Pathological image high-resolution reconstruction method based on deep learning
By constructing a deep learning model, preprocessing, training and optimization of pathological images, the problem of insufficient texture and detail recovery capabilities in high-resolution reconstruction of pathological images is solved, and a higher quality reconstruction effect is achieved.
Patent Information
- Application Number
- CN202411764111.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-04
- Publication Date
- 2025-04-22
AI Technical Summary
The prior art has problems with limited texture and detail recovery capabilities in high-resolution reconstruction of pathological images, resulting in blurred or distorted reconstruction results.
Using a deep learning method, a deep learning model of the generator network and discriminator network is constructed, and high-resolution reconstruction of pathological target images is achieved by preprocessing, training and optimizing the pathological images.
Improve the image quality and detail display capabilities of pathological images, and achieve more accurate high-resolution reconstruction, avoiding blur or distortion of reconstruction results.
Smart Images
Figure CN119863363A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing technology, and in particular to a high-resolution reconstruction method of pathological images based on deep learning. Background Art
[0002] Pathological images are high-resolution digital images acquired by scanning with a fully automatic microscope or optical magnification system, and are then stitched and processed by a computer with high precision and without gaps in multiple fields of view, thereby obtaining a multi-level visualized image, which usually has complex details and structures. At present, when processing pathological images, image processing techniques such as interpolation methods, frequency domain-based methods, and super-resolution reconstruction methods are usually used. However, these methods have certain limitations in the high-resolution reconstruction of pathological images, and have limited ability to restore image texture and details, which can easily lead to blurred or distorted reconstruction results. Therefore, the present invention proposes a high-resolution reconstruction method for pathological images based on deep learning. The deep learning-based algorithm learns the complex features in the pathological images, which can not only achieve more accurate high-resolution reconstruction, but also restore details based on the complex features in the pathological images, thereby improving the image quality and detail display capability of the pathological images. Summary of the invention
[0003] The purpose of the present invention is to provide a high-resolution reconstruction method for pathological images based on deep learning to solve the problems raised in the above-mentioned background technology.
[0004] To achieve the above object, the present invention provides the following technical solution: a high-resolution reconstruction method of pathological images based on deep learning, comprising:
[0005] Acquire a pathological image, and preprocess the pathological image to obtain a first processed pathological image; the first processed pathological image includes: a first processed pathological training image and a first processed pathological target image;
[0006] Build a deep learning model to obtain a deep learning model;
[0007] Using the first processed pathological training image to train the deep learning model to obtain deep learning model training data;
[0008] Performing training analysis based on the first processed pathological training image and the deep learning model training data to obtain a model training analysis result, and performing model optimization on the deep learning model according to the model training analysis result to obtain a deep learning optimization model;
[0009] The first processed pathological target image is input into the deep learning optimization model for high-resolution reconstruction to obtain a reconstructed image of the pathological target.
[0010] Furthermore, there are multiple pathological images, and after preprocessing the pathological images, the first processed pathological images are screened, the first processed pathological training images are screened into the training image set, and the first processed pathological target images are screened into the image set to be processed.
[0011] Furthermore, when there are a plurality of first processed pathology target images in the set of images to be processed, the first processed pathology target images are arranged in the set of images to be processed in the order of their acquisition time to form a queue of images to be processed.
[0012] Furthermore, the deep learning model is used to learn the features and texture information of the pathological image, and reconstruct the pathological image based on the features and texture information of the pathological image to obtain a high-resolution pathological image, wherein the deep learning model includes: a generator network and a discriminator network. When determining the generator network and the discriminator network, the number of layers and the number of nodes of the generator network and the discriminator network are determined according to the requirement standards of the pathological image, thereby determining the deep learning model.
[0013] Further, training analysis is performed based on the first processed pathological training image and the deep learning model training data, including:
[0014] Determine loss analysis parameters, and select parameter loss functions based on the loss analysis parameters;
[0015] Performing model loss analysis based on the parameter loss function in combination with the first processed pathological training image and the deep learning model training data to obtain parameter loss data;
[0016] Analyze the parameter loss data to determine the model training analysis results.
