Lupus nephritis therapeutic effect prediction system based on multiple staining kidney biopsy pathological images
By extracting features and performing similarity analysis on multi-stain renal biopsy pathological images, combined with model training and optimization of the prediction model, the problem of insufficient information mining from biopsy samples of lupus nephritis was solved, and more accurate patient classification and severity assessment were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SECOND AFFILIATED HOSPITAL OF XIAN MEDICAL UNIV
- Filing Date
- 2026-03-06
- Publication Date
- 2026-05-29
AI Technical Summary
In existing technologies, the image information mining effect of biopsy samples in lupus nephritis is poor, making it difficult for doctors to accurately assess the patient's subtype and severity.
By extracting features, performing similarity analysis, and training models from multi-stain renal biopsy pathological images, reference patients similar to the target patient's biopsy samples are identified. Their image features and labels are fused to form training data, and adjustment coefficients are used to optimize the prediction model to predict the cure probability of the target patient.
It improves the information mining effect of biopsy sample images of lupus nephritis, assists doctors in accurately assessing the patient's subtype and severity, and improves the accuracy and efficiency of diagnosis.
Smart Images
Figure CN121810673B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology, specifically to a lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images. Background Technology
[0002] Lupus nephritis is one of the most common and serious complications of systemic lupus erythematosus. It is a disease characterized by immune damage to both kidneys with different pathological types, accompanied by significant clinical manifestations of kidney damage. The disease is highly heterogeneous, which brings great challenges to its treatment. Clinically, patients with lupus nephritis have long faced the dilemma of limited efficacy of traditional immunosuppressants and high relapse rate, and patients' responses to treatment vary significantly.
[0003] Currently, renal biopsy is the gold standard for the diagnosis and classification of lupus nephritis. It assists doctors in determining the specific type and severity of lupus nephritis through histological examination. Existing technology often uses high-resolution scanners to scan traditionally stained biopsy sections to form full-wave images. By analyzing the patient's full-wave images and combining them with the results of other examinations, doctors can assess the specific type and severity of the patient's lupus nephritis and carry out corresponding treatment accordingly.
[0004] However, it should be noted that due to the high heterogeneity of lupus nephritis, relying solely on manual analysis of full-wavelength images can easily lead to misjudgments by doctors regarding the specific subtype and severity of lupus nephritis in patients. In other words, current technologies are not very effective at extracting information from biopsy sample images of lupus nephritis. Summary of the Invention
[0005] The purpose of this invention is to provide a lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images, which solves the technical problem of poor information mining effect of existing technologies on biopsy sample images of lupus nephritis.
[0006] In a first aspect, one embodiment of the present invention provides a lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images, the system comprising:
[0007] The feature extraction module is used to extract features from the stained biopsy images of each patient in the lupus nephritis patient set under each staining method, so as to obtain multiple image features of each patient. The patient set includes a target patient and multiple historical patients. The multiple image features of a patient correspond one-to-one with multiple staining methods. The stained biopsy image is the image acquired after the biopsy sample of the corresponding patient has been processed by the corresponding staining method.
[0008] The similarity analysis module is used to perform similarity analysis between multiple image features of the target patient and multiple image features of each historical patient, so as to distinguish reference patients from non-reference patients from multiple historical patients. The sample similarity of the reference patient is greater than that of the non-reference patient. The sample similarity is used to represent the degree of similarity between the biopsy samples of the historical patients and the biopsy samples of the target patient.
[0009] The data preparation module is used to fuse multiple image features of the same reference patient and their corresponding patient labels to obtain multiple training data, wherein the patient labels are used to indicate whether the corresponding reference patient has been cured.
[0010] The prediction module is used to train a model based on multiple training data to obtain a prediction model, and to process multiple image features of the target patient based on the prediction model to obtain prediction information, wherein the prediction information is used to indicate the probability of the target patient being cured. The loss function of the prediction model includes an adjustment coefficient, which is used to adjust the prediction loss of the corresponding training data. The adjustment coefficient is determined based on the sample similarity and sample specificity of the corresponding reference patient. The sample specificity is used to indicate the specificity of the biopsy sample of the corresponding reference patient among multiple reference patients.
