, and the second pancreatic cancer pathological feature in the first pancreatic cancer pathological feature is retained. The first pancreatic cancer pathology feature and the second pancreatic cancer pathology feature are flattened to obtain a corresponding input pancreatic cancer pathology feature.

2. The pancreatic cancer artificial intelligence-assisted pathological diagnosis system based on deep learning according to claim 1 is characterized in that: The pre-processing operation includes: Cleaning the acquired original pancreatic cancer pathology image, that is, removing blurred and repeated original pancreatic cancer pathology images, and generating a corresponding first pancreatic cancer pathology image; removing noise in the first pancreatic cancer pathology image based on a Gaussian filtering algorithm to generate a corresponding second pancreatic cancer pathology image; using a histogram equalization method to enhance the contrast of the second pancreatic cancer pathology image to generate a corresponding third pancreatic cancer pathology image; The third pancreatic cancer pathology image is normalized to generate the corresponding target pancreatic cancer pathology image.

3. The pancreatic cancer artificial intelligence-assisted pathological diagnosis system based on deep learning according to claim 1 is characterized in that: The construction module is used to adopt a deep learning algorithm to construct and train a pancreatic cancer pathology diagnosis model based on the training pancreatic cancer pathology images, and adjust the model parameters of the pancreatic cancer pathology diagnosis model in real time, specifically including: Using a deep learning algorithm, the training pancreatic cancer pathology image is input into the input layer of the pancreatic cancer pathology diagnosis model, and the size of the training pancreatic cancer pathology image is ,in, represents the height of the training pancreatic cancer pathology image; represents the width of the training pancreatic cancer pathology image; represents the number of channels of the training pancreatic cancer pathology image; The pancreatic cancer pathology diagnosis model includes multiple convolutional layers, each of which includes a preset number of convolutional kernels. By adjusting the size, number and step size of the convolutional kernels, introducing corresponding bias terms and activation functions, extracting local features of different scales and complexities in the training pancreatic cancer pathology image, and summing all the local features to obtain the corresponding first pancreatic cancer pathology feature. The mathematical formula of the first pancreatic cancer pathology feature is: In the formula, Indicates the location of the first pancreatic cancer pathological feature The pixel value of Represents the training pancreatic cancer pathology image at position The pixel value of Represents the position in the convolution kernel The weight of represents the activation function of the convolutional layer; Represents the bias term of the convolutional layer.

4. The pancreatic cancer artificial intelligence-assisted pathological diagnosis system based on deep learning according to claim 3 is characterized in that: The dimension of the first pancreatic cancer pathological feature is reduced by the pooling layer of the pancreatic cancer pathological diagnosis model, and important feature information in the first pancreatic cancer pathological feature is retained, that is, the second pancreatic cancer pathological feature. The mathematical formula of the second pancreatic cancer pathological feature is: In the formula, Indicates the second pancreatic cancer pathological feature at the location The pixel value of Indicates the location of the first pancreatic cancer pathological feature The pixel value of represents the pooling window; Indicates taking the maximum value among all positions within the pooling window; The first pancreatic cancer pathological feature and second pancreatic cancer pathological features After flattening, the corresponding input pancreatic cancer pathological features are obtained .

5. The deep learning-based artificial intelligence-assisted pathological diagnosis system for pancreatic cancer according to claim 4 is characterized in that: The input pancreatic cancer pathological features are subjected to feature fusion and dimensionality reduction processing through the fully connected layer of the pancreatic cancer pathological diagnosis model to generate corresponding output features. The mathematical formula of the output features is: In the formula, Represents the output features of the fully connected layer; represents the transpose of the weights of the fully connected layer; Indicates the input of pancreatic cancer pathological characteristics; represents the bias term of the fully connected layer; Based on the output layer of the pancreatic cancer pathological diagnosis model, a softmax classifier is used to map the output features to the probability distribution of the pancreatic cancer diagnosis results to obtain the corresponding pancreatic cancer diagnosis category and diagnosis probability: In the formula, Indicates pancreatic cancer diagnosis category The probability of diagnosis; Indicates pancreatic cancer diagnosis category The corresponding output features; Represents the exponential sum of the output features corresponding to all pancreatic cancer diagnosis classes.

6. The deep learning-based artificial intelligence-assisted pathological diagnosis system for pancreatic cancer according to claim 5 is characterized in that: The loss function corresponding to the pancreatic cancer pathological diagnosis model is determined to measure the difference between the pancreatic cancer diagnosis result and the actual pathological result. The mathematical formula of the loss function is: In the formula, represents the loss function; Represents the number of input training pancreatic cancer pathology images; Represents the loss value corresponding to a single training pancreatic cancer pathology image; Indicates the actual pathological results corresponding to the training pancreatic cancer pathological images; Represents the pancreatic cancer diagnosis results corresponding to the training pancreatic cancer pathology images.