[0017] Furthermore, the deep learning model is optimized based on the model training analysis results, including:
[0018] Adjust the model parameters based on the parameter loss data in the model training analysis results to obtain an adjusted deep learning model;
[0019] Retrain the adjusted deep learning model, obtain the adjusted deep learning model training data, and obtain the adjusted parameter loss data;
[0020] Perform performance analysis on the deep learning model using cross-validation based on the parameter loss data and the adjusted parameter loss data to determine the optimal model parameter data;
[0021] The deep learning model is optimized according to the optimal data of model parameters to obtain a deep learning optimization model.
[0022] Furthermore, the number of times the model parameters are adjusted is not unique. When adjusting the model parameters, the parameter adjustment interval is determined, and the single variable control method is used to adjust the model parameters of the deep learning model in a positive or negative direction according to the parameter adjustment interval to obtain multiple adjusted deep learning models, and then the multiple adjusted deep learning models are retrained to obtain multiple groups of adjusted deep learning model training data.
[0023] Furthermore, after obtaining the pathological target reconstructed image, the pathological target reconstructed image is evaluated, including:
[0024] Acquiring image attribute data for the pathological target reconstructed image to obtain the pathological target reconstructed image attribute data;
[0025] Perform image quality index analysis based on the image attribute data reconstructed according to the pathological target to obtain image quality index data;
[0026] The image quality index data is integrated to obtain the pathological target reconstructed image evaluation data, and the pathological target reconstructed image evaluation result is obtained according to the pathological target reconstructed image evaluation data.
[0027] Furthermore, when performing image quality index analysis based on the attribute data of the pathological target reconstructed image, the pathological target reconstructed image is compared with the original high-resolution image, including:
[0028] Determine the original high-resolution image and use the original high-resolution image as a reference standard for comparative analysis;
[0029] Obtain original high-resolution image attribute data;
[0030] According to the pathological target reconstructed image attribute data and the original high-resolution image attribute data, effective data extraction is performed according to the image quality index, and the image quality index is calculated based on the extraction results to obtain the image quality index data.
[0031] Further, after obtaining the evaluation result of the pathological target reconstructed image, when the evaluation result of the pathological target reconstructed image is unqualified, the pathological target reconstructed image is corrected according to the evaluation result of the pathological target reconstructed image, including:
[0032] In the process of high-resolution reconstruction of the first processed pathological target image corresponding to the pathological target reconstructed image by the deep learning optimization model, features and texture information of the pathological image are retrieved;
[0033] The pathological target reconstructed image is disassembled according to the image information to obtain a plurality of pathological target reconstructed image area blocks;
[0034] Screening a plurality of pathological target reconstructed image region blocks to be corrected to determine a pathological target reconstructed image region block to be corrected;
[0035] Correction and restoration are performed on the reconstructed image region block of the pathological target to be corrected according to the characteristics and texture information of the pathological image, so as to obtain the corrected reconstructed image region block of the pathological target;
[0036] A corrected pathological target reconstructed image is obtained based on the corrected pathological target reconstructed image region block.
[0037] The present invention constructs a deep learning model so that high-resolution reconstruction of pathological target images can be achieved when processing pathological images, thereby improving the resolution of pathological target images. The deep learning model is also optimized before high-resolution reconstruction of pathological target images, thereby improving the accuracy of the deep learning model, so that the deep learning optimization model can better achieve more accurate high-resolution reconstruction of pathological images, and at the same time can avoid blurring or distortion of the reconstructed image of the pathological target, thereby ensuring the image quality and detail display capability of the reconstructed image of the pathological target.
[0038] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the application documents.