[0011] In one embodiment, the step of performing similarity analysis based on multiple image features of the target patient and multiple image features of each historical patient to distinguish reference patients from non-reference patients among multiple historical patients includes:
[0012] In multiple historical patients, the similarity between different image features of each historical patient and the target patient under each staining method is analyzed to obtain multiple feature similarity values of each historical patient. The multiple feature similarity values of a historical patient correspond one-to-one with multiple staining methods.
[0013] Analyze the central tendency of multiple feature similarity values for each historical patient to obtain the sample similarity for each historical patient;
[0014] Based on the sample similarity of each historical patient, reference patients and non-reference patients are distinguished from multiple historical patients.
[0015] In one embodiment, the step of obtaining the adjustment coefficient includes:
[0016] In multiple reference patients, the differences between each reference patient and other reference patients under each staining mode are analyzed to obtain multiple feature difference values for each reference patient. The multiple feature difference values of a reference patient correspond one-to-one with multiple staining modes.
[0017] Analyze the central tendency of multiple characteristic differences for each reference patient to obtain the sample specificity for each reference patient;
[0018] The adjustment coefficient for each reference patient is determined based on the sample specificity and sample similarity of each reference patient.
[0019] In one embodiment, the step of analyzing the differences between different image features of each reference patient and other reference patients under each staining modality, and obtaining multiple feature difference values for each reference patient, includes:
[0020] In multiple reference patients, the outlier degree of each image feature of each reference patient in the corresponding feature set is analyzed to obtain multiple feature difference values for each reference patient. The feature set consists of multiple image features of multiple reference patients under the corresponding staining mode.
[0021] In one embodiment, the sample specificity is positively correlated with the corresponding adjustment coefficient, and the sample similarity is positively correlated with the corresponding adjustment coefficient.
[0022] In one embodiment, the step of determining the adjustment coefficient for each reference patient based on the sample specificity and sample similarity of each reference patient includes:
[0023] Calculate the sum of sample specificity and sample similarity for each reference patient to obtain the initial coefficient value for each reference patient;
[0024] The initial coefficient values for each reference patient were normalized to obtain the adjusted coefficients for each reference patient.
[0025] In one embodiment, the step of normalizing the initial coefficient values for each reference patient to obtain the adjusted coefficients for each reference patient includes:
[0026] Based on the training batch corresponding to each reference patient, the initial coefficient value of each reference patient is normalized to obtain the adjustment coefficient of each reference patient.
[0027] Each training batch corresponds to at least two reference patients. Different training batches have the same number of reference patients, but the reference patients in different training batches are completely different. The sum of the adjustment coefficients for all reference patients in the same training batch is 1.
[0028] In one embodiment, the step of extracting features from stained biopsy images of each patient in a set of lupus nephritis patients under each staining modality to obtain multiple image features for each patient includes:
[0029] In a patient set with lupus nephritis, corner detection was performed on the stained biopsy images of each patient under each staining method to obtain the corner information corresponding to each patient under each staining method.
[0030] The corner point information of each patient under multiple staining methods is vectorized to obtain multiple image features for each patient.
[0031] In one embodiment, the prediction module is further configured to: fuse the prediction information with the medical history data of a key reference patient and output the result, wherein the key reference patient is the reference patient with the highest initial coefficient value among multiple reference patients.
[0032] In one embodiment, the medical history data includes key reference patient stained biopsy images under multiple staining methods, symptom description text, and treatment plans.
[0033] Secondly, another embodiment of the present invention provides a method for predicting the efficacy of treatment for lupus nephritis based on multi-stain renal biopsy pathological images, the method comprising:
[0034] In the patient set of lupus nephritis, feature extraction was performed on the stained biopsy images of each patient under each staining method to obtain multiple image features of each patient. The patient set includes a target patient and multiple historical patients. Multiple image features of a patient correspond one-to-one with multiple staining methods. The stained biopsy image is the image acquired after the biopsy sample of the corresponding patient has been processed by the corresponding staining method.
