Pathological image intelligent diagnosis method based on deep learning
By using a deep learning-based intelligent diagnostic method for pathological images, which automatically extracts features from pathological images using convolutional neural networks, the problem of reliance on manual diagnosis and inconsistencies in diagnostic results is solved, thus improving the efficiency and accuracy of pathological diagnosis and providing reliable auxiliary diagnostic support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGRAO KANGWAN MEDICAL TESTING LABORATORY CO LTD
- Filing Date
- 2026-01-20
- Publication Date
- 2026-05-01
AI Technical Summary
In existing technologies, pathological diagnosis relies on manual observation and experience-based judgment, which is labor-intensive, highly subjective, and results in significant differences in diagnosis between different doctors. Furthermore, existing image processing or machine learning-based methods are not robust enough to adapt to different staining conditions and tissue morphology changes, resulting in low accuracy in identifying and classifying complex pathological structures.
A deep learning-based intelligent diagnostic method for pathological images is adopted, including preprocessing, image segmentation, deep learning diagnostic model training, and result labeling. Convolutional neural networks (CNNs) are used for feature extraction and classification to generate pathological diagnostic results and their confidence labels.
It enables automatic extraction of pathological image features, reduces reliance on human experience, improves the efficiency and consistency of pathological diagnosis, and provides a reliable auxiliary diagnostic tool.
Smart Images
Figure CN121962097A_ABST
Abstract
Description
A Deep Learning-Based Intelligent Diagnostic Method for Pathological Images Technical Field
[0001] This invention belongs to the field of medical image processing and artificial intelligence technology, specifically referring to an intelligent diagnostic method for pathological images based on deep learning. Background Technology
[0002] Pathological diagnosis is a crucial basis for confirming diseases in clinical medicine. Traditional pathological diagnosis mainly relies on pathologists' manual observation and experience in interpreting pathological slides. With the continuous improvement of pathological slide resolution, especially the widespread application of digital full-view pathological slide images, the amount of information contained in a single pathological image has increased significantly. Manual slide interpretation is labor-intensive, highly subjective, and there are certain differences in diagnostic results between different doctors.
[0003] While existing pathological analysis methods based on image processing or machine learning have emerged, most of these methods rely on manually designed features, resulting in poor robustness and difficulty in adapting to changes in different staining conditions and tissue morphology. Furthermore, the accuracy of automatic identification and classification of complex pathological structures still needs improvement. Summary of the Invention
[0004] In order to overcome the shortcomings of the prior art, the present invention provides a method for intelligent diagnosis of pathological images based on deep learning.
[0005] The technical solution adopted in this invention is as follows: including the following steps: S1, acquiring a pathological image to be diagnosed; S2, preprocessing the pathological image, the preprocessing including at least image denoising, color standardization, and size normalization; S3, segmenting the preprocessed pathological image to obtain at least one pathological tissue region to be analyzed; S4, inputting the pathological tissue region to be analyzed into a pre-trained deep learning diagnostic model for feature extraction and classification prediction to obtain the corresponding pathological diagnosis result and its confidence level; S5, when the confidence level is greater than a preset threshold, generating pathological diagnosis marker information at the corresponding position in the pathological image; S6, outputting a pathological image containing the pathological diagnosis marker information.
[0006] Furthermore, the pathological images include hematoxylin-eosin (HE) stained pathological slide images, immunohistochemical stained pathological images, or digital full-view pathological slide images.
[0007] Furthermore, the image segmentation in step S3 adopts any one of threshold segmentation, region growing, cluster analysis, graph model segmentation, or deep learning segmentation network.
[0008] Furthermore, the deep learning diagnostic model is a model based on the Convolutional Neural Network (CNN) architecture.
[0009] Furthermore, the deep learning diagnostic model includes the following steps during the training phase: T1, acquiring a large number of labeled pathological image samples and their corresponding pathological diagnostic labels; T2, using the pathological image samples as model input and the corresponding pathological diagnostic labels as supervision signals to train the deep learning diagnostic model; T3, iteratively optimizing the model parameters based on the difference between the model prediction results and the true labels until the model performance meets the preset training requirements.
[0010] Furthermore, the pathological diagnosis results include at least one of normal tissue, benign lesions, or malignant lesions.
[0011] Furthermore, the malignant lesion includes pathological grading results and tumor differentiation information.
[0012] Furthermore, the pathological diagnostic marker information includes the diagnostic category, confidence level value, and corresponding spatial location information.