[0039] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0041] Figure 1 A schematic diagram of the steps of the method for high-resolution reconstruction of pathological images according to the present invention;
[0042] Figure 2 It is a schematic diagram of some steps of step 4 in the method for high-resolution reconstruction of pathological images of the present invention;
[0043] Figure 3 It is another partial step schematic diagram of step 4 in the method for high-resolution reconstruction of pathological images of the present invention;
[0044] Figure 4 A schematic diagram of the steps of evaluating the reconstructed image of the pathological target in the high-resolution reconstruction method of the pathological image of the present invention;
[0045] Figure 5 A schematic diagram of the steps of step C2 for evaluating the reconstructed image of the pathological target in the high-resolution reconstruction method of the pathological image of the present invention;
[0046] Figure 6 It is a schematic diagram of the steps of correcting the pathological target reconstructed image according to the evaluation result of the pathological target reconstructed image in the high-resolution pathological image reconstruction method of the present invention. DETAILED DESCRIPTION
[0047] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.
[0048] like Figure 1 As shown, an embodiment of the present invention provides a method for high-resolution reconstruction of pathological images based on deep learning, comprising:
[0049] Step 1: Acquire a pathological image, and pre-process the pathological image to obtain a first processed pathological image; the first processed pathological image includes: a first processed pathological training image and a first processed pathological target image;
[0050] Step 2: Build a deep learning model to obtain a deep learning model;
[0051] Step 3: Use the first processed pathological training image to train the deep learning model to obtain deep learning model training data;
[0052] Step 4: Perform training analysis based on the first processed pathological training image and the deep learning model training data to obtain a model training analysis result, and optimize the deep learning model according to the model training analysis result to obtain a deep learning optimization model;
[0053] Step 5: Input the first processed pathological target image into the deep learning optimization model for high-resolution reconstruction to obtain a reconstructed image of the pathological target.
[0054] In the above technical solution, the obtained pathological images are divided into pathological training images and pathological target images according to the image usage, wherein the pathological target images are the pathological images that need to be processed.
[0055] In the above technical solution, the preprocessing of the pathological image includes: image denoising, normalization and other operations.
[0056] In the above technical solution, the deep learning model is used to learn the complex features in the pathological images.
[0057] In the above technical solution, when constructing a deep learning model, a suitable resolution is selected to reconstruct the deep learning model.
[0058] In the above technical solution, the number of the first processed pathology training images is multiple.
[0059] In the above technical solution, when optimizing the deep learning model according to the model training analysis results, it includes: adjusting the model parameters, the number of layers and nodes of the network structure, etc.
[0060] The above technical solution constructs a deep learning model so that when processing pathological images, high-resolution reconstruction of pathological target images can be achieved, thereby improving the resolution of pathological target images, and then improving the image quality and detail display capability of pathological images, so that medical staff can more accurately analyze pathological images or extract key information from pathological images. In addition, the first processed pathological target image is input into the deep learning optimization model for high-resolution reconstruction. The deep learning optimization model is used for processing, which can not only process a large number of pathological images and improve the utilization efficiency of data, but also identify patterns or laws in pathological images, so that the pathological target reconstructed image is more comprehensive and accurate. The deep learning model is also optimized before the high-resolution reconstruction of the pathological target image, so as to improve the accuracy of the deep learning model, so that the deep learning optimization model can better achieve more accurate high-resolution reconstruction for pathological images, and at the same time, it can also avoid blurring or distortion of the pathological target reconstructed image, and ensure the image quality and detail display capability of the pathological target reconstructed image, so that medical staff can more accurately analyze pathological images, obtain more and more accurate information from pathological images, and provide information support for patient condition analysis.
[0061] In one embodiment provided by the present invention, the number of the pathological images is multiple, and after preprocessing the pathological images, the first processed pathological images are screened, the first processed pathological training images are screened into the training image set, and the first processed pathological target images are screened into the image set to be processed.
[0062] In the above technical solution, when acquiring pathological images, both pathological training images used for training deep learning models and pathological target images reconstructed with high resolution through deep learning models can be acquired.
[0063] In the above technical solution, the number of the first processed pathology training images in the training image set is multiple.