[0035] Similarity analysis is performed based on multiple image features of the target patient and multiple image features of each historical patient to distinguish reference patients from non-reference patients among multiple historical patients. The sample similarity of reference patients is greater than that of non-reference patients. The sample similarity is used to represent the degree of similarity between biopsy samples of historical patients and biopsy samples of the target patient.
[0036] Multiple image features of the same reference patient and their corresponding patient labels are fused to obtain multiple training data, where the patient labels are used to indicate whether the corresponding reference patient has been cured.
[0037] A model is trained based on multiple training data to obtain a prediction model. The prediction model is then used to process multiple image features of the target patient to obtain prediction information, which indicates the probability of the target patient being cured. The loss function of the prediction model includes an adjustment coefficient, which is used to adjust the prediction loss of the corresponding training data. The adjustment coefficient is determined based on the sample similarity and sample specificity of the corresponding reference patient. The sample specificity indicates the degree of specificity of the biopsy sample of the corresponding reference patient among multiple reference patients.
[0038] Thirdly, in another embodiment of the present invention, an electronic device is provided, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method described in the second aspect above.
[0039] Fourthly, in another embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in the second aspect above.
[0040] The present invention has the following beneficial effects:
[0041] This invention first uses image feature extraction and feature similarity analysis to screen a patient set of lupus nephritis patients, identifying several reference patients whose biopsy samples are highly similar to those of the target patient. Then, multiple image features of the same reference patient and their corresponding patient labels are fused to form training data, which is used to train the model. Finally, the trained model is used to predict the probability of the target patient being cured. During the model training phase, the loss value of each training sample is adaptively adjusted based on the sample similarity and specificity to the reference patients. This helps the model focus more on training data that are similar to the target patient's biopsy sample and effectively distinguishable from other reference patients' biopsy samples, matching the highly heterogeneous nature of lupus nephritis. This better mines disease-related information from stained biopsy images, improving the information mining effect of lupus nephritis biopsy sample images and assisting doctors in accurately assessing the specific subtype and severity of lupus nephritis in patients. Attached Figure Description
[0042] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0043] Figure 1 This is a schematic diagram of the structure of a lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images provided in an embodiment of the present invention;
[0044] Figure 2 This is a flowchart illustrating a method for predicting the treatment efficacy of lupus nephritis based on multi-stain renal biopsy pathological images provided in an embodiment of the present invention.
[0045] Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0046] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0048] The following description, in conjunction with the accompanying drawings, details the specific scheme of the lupus nephritis efficacy prediction system based on multi-stain renal biopsy pathological images provided by the present invention.
[0049] In one embodiment, the present invention provides a lupus nephritis treatment efficacy prediction system 100 based on multi-stain renal biopsy pathological images, such as... Figure 1 As shown, the system 100 includes:
[0050] The feature extraction module 101 is used to extract features from the stained biopsy images of each patient in the lupus nephritis patient set under each staining method, so as to obtain multiple image features for each patient.
[0051] The patient set includes a target patient and multiple historical patients. Multiple image features of a patient correspond one-to-one with multiple staining methods. The stained biopsy image is an image acquired after the biopsy sample of the corresponding patient has been processed by the corresponding staining method.
[0052] In this invention, "target patient" refers to any lupus nephritis patient for whom disease status confirmation (determining the specific subtype and severity of lupus nephritis) is required; "historical patient" refers to lupus nephritis patients who have completed treatment (i.e., discharged from hospital); and "biopsy sample" refers to the sample collected from the corresponding patient through a kidney biopsy.
[0053] The aforementioned staining methods include at least four: HE staining, PAS staining, Mason staining, and silver staining. In application, after staining the tissue sections of samples collected from a patient's renal biopsy using the corresponding staining method, a high-resolution scanner is used to scan the stained renal biopsy tissue sections to obtain the stained biopsy image of the corresponding patient under the corresponding staining method.
[0054] Specifically, in a patient set with lupus nephritis, the steps of extracting features from stained biopsy images of each patient under each staining modality to obtain multiple image features for each patient include:
[0055] In a patient set with lupus nephritis, corner detection was performed on the stained biopsy images of each patient under each staining method to obtain the corner information corresponding to each patient under each staining method.