7. The deep learning-based artificial intelligence-assisted pathological diagnosis system for pancreatic cancer according to claim 6 is characterized in that: The gradient of the loss function to the model parameters of the pancreatic cancer pathology diagnosis model is calculated based on the back propagation algorithm: In the formula, represents the loss function; represents the weight of the fully connected layer; Indicates the pancreatic cancer diagnosis results corresponding to the training pancreatic cancer pathology images; Indicates the input of pancreatic cancer pathological characteristics; Represents the loss function Weights for the fully connected layers The gradient of Represents the loss function Diagnosis of pancreatic cancer The gradient of Indicates the input of pancreatic cancer pathological characteristics Weights for the fully connected layers gradient.

8. The deep learning-based artificial intelligence-assisted pathological diagnosis system for pancreatic cancer according to claim 7 is characterized in that: The model parameters of the pancreatic cancer pathology diagnosis model are updated based on the gradient descent method to minimize the loss function: In the formula, Indicates The weight of the fully connected layer after iterations; Indicates The weight of the fully connected layer after iterations; represents the learning rate, which is used to control the update step size of the model parameters of the pancreatic cancer pathology diagnosis model; represents the loss function; represents the weight of the fully connected layer; Represents the loss function Weights for the fully connected layers In the The gradient at iteration .

9. The pancreatic cancer artificial intelligence-assisted pathological diagnosis system based on deep learning according to claim 1 is characterized in that: The evaluation module is used to evaluate the diagnostic performance of the pancreatic cancer pathology diagnosis model based on the verified pancreatic cancer pathology image, and optimize the pancreatic cancer pathology diagnosis model according to the evaluation result, and specifically includes: Inputting the verified pancreatic cancer pathology image into the pancreatic cancer pathology diagnosis model, determining a confusion matrix corresponding to the verification result, wherein the elements of the confusion matrix include true positive examples, false positive examples, true negative examples and false negative examples, wherein the true positive examples are correctly diagnosed pancreatic cancer pathology results; the false positive examples are incorrectly diagnosed pancreatic cancer pathology results; the true negative examples are correctly diagnosed non-pancreatic cancer pathology results; and the false negative examples are incorrectly diagnosed non-pancreatic cancer pathology results; Based on the confusion matrix, the diagnostic indicators corresponding to the pancreatic cancer pathological diagnosis model are calculated, wherein the diagnostic indicators include accuracy, precision, recall, F1 score and false positive rate; The calculation formula of the accuracy is: In the formula, Indicates the accuracy of the pancreatic cancer pathological diagnosis model; represents a true case, i.e., a correct diagnosis of pancreatic cancer pathology result; represents a true negative example, i.e., a correct diagnosis of non-pancreatic cancer pathology results; represents a false positive, i.e., an incorrect diagnosis of pancreatic cancer pathology result; represents a false negative example, i.e., an incorrect diagnosis of non-pancreatic cancer pathology results; The calculation formula of the accuracy is: In the formula, Indicates the accuracy rate corresponding to the pancreatic cancer pathological diagnosis model; represents a true case, i.e., a correct diagnosis of pancreatic cancer pathology result; represents a false positive, i.e., an incorrect diagnosis of pancreatic cancer pathology result; The calculation formula of the recall rate is: In the formula, Represents the recall rate corresponding to the pancreatic cancer pathological diagnosis model; represents a true case, i.e., a correct diagnosis of pancreatic cancer pathology result; represents a false negative example, i.e., an incorrect diagnosis of non-pancreatic cancer pathology results; The calculation formula of the F1 score is: In the formula, Indicates the F1 score corresponding to the pancreatic cancer pathological diagnosis model; Indicates the accuracy rate corresponding to the pancreatic cancer pathological diagnosis model; Represents the recall rate corresponding to the pancreatic cancer pathological diagnosis model; The calculation formula of the false positive rate is: In the formula, Represents the false positive rate corresponding to the pancreatic cancer pathological diagnosis model; represents a false positive, i.e., an incorrect diagnosis of pancreatic cancer pathology result; represents a true negative example, i.e., a correct diagnosis of non-pancreatic cancer pathology results; The diagnostic indicators are analyzed, and the precision and recall are weighed to evaluate the diagnostic performance of the pancreatic cancer pathology diagnosis model, and the pancreatic cancer pathology diagnosis model is optimized according to the evaluation results.

Images