[0064] In the above technical solution, when the first processed pathological images are screened after the pathological images are pre-processed, the screening is performed according to the purpose of the pathological images.
[0065] In the above technical solution, the training image set and the to-be-processed image set are stored for the first processed pathology training image and the first processed pathology target image respectively.
[0066] The above technical solution realizes the division of pathological images through a training image set and a to-be-processed image set, so that after obtaining the preprocessed pathological images, the pathological images to be trained can be distinguished from the target pathological images to be processed, avoiding confusion when there are many pathological images.
[0067] In an embodiment provided by the present invention, when there are multiple first processed pathology target images in the to-be-processed image set, the first processed pathology target images are arranged in the to-be-processed image set in the order of their acquisition time to form an to-be-processed image queue.
[0068] In the above technical solution, the number of first processed pathology target images in the image set to be processed may also be one. When there is only one first processed pathology target image in the image set to be processed, there is no need to perform time analysis on the first processed pathology target image to form a queue.
[0069] The above technical solution forms a queue of images to be processed in the set of images to be processed according to the chronological order of acquisition of the first processed pathological target images, so that when there are multiple first processed pathological target images, high-resolution reconstruction can be performed in sequence through the deep learning model, thereby avoiding the medical staff providing the pathological images waiting for too long without receiving a response from the deep learning module for the first processed pathological target images corresponding to the pathological images, thereby improving the experience of the medical staff.
[0070] In one embodiment provided by the present invention, the deep learning model is used to learn the features and texture information of pathological images, and reconstruct the pathological images based on the features and texture information of the pathological images to obtain high-resolution pathological images, wherein the deep learning model includes: a generator network and a discriminator network. When determining the generator network and the discriminator network, the number of layers and the number of nodes of the generator network and the discriminator network are determined according to the requirement standards of the pathological images, thereby determining the deep learning model.
[0071] In the above technical solution, the deep learning model is a model based on a generative adversarial network (GAN), which is a deep learning model composed of two neural networks.
[0072] In the above technical solution, the generator network learns to generate samples similar to real data, and the discriminator network determines whether a sample is real data or forged data generated by the generator. The two networks compete with each other, and through repeated iterations, the generator's ability to generate realistic samples is improved, while the discriminator network is used to more accurately distinguish between real and forged samples.
[0073] In the above technical solution, when determining the number of layers and nodes of the generator network and the discriminator network according to the required standards of the pathological image, the initial number of layers and the initial number of nodes of the generator network and the discriminator network are determined, and the initial values of the hyperparameters are also determined.
[0074] The above technical solution constructs a deep learning model so that the deep learning model can reconstruct the first processed pathological target image with high resolution. It can not only learn the complex features in the first processed pathological target image and enhance the visibility of microstructures and cytological features, but also improve the resolution of the first processed pathological target image, which helps to analyze pathological images more comprehensively and accurately and promote medical research by medical staff.
[0075] like Figure 2 As shown, in one embodiment provided by the present invention, training analysis is performed based on the first processed pathological training image and the deep learning model training data, including:
[0076] A1. Determine loss analysis parameters and select parameter loss function based on the loss analysis parameters;
[0077] A2. Perform model loss analysis based on the parameter loss function in combination with the first processed pathological training image and the deep learning model training data to obtain parameter loss data;
[0078] A3. Analyze the parameter loss data and determine the model training analysis results.
[0079] In the above technical solution, the loss analysis parameters include: pixels, perception, etc., and the parameter loss function includes pixel-level loss, perceptual loss, etc.
[0080] In the above technical solution, when analyzing parameter loss data, numerical value analysis is performed on the loss data of the same parameter.
[0081] The above technical solution guides the training of the deep learning model through the parameter loss function, so that the deep learning model can have a smaller loss error and improve the accuracy of the deep learning model.