[0056] The corner point information of each patient under multiple staining methods is vectorized to obtain multiple image features for each patient.
[0057] The specific algorithm used for corner detection can be adaptively determined according to the actual situation. In this invention, the algorithm used... The corner detection algorithm completes the corner detection operation described above.
[0058] The corner information includes at least multiple corners of the stained biopsy image of the corresponding patient under the corresponding staining method, as well as a descriptor for each corner.
[0059] In this invention, the process of vectorizing corner information into corresponding image features is based on the bag-of-words model. Specifically, the process involves: clustering all descriptors corresponding to the patient set to obtain multiple clusters (the number of clusters can be specified based on empirical adaptability, and is set to 500 in this invention); then summarizing the center points of multiple clusters (the descriptor with the smallest sum of Euclidean distances to other descriptors in the cluster, which can be understood as visual words in the visual dictionary referred to below) into a visual dictionary; finally, encoding multiple descriptors corresponding to each patient under each staining mode according to the visual dictionary, thereby obtaining the image features corresponding to each patient under each staining mode (specifically, the frequency of occurrence of each visual word in the visual dictionary).
[0060] The similarity analysis module 102 is used to perform similarity analysis based on multiple image features of the target patient and multiple image features of each historical patient, so as to distinguish the reference patient from the non-reference patient among multiple historical patients.
[0061] Among them, the sample similarity of the reference patient was greater than that of the non-reference patient. The sample similarity was used to represent the degree of similarity between biopsy samples of historical patients and biopsy samples of the target patient.
[0062] Specifically, the step of performing similarity analysis based on multiple image features of the target patient and multiple image features of each historical patient to distinguish reference patients from non-reference patients among multiple historical patients includes:
[0063] In multiple historical patients, the similarity between different image features of each historical patient and the target patient under each staining method is analyzed to obtain multiple feature similarity values of each historical patient. The multiple feature similarity values of a historical patient correspond one-to-one with multiple staining methods.
[0064] Analyze the central tendency of multiple feature similarity values for each historical patient to obtain the sample similarity for each historical patient;
[0065] Based on the sample similarity of each historical patient, reference patients and non-reference patients are distinguished from multiple historical patients.
[0066] In this invention, the feature similarity value is specifically the absolute value of the cosine similarity between two different image features of the corresponding historical patient and the target patient under the corresponding staining method. The sample similarity is specifically the average of multiple feature similarity values of the historical patient.
[0067] Since the same biopsy sample will show different dimensions of disease information under different staining methods, the above measures can effectively integrate the feature information of biopsy samples under different dimensions, so as to accurately quantify the similarity between different biopsy samples and ensure the accuracy and reliability of the calculated sample similarity.
[0068] The higher the sample similarity, the greater the similarity between the biopsy sample of the corresponding historical patient and the biopsy sample of the target patient. In other words, the higher the reference value of the corresponding historical patient's medical history data (e.g., stained biopsy images under multiple staining methods, symptom description text, symptom diagnosis results, and treatment plan) for the diagnosis of the target patient's disease.
[0069] In one example, several historical patients (e.g., 1000) with the highest sample similarity among multiple historical patients can be identified as reference patients.
[0070] In another example, historical patients whose sample similarity is greater than a similarity threshold (set to 0.6 in this invention based on experience) can be identified as reference patients.
[0071] In the above example, among multiple historical patients, those who were not identified as reference patients are considered non-reference patients.
[0072] The data preparation module 103 is used to fuse multiple image features of the same reference patient and their corresponding patient labels to obtain multiple training data.
[0073] The patient tag is used to indicate whether the corresponding reference patient has been cured.
[0074] In this invention, the determination of whether a reference patient has been cured is as follows:
[0075] For the corresponding reference patient, the values of each key monitoring indicator in the last data monitoring operation before discharge after treatment are all within the normal reference range. If they are all within the normal reference range, the corresponding reference patient is determined to be cured; otherwise, the corresponding reference patient is determined to be not cured.