[0013] Furthermore, prior to step S4, the area of the pathological tissue to be analyzed is segmented into image blocks to make its size consistent with the input size of the deep learning diagnostic model.
[0014] Furthermore, the pathological diagnosis report is output in the form of an electronic report or a visual image to assist pathologists in making clinical diagnostic decisions.
[0015] The beneficial effects achieved by the present invention using the above structure are as follows: The present invention provides a pathological image intelligent diagnosis method based on deep learning, which achieves the following beneficial effects: (1) Automatically extracting pathological image features through deep learning model, reducing the dependence on human experience.
[0016] (2) It can automatically locate and mark the lesion area in large-size pathological images.
[0017] (3) Improve the efficiency and consistency of pathological diagnosis.
[0018] (4) It can be used as an auxiliary diagnostic tool to provide reliable reference for pathologists. Attached Figure Description
[0019] Figure 1 is a flowchart of a pathological image intelligent diagnosis method based on deep learning proposed in this invention.
[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0022] (a) Acquisition of pathological images The pathological images to be diagnosed can be acquired through digital pathological scanning equipment, such as fully automated slide scanners, which convert traditional glass slide pathological sections into digital pathological images.
[0023] Pathological images include: HE (hematoxylin-eosin) stained pathological slide images, immunohistochemical stained pathological images, and digital full-view pathological slide images. Commonly used formats include .svs, .GIFf, and .ndpi.
[0024] Example parameters: Scan resolution: 20×(0.5μm / pixel); Image color depth: 24-bit RGB; Scan mode: Bright field.
[0025] The above-mentioned image acquisition methods are conventional techniques in the field of pathological image acquisition, and this invention does not limit them.
[0026] (II) Pathological Image Preprocessing To reduce the impact of noise, staining differences and size differences on subsequent analysis, pathological images are preprocessed, including the following steps: Image denoising, using Gaussian filtering (e.g., σ=1~2) and median filtering (window 5×5) to reduce scanning noise or random noise.
[0027] Algorithm selection is based on noise type: Gaussian noise is filtered using Gaussian filtering, and salt-and-pepper noise is filtered using median filtering.
[0028] Color normalization is based on the Reinhard method: mapping the image color space to the mean and standard deviation of a reference template; the Macenko method: estimating the staining basis matrix through singular value decomposition, achieving normalization after stain separation; and adjusting the brightness of the HE image to the range of 0~255.
[0029] Size normalization is performed by scaling local slice regions to ensure that the width and height are consistent with the model input size (512×512 pixels); bilinear interpolation is used to ensure that the image texture features are not distorted as much as possible.
[0030] (III) Pathological Image Segmentation: After preprocessing, the pathological images are segmented to obtain the pathological tissue regions to be analyzed. The following techniques can be used: threshold-based methods: Otsu adaptive thresholding algorithm for grayscale image segmentation, with a threshold range of 0~255; converting RGB to grayscale before segmentation.
[0031] Image segmentation separates pathological tissue areas from background areas, thereby reducing the interference of irrelevant areas on diagnostic results.
[0032] (iv) Input preparation for deep learning diagnostic analysis model: The segmented pathological tissue region is divided into multiple image blocks, with a common size of 256×256 pixels; the image blocks can be data augmented: rotate ±90°, horizontal / vertical flip, random brightness / contrast adjustment, and color jitter.
[0033] The deep learning diagnostic model is based on a convolutional neural network (CNN), specifically ResNet50. The model extracts texture, morphology, and structural features through multiple convolutions. The output is the diagnostic category and its corresponding confidence level.
[0034] Prediction and fusion are performed separately for each image patch; the prediction results for the entire pathological tissue region can be obtained by majority voting, average confidence, or weighted fusion to obtain the final diagnostic result.
[0035] (iv) Model training process description Data preparation Obtain a large number of labeled pathological image samples and their corresponding diagnostic labels (benign / malignant, grade type, etc.); it is recommended that the sample size be at least several thousand to tens of thousands of image blocks.
[0036] The training method uses image patches as input and diagnostic labels as supervision signals; it uses the backpropagation algorithm and optimization algorithm (SGD or Adam); the initial learning rate can be set to 0.001, and the training is carried out for 50~100 epochs, which are then adjusted according to the performance on the validation set; the loss function is cross-entropy loss to handle the class imbalance problem.