[0082] like Figure 3 As shown, in one embodiment provided by the present invention, model optimization is performed on a deep learning model according to the model training analysis result, including:
[0083] B1. Adjust the model parameters based on the parameter loss data in the model training analysis results to obtain the adjusted deep learning model;
[0084] B2. Retrain the adjusted deep learning model, obtain the adjusted deep learning model training data, and obtain the adjusted parameter loss data;
[0085] B3. Perform performance analysis on the deep learning model using cross-validation based on the parameter loss data and the adjusted parameter loss data to determine the optimal model parameter data;
[0086] B4. Optimize the deep learning model according to the optimal data of model parameters to obtain a deep learning optimization model.
[0087] In the above technical solution, adjusting the model parameters includes: adjusting hyperparameters, the number of layers and nodes of the network structure, etc.
[0088] In the above technical solution, when performing performance analysis on the deep learning model, cross-validation or other evaluation indicators can also be used for performance analysis.
[0089] In the above technical solution, when performing performance analysis on the deep learning model, the optimal parameter loss data is determined, and based on the optimal parameter loss data, the data of the model parameters in the corresponding state is obtained, and the model parameters in this state are used as the optimal model parameter data.
[0090] The above technical solution optimizes the deep learning model according to the model training analysis results, so that the model parameters in the deep learning optimization model can cause smaller losses, thereby improving the effect of the deep learning optimization model in improving the quality and details of the image, ensuring the reconstruction effect of the pathological target reconstructed image, and reducing image distortion and information loss.
[0091] In one embodiment provided by the present invention, the number of times the model parameters are adjusted is not unique. When adjusting the model parameters, the parameter adjustment interval is determined, and the model parameters are positively or negatively adjusted for the deep learning model using a single variable control method according to the parameter adjustment interval to obtain multiple adjusted deep learning models, and then the multiple adjusted deep learning models are re-trained to obtain multiple groups of adjusted deep learning model training data.
[0092] In the above technical solution, the parameter adjustment interval can be set according to needs.
[0093] In the above technical solution, the parameter adjustment intervals of different parameters may be the same or different.
[0094] In the above technical solution, each adjusted deep learning model can acquire training data by performing high-resolution reconstruction on the first processed pathology training image.
[0095] The above technical solution facilitates the adjustment of model parameters by determining the parameter adjustment interval, thereby improving the efficiency of model training, and adopts a single variable control method to adjust the model parameters in a positive or negative direction for the deep learning model according to the parameter adjustment interval, thereby ensuring the comprehensiveness of the model parameter adjustment and enabling the acquisition of more comprehensive training data, so that when the deep learning model is optimized according to the model training analysis results, the overall optimization of the deep learning model can be ensured, avoiding large deviations in the optimization effects of different parameters that lead to more defects in the reconstructed image of the pathological target, thereby ensuring the image quality and detail display capability of the reconstructed image of the pathological target.
[0096] like Figure 4 As shown, in one embodiment provided by the present invention, after obtaining the pathological target reconstructed image, the pathological target reconstructed image is also evaluated, including:
[0097] C1. Acquiring image attribute data for the pathological target reconstructed image to obtain the pathological target reconstructed image attribute data;
[0098] C2, performing image quality index analysis based on the image attribute data reconstructed according to the pathological target to obtain image quality index data;
[0099] C3. Obtain pathological target reconstructed image evaluation data by integrating the image quality index data, and obtain the pathological target reconstructed image evaluation result according to the pathological target reconstructed image evaluation data.
[0100] In the above technical solution, the image attributes include: image structure, pixels, resolution, saturation, etc.
[0101] In the above technical solution, the image quality indicators include: structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), etc.
[0102] In the above technical solution, when the image quality index data is integrated, the image quality index data is combined with the image quality index weight to obtain the pathological target reconstructed image evaluation data, wherein the image quality index weight is usually equal, but can also be unequal.