[0076] The key detection indicators mentioned above include, but are not limited to, urine protein / creatinine ratio, serum albumin, serum creatinine, and anti-dsDNA antibody.
[0077] Multiple reference patients and multiple training data are matched one-to-one. The training data is formed by splicing multiple image features of the corresponding reference patients in a set order (such as the order of HE staining, PAS staining, Mason staining and silver staining). The ground truth label of the training data is the patient label of the corresponding reference patient (the label value is 1 if cured, and 0 if not cured).
[0078] The prediction module 104 is used to train a model based on multiple training data to obtain a prediction model, and to process multiple image features of the target patient based on the prediction model to obtain prediction information.
[0079] The prediction information is used to indicate the probability of the target patient being cured. The loss function of the prediction model includes an adjustment coefficient, which is used to adjust the prediction loss of the corresponding training data. The adjustment coefficient is determined based on the sample similarity and sample specificity of the corresponding reference patient. The sample specificity is used to indicate the specificity of the biopsy sample of the corresponding reference patient among multiple reference patients.
[0080] The higher the sample specificity, the higher the specificity of the biopsy sample of the corresponding reference patient among multiple reference patients. In other words, the higher the probability that the biopsy sample of the corresponding reference patient indicates a rare subtype, and thus the higher the medical reference value of the biopsy sample of the corresponding reference patient among all biopsy samples corresponding to lupus nephritis.
[0081] In this invention, ResNet convolutional neural network is specifically used to construct the above prediction model. In practical applications, other model frameworks that support prediction functions can also be used to construct the above prediction model, and this invention does not limit this.
[0082] This invention first uses image feature extraction and feature similarity analysis to screen a patient set of lupus nephritis patients, identifying several reference patients whose biopsy samples are highly similar to those of the target patient. Then, multiple image features of the same reference patient and their corresponding patient labels are fused to form training data, which is used to train the model. Finally, the trained model is used to predict the probability of the target patient being cured. During the model training phase, the loss value of each training sample is adaptively adjusted based on the sample similarity and specificity to the reference patients. This helps the model focus more on training data that are similar to the target patient's biopsy sample and effectively distinguishable from other reference patients' biopsy samples, matching the highly heterogeneous nature of lupus nephritis. This better mines disease-related information from stained biopsy images, improving the information mining effect of lupus nephritis biopsy sample images and assisting doctors in accurately assessing the specific subtype and severity of lupus nephritis in patients.
[0083] Specifically, the step of obtaining the adjustment coefficient includes:
[0084] In multiple reference patients, the differences between each reference patient and other reference patients under each staining mode are analyzed to obtain multiple feature difference values for each reference patient. The multiple feature difference values of a reference patient correspond one-to-one with multiple staining modes.
[0085] Analyze the central tendency of multiple characteristic differences for each reference patient to obtain the sample specificity for each reference patient;
[0086] The adjustment coefficient for each reference patient is determined based on the sample specificity and sample similarity of each reference patient.
[0087] Furthermore, the step of analyzing the differences between each reference patient and other reference patients in different image features under each staining method, and obtaining multiple feature difference values for each reference patient, includes:
[0088] In multiple reference patients, the outlier degree of each image feature of each reference patient in the corresponding feature set is analyzed to obtain multiple feature difference values for each reference patient. The feature set consists of multiple image features of multiple reference patients under the corresponding staining mode.
[0089] The feature difference value can be the average of multiple image difference values for the corresponding image feature in the corresponding feature set, where the image difference value is the difference between the numerical value 1 and the absolute value of the cosine similarity between the corresponding two image features.
[0090] For example, the first The reference patient at the 1st Feature difference values of image features corresponding to each staining method It can be represented as:
[0091]
[0092] in, Indicates the first The reference patient at the 1st In the feature set corresponding to the staining method, except for the , The reference patient at the 1st The number of image features other than the image features corresponding to each coloring method. Indicates the first The reference patient at the 1st The image features corresponding to the i-th coloring method and their corresponding feature set in the i-th coloring method The absolute value of the cosine similarity between image features.