[0037] Model evaluation and optimization: Accuracy, F1 score, and AUC are evaluated on the validation set; EarlyStopping is used to optimize the training process.
[0038] (vi) Diagnostic result labeling and output labeling: For image blocks with a confidence level ≥ threshold (e.g., 0.8) of the model output, diagnostic labels are generated in the original pathological image; the labeling information includes: diagnostic category, confidence level value, and image coordinates.
[0039] Output method: Output an electronic report (PDF / HTML) that includes diagnostic categories, confidence statistics, and sample image patches.
[0040] Gaussian filter kernel: Convolution calculation: .
[0041] Reinhard method ; ; ;in , For the source image and reference image in , , The mean and standard deviation of the channel.
[0042] The Otsu thresholding algorithm for image segmentation aims to maximize the inter-class variance. : ; , The percentage of pixels in the foreground and background; , Mean of foreground and background; threshold : .
[0043] Deep learning diagnostic model formula 1) Convolutional layer operation ; Input feature map; : Convolution kernel; : Bias; Output is passed through the non-linear activation function ReLU: 2) Softmax output probability ; This is the output of the final fully connected layer; This represents the number of categories.
[0044] 3) Cross-entropy loss function ; : Real tags; : Predicted probability.
[0045] Model training optimization 1) Gradient descent parameter update SGD: ; Here is the learning rate; Adam can be written as: ; ; .
[0046] Confidence labeling and output formula 1) Confidence threshold determination: ; Set to 0.8~0.9; Color mapping: Directly using RGB mapping: .
[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
[0048] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A deep learning-based intelligent diagnostic method for pathological images, characterized in that, The process includes the following steps: S1, acquiring a pathological image to be diagnosed; S2, preprocessing the pathological image, the preprocessing including at least image denoising, color normalization and size normalization; S3, segmenting the preprocessed pathological image to obtain at least one pathological tissue region to be analyzed. S4. Input the pathological tissue region to be analyzed into a pre-trained deep learning diagnostic model to perform feature extraction and classification prediction, and obtain the corresponding pathological diagnosis result and its confidence level; S5. When the confidence level is greater than a preset threshold, generate pathological diagnosis label information at the corresponding position in the pathological image; S6. Output the pathological image containing the pathological diagnosis label information.
2. The intelligent pathological image diagnosis method based on deep learning according to claim 1, characterized in that: The pathological images include hematoxylin-eosin (HE) stained pathological slide images, immunohistochemical stained pathological images, or digital full-view pathological slide images.
3. The intelligent pathological image diagnosis method based on deep learning according to claim 2, characterized in that: The image segmentation in step S3 uses any one of the following: threshold segmentation, region growing, cluster analysis, graphical model segmentation, or deep learning segmentation network.
4. The intelligent diagnostic method for pathological images based on deep learning according to claim 3, characterized in that: The deep learning diagnostic model is based on the Convolutional Neural Network (CNN) architecture.
5. The intelligent pathological image diagnosis method based on deep learning according to claim 4, characterized in that: The deep learning diagnostic model includes the following steps during the training phase: T1, acquiring a large number of labeled pathological image samples and their corresponding pathological diagnostic labels; T2, using the pathological image samples as model input and the corresponding pathological diagnostic labels as supervision signals to train the deep learning diagnostic model. T3. Iteratively optimize the model parameters based on the difference between the model's prediction results and the true labels until the model performance meets the preset training requirements.
6. The intelligent diagnostic method for pathological images based on deep learning according to claim 5, characterized in that: The pathological diagnosis results include at least one of normal tissue, benign lesions, or malignant lesions.
7. The intelligent pathological image diagnosis method based on deep learning according to claim 6, characterized in that: The malignant lesions include pathological grading results and tumor differentiation information.
8. The intelligent pathological image diagnosis method based on deep learning according to claim 7, characterized in that: The pathological diagnostic marker information includes the diagnostic category, confidence level value, and corresponding spatial location information.
9. The intelligent diagnostic method for pathological images based on deep learning according to claim 8, characterized in that: Before step S4, the area of the pathological tissue to be analyzed is segmented into image blocks to make its size consistent with the input size of the deep learning diagnostic model.
10. The intelligent diagnostic method for pathological images based on deep learning according to claim 9, characterized in that: The pathology diagnosis report is output in the form of an electronic report or a visual image to assist pathologists in making clinical diagnostic decisions.