[0103] In the above technical solution, after the pathological target reconstructed image evaluation data is obtained, the pathological target reconstructed image evaluation data is combined with a preset judgment threshold to determine whether the pathological target reconstructed image is qualified, thereby obtaining the pathological target reconstructed image evaluation result. If the pathological target reconstructed image evaluation data is not less than the preset judgment threshold, the pathological target reconstructed image evaluation result is qualified; if the pathological target reconstructed image evaluation data is less than the preset judgment threshold, the pathological target reconstructed image evaluation result is unqualified.
[0104] The above technical solution evaluates the reconstructed image of the pathological target to clarify the image quality and detail display capability of the reconstructed image of the pathological target, thereby realizing the evaluation of the reconstructed image of the pathological target and ensuring the effect of obtaining the reconstructed image of the pathological target.
[0105] like Figure 5 As shown, in one embodiment provided by the present invention, when performing image quality index analysis based on the attribute data of the pathological target reconstructed image, the pathological target reconstructed image is compared with the original high-resolution image, including:
[0106] D1. Determine the original high-resolution image and use it as a reference standard for comparative analysis;
[0107] D2, obtaining original high-resolution image attribute data;
[0108] D3. Extract effective data according to the image quality index based on the reconstructed image attribute data of the pathological target and the original high-resolution image attribute data, and calculate the image quality index based on the extraction results to obtain the image quality index data.
[0109] In the above technical solution, the original high-resolution image is a high-resolution image among the acquired pathological images.
[0110] In the above technical solution, the image quality indicators include: structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), etc.
[0111] In the above technical scheme, when effective data is extracted according to the image quality index based on the pathological target reconstructed image attribute data and the original high-resolution image attribute data, if the image quality index is a structural similarity index, image structure data is extracted from the pathological target reconstructed image attribute data and the original high-resolution image attribute data, and the image structure data is used to calculate the structural similarity between the pathological target reconstructed image and the original high-resolution image; if the image quality index is a peak signal-to-noise ratio index, image peak data is extracted from the pathological target reconstructed image attribute data and the original high-resolution image attribute data, and the image peak data is used to calculate the peak signal-to-noise ratio between the pathological target reconstructed image and the original high-resolution image.
[0112] The above technical solution determines the original high-resolution image and uses the original high-resolution image as a reference standard for comparative analysis, thereby specifically comparing the pathological target reconstructed image with the original high-resolution image, measuring the accuracy and fidelity of the reconstruction result, and thus more intuitively clarifying the difference between the original high-resolution image and the pathological target reconstructed image. In addition, effective data extraction is performed according to the image quality index based on the attribute data of the pathological target reconstructed image and the attribute data of the original high-resolution image, so that when calculating the image quality index, only the image attribute data related to the image quality index is calculated, avoiding the redundancy of irrelevant data, thereby reducing the confusion of the image quality index calculation, and then reducing the probability of error in the image quality index calculation, thereby ensuring the accuracy of the image quality index data.
[0113] like Figure 6 As shown, in one embodiment provided by the present invention, after obtaining the pathological target reconstructed image evaluation result, when the pathological target reconstructed image evaluation result is unqualified, the pathological target reconstructed image is corrected according to the pathological target reconstructed image evaluation result, including:
[0114] S1. In the process of high-resolution reconstruction of the first processed pathological target image corresponding to the pathological target reconstructed image by the deep learning optimization model, extracting features and texture information of the pathological image;
[0115] S2, disassembling the pathological target reconstructed image according to the image information to obtain a plurality of pathological target reconstructed image area blocks;
[0116] S3, screening the pathological target reconstructed image region blocks to be corrected in the plurality of pathological target reconstructed image region blocks to be corrected, and determining the pathological target reconstructed image region blocks to be corrected;
[0117] S4, performing correction and restoration on the pathological target reconstructed image region block to be corrected according to the characteristics and texture information of the pathological image, to obtain a corrected pathological target reconstructed image region block;
[0118] S5. Obtain a corrected pathological target reconstructed image based on the corrected pathological target reconstructed image region block.