[0093] Sample specificity is specifically the average of multiple characteristic differences corresponding to the reference patient.
[0094] Specifically, the sample specificity is positively correlated with the corresponding adjustment coefficient, and the sample similarity is positively correlated with the corresponding adjustment coefficient.
[0095] It should be understood that the higher the sample specificity, the more important the training sample corresponding to the reference patient is in participating in model training, and the higher the adjustment coefficient should be. Similarly, the higher the sample similarity, the more important the training sample corresponding to the reference patient is in participating in model training, and the higher the adjustment coefficient should be.
[0096] Furthermore, the step of determining the adjustment coefficient for each reference patient based on the sample specificity and sample similarity includes:
[0097] Calculate the sum of sample specificity and sample similarity for each reference patient to obtain the initial coefficient value for each reference patient;
[0098] The initial coefficient values for each reference patient were normalized to obtain the adjusted coefficients for each reference patient.
[0099] The step of normalizing the initial coefficient values for each reference patient to obtain the adjusted coefficients for each reference patient includes:
[0100] Based on the training batch corresponding to each reference patient, the initial coefficient value of each reference patient is normalized to obtain the adjustment coefficient of each reference patient.
[0101] Each training batch corresponds to at least two reference patients. Different training batches have the same number of reference patients, but the reference patients in different training batches are completely different. The sum of the adjustment coefficients for all reference patients in the same training batch is 1.
[0102] Based on the above settings, the prediction loss (the difference between the predicted value and the ground truth label) of at least two training samples in a training batch is calculated by weighting. This amplifies the proportion of loss calculation for training samples with higher importance when participating in model training, while adaptively reducing the proportion of loss calculation for training samples with lower importance when participating in model training. This avoids model overfitting and helps the model better learn the mapping relationship between the multi-image features of patient biopsy samples under multiple staining methods and the probability of patient cure, thereby improving the prediction accuracy of the final trained prediction model.
[0103] Specifically, based on the training batch corresponding to each reference patient, the initial coefficient value of each reference patient is normalized to obtain the adjusted coefficient for each reference patient as follows:
[0104] Calculate the sum of the initial coefficient values of at least two reference patients corresponding to each training batch to obtain the cumulative coefficient value for each training batch;
[0105] The ratio of the initial coefficient value of each reference patient to the cumulative coefficient value of its corresponding training batch is calculated to obtain the adjustment coefficient of each reference patient.
[0106] In this invention, the termination condition for model training is: the loss function value of a certain training batch is less than a set loss threshold and / or the number of training batches processed is greater than a set batch number threshold.
[0107] Specifically, the loss function value of a training batch is the weighted sum of the predicted losses of at least two training samples in the training batch, where the weights for the predicted losses of the training samples are their adjustment coefficients.
[0108] In application, the above-mentioned multiple training batches can be obtained by completely random sampling of multiple training data. For example, the number of patients corresponding to a single training batch can be set to 100 based on experience.
[0109] It should be added that, in practical applications, when the number of training samples is less than a set threshold (such as 1000), data augmentation measures can be used to expand the number of existing training samples to reduce the risk of model overfitting.
[0110] In some embodiments, the prediction module is further configured to: fuse and output the prediction information with the medical history data of a key reference patient, wherein the key reference patient is the reference patient with the highest initial coefficient value among multiple reference patients. The medical history data includes stained biopsy images of the key reference patient under multiple staining methods, symptom description text, and treatment plans.
[0111] In this implementation, by fusing and outputting predictive information with the medical history data of key reference patients, the system not only indicates the probability of the target patient being cured, but also provides doctors with high-value cases that can assist in diagnosis (the specific subtype and severity of the target patient's lupus nephritis). These high-value cases not only corroborate the credibility of the output predictive information, but also help doctors break possible preconceived notions (such as the possibility that recently treated patients may interfere with the diagnosis of the current target patient), thus assisting doctors in better assessing the specific subtype and severity of the target patient's lupus nephritis.
[0112] In practical applications, when the predictive information indicates a low probability of a target patient being cured, it can remind doctors to conduct a more detailed diagnosis of lupus nephritis.
[0113] Furthermore, when predictive information indicates a high probability of the target patient being cured, and the critical reference patient's condition is similar to that of the target patient, the critical reference patient's medical history data can effectively reduce the complexity for doctors in developing subsequent treatment plans and improve diagnostic efficiency while ensuring diagnostic accuracy.
[0114] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above.
[0115] In one embodiment, the present invention provides a method for predicting the treatment efficacy of lupus nephritis based on multi-stain renal biopsy pathological images, such as... Figure 2 As shown, the method includes:
[0116] Step S1: In the patient set of lupus nephritis, feature extraction is performed on the stained biopsy images of each patient under each staining method to obtain multiple image features for each patient.
[0117] The patient set includes a target patient and multiple historical patients. Multiple image features of a patient correspond one-to-one with multiple staining methods. The stained biopsy image is an image acquired after the biopsy sample of the corresponding patient has been processed by the corresponding staining method.
[0118] Step S2: Perform similarity analysis based on multiple image features of the target patient and multiple image features of each historical patient to distinguish reference patients from non-reference patients among multiple historical patients.
[0119] Among them, the sample similarity of the reference patient was greater than that of the non-reference patient. The sample similarity was used to represent the degree of similarity between biopsy samples of historical patients and biopsy samples of the target patient.
[0120] Step S3: Fuse multiple image features of the same reference patient and their corresponding patient labels to obtain multiple training data.
[0121] The patient tag is used to indicate whether the corresponding reference patient has been cured.
[0122] Step S4: Train the model based on multiple training data to obtain a prediction model, and process multiple image features of the target patient based on the prediction model to obtain prediction information.
[0123] The prediction information is used to indicate the probability of the target patient being cured. The loss function of the prediction model includes an adjustment coefficient, which is used to adjust the prediction loss of the corresponding training data. The adjustment coefficient is determined based on the sample similarity and sample specificity of the corresponding reference patient. The sample specificity is used to indicate the specificity of the biopsy sample of the corresponding reference patient among multiple reference patients.
[0124] Furthermore, the lupus nephritis efficacy prediction system based on multi-stain renal biopsy pathological images provided in the above embodiments and the lupus nephritis efficacy prediction method based on multi-stain renal biopsy pathological images belong to the same concept. The specific implementation process is detailed in the system embodiments and will not be repeated here.
[0125] This invention also provides an electronic device. Please refer to [link to relevant documentation]. Figure 3 The electronic device may include a processor 301, a memory 302, and a program 3021 stored in the memory 302 and capable of running on the processor 301.
[0126] When program 3021 is executed by processor 301, it can achieve the following: Figure 2 Any steps in the corresponding method embodiments and the achievement of the same beneficial effects will not be repeated here.
[0127] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.
[0128] This invention also provides a readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described functions. Figure 2 Any step in the corresponding method embodiment can achieve the same technical effect, and will not be repeated here to avoid repetition.
[0129] The computer-readable storage medium of this invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0130] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.
[0131] The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0132] Computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or terminal. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0133] This invention also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to achieve the method for predicting the efficacy of lupus nephritis based on multi-stain renal biopsy pathological images provided in the above embodiments.
[0134] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0135] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A system for predicting the treatment efficacy of lupus nephritis based on multi-stain renal biopsy pathological images, characterized in that, The system includes: The feature extraction module is used to extract features from the stained biopsy images of each patient in the lupus nephritis patient set under each staining method, so as to obtain multiple image features of each patient. The patient set includes a target patient and multiple historical patients. The multiple image features of a patient correspond one-to-one with multiple staining methods. The stained biopsy image is the image acquired after the biopsy sample of the corresponding patient has been processed by the corresponding staining method. The similarity analysis module is used to perform similarity analysis between multiple image features of the target patient and multiple image features of each historical patient, so as to distinguish reference patients from non-reference patients from multiple historical patients. The sample similarity of the reference patient is greater than that of the non-reference patient. The sample similarity is used to represent the degree of similarity between the biopsy samples of the historical patients and the biopsy samples of the target patient. The data preparation module is used to fuse multiple image features of the same reference patient and their corresponding patient labels to obtain multiple training data, wherein the patient labels are used to indicate whether the corresponding reference patient has been cured. A prediction module is used to train a model based on multiple training data to obtain a prediction model, and to process multiple image features of the target patient based on the prediction model to obtain prediction information, wherein the prediction information is used to indicate the probability of the target patient being cured. The loss function of the prediction model includes an adjustment coefficient, which is used to adjust the prediction loss of the corresponding training data. The adjustment coefficient is determined based on the sample similarity and sample specificity of the corresponding reference patient. The sample specificity is used to indicate the specificity of the biopsy sample of the corresponding reference patient among multiple reference patients. The step of obtaining the adjustment coefficient includes: In multiple reference patients, the differences between each reference patient and other reference patients under each staining mode are analyzed to obtain multiple feature difference values for each reference patient. The multiple feature difference values of a reference patient correspond one-to-one with multiple staining modes. Analyze the central tendency of multiple characteristic differences for each reference patient to obtain the sample specificity for each reference patient; The adjustment coefficient for each reference patient is determined based on the sample specificity and sample similarity of each reference patient. The step of analyzing the differences between each reference patient and other reference patients under each staining modality, and obtaining multiple feature difference values for each reference patient, includes: In multiple reference patients, the outlier degree of each image feature of each reference patient in the corresponding feature set is analyzed to obtain multiple feature difference values for each reference patient. The feature set consists of multiple image features of multiple reference patients under the corresponding staining mode.
2. The lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images according to claim 1, characterized in that, The step of performing similarity analysis based on multiple image features of the target patient and multiple image features of each historical patient to distinguish reference patients from non-reference patients from multiple historical patients includes: In multiple historical patients, the similarity between different image features of each historical patient and the target patient under each staining method is analyzed to obtain multiple feature similarity values of each historical patient. The multiple feature similarity values of a historical patient correspond one-to-one with multiple staining methods. Analyze the central tendency of multiple feature similarity values for each historical patient to obtain the sample similarity for each historical patient; Based on the sample similarity of each historical patient, reference patients and non-reference patients are distinguished from multiple historical patients.
3. The lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images according to claim 1, characterized in that, The sample specificity is positively correlated with the corresponding adjustment coefficient, and the sample similarity is positively correlated with the corresponding adjustment coefficient.
4. The lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images according to claim 3, characterized in that, The steps for determining the adjustment coefficient for each reference patient based on sample specificity and sample similarity include: Calculate the sum of sample specificity and sample similarity for each reference patient to obtain the initial coefficient value for each reference patient; The initial coefficient values for each reference patient were normalized to obtain the adjusted coefficients for each reference patient.
5. The lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images according to claim 4, characterized in that, The steps for normalizing the initial coefficient values for each reference patient to obtain the adjusted coefficients for each reference patient include: Based on the training batch corresponding to each reference patient, the initial coefficient value of each reference patient is normalized to obtain the adjustment coefficient of each reference patient. Each training batch corresponds to at least two reference patients. Different training batches have the same number of reference patients, but the reference patients in different training batches are completely different. The sum of the adjustment coefficients for all reference patients in the same training batch is 1.
6. The lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images according to claim 1, characterized in that, In a patient dataset with lupus nephritis, the steps for feature extraction of stained biopsy images for each patient under each staining modality to obtain multiple image features for each patient include: In a patient set with lupus nephritis, corner detection was performed on the stained biopsy images of each patient under each staining method to obtain the corner information corresponding to each patient under each staining method. The corner point information of each patient under multiple staining methods is vectorized to obtain multiple image features for each patient.
7. The lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images according to claim 4, characterized in that, The prediction module is also used to: fuse the prediction information with the medical history data of a key reference patient and output the result, wherein the key reference patient is the reference patient with the highest initial coefficient value among multiple reference patients.
8. The lupus nephritis treatment efficacy prediction system based on multi-stain renal biopsy pathological images according to claim 7, characterized in that, The medical history data includes key reference patient stained biopsy images under multiple staining methods, symptom description text, and treatment plans.