[0119] In the above technical solution, if the evaluation result of the pathological target reconstructed image is qualified, there is no need to correct the pathological target reconstructed image.
[0120] In the above technical solution, when the reconstructed image of the pathological target is disassembled according to the image information, the object image area and the background area are disassembled, and then the object image area is further disassembled into multiple area blocks according to the object image characteristics and distribution.
[0121] In the above technical scheme, when screening for correction among multiple pathological target reconstructed image area blocks, the pathological target reconstructed image area blocks containing only the background are filtered out to obtain a first screening result; in the first screening result, image content judgment is performed on the pathological target reconstructed image area blocks to determine whether target structures or target cells appear in the pathological target reconstructed image area blocks, and the pathological target reconstructed image area blocks where the target structures or target cells appear are screened out to obtain a second screening result; image quality index analysis is performed on the pathological target reconstructed image area blocks in the second screening result to determine the pathological target reconstructed image area blocks that need to be corrected, and obtain the pathological target reconstructed image area blocks to be corrected.
[0122] In the above technical solution, when obtaining a corrected pathological target reconstructed image based on the corrected pathological target reconstructed image area block, the corrected pathological target reconstructed image area block is combined and restored with the pathological target reconstructed image area block that is filtered out during the correction screening to obtain the corrected pathological target reconstructed image.
[0123] In the above technical scheme, when the correction and restoration is performed on the reconstructed image region block of the pathological target to be corrected according to the characteristics and texture information of the pathological image, it is divided into the characteristic correction of the pathological image and the texture information correction, and different correction steps are performed according to the number of the reconstructed image region blocks of the pathological target to be corrected. When the number of the reconstructed image region blocks of the pathological target to be corrected is unique, the characteristic enhancement is performed on the reconstructed image region blocks of the pathological target to be corrected according to the characteristics of the pathological image to obtain the first corrected reconstructed image region block of the pathological target, and then the texture target judgment is performed in the first corrected reconstructed image region block of the pathological target in combination with the texture information. If the target texture exists in the first corrected reconstructed image region block of the pathological target, the target texture correction is performed on the first corrected reconstructed image region block of the pathological target according to the texture information to obtain the corrected reconstructed image region block of the pathological target; when the number of the reconstructed image region blocks of the pathological target to be corrected is not unique, texture analysis is performed on the reconstructed image region blocks of the pathological target to be corrected to determine whether texture correction is required. The pathological target reconstructed image area block to be corrected is obtained to obtain a first correction set and a second correction set, wherein the first correction set includes all the pathological target reconstructed image area blocks to be corrected that need texture correction, and the second correction set includes all the pathological target reconstructed image area blocks to be corrected that do not need texture correction, then texture correction is performed on the first correction set according to the texture information to obtain the second corrected pathological target reconstructed image area block, and at the same time, feature enhancement is performed on the second correction set according to the characteristics of the pathological image to obtain the second partially corrected pathological target reconstructed image area block, and then after obtaining the second corrected pathological target reconstructed image area block, feature enhancement is performed on the second corrected pathological target reconstructed image area block using the characteristics of the pathological image to obtain the first partially corrected pathological target reconstructed image area block, thereby determining the corrected pathological target reconstructed image area block according to the first partially corrected pathological target reconstructed image area block and the second partially corrected pathological target reconstructed image area block.
[0124] The above technical solution ensures the accuracy and fidelity of the final pathological target reconstructed image by correcting the pathological target reconstructed image when the evaluation result of the pathological target reconstructed image is unqualified, and disassembles the pathological target reconstructed image according to the image information, so that when correcting the pathological target reconstructed image, only the pathological target reconstructed image area block to be corrected is corrected, thereby improving the correction efficiency, while also ensuring the image quality and detail display capability of the pathological target reconstructed image, thereby enabling medical staff to analyze the pathological image more accurately, obtain more and more accurate information from the pathological image, and provide information support for the patient's condition analysis.
[0125] Those skilled in the art should understand that the first and second in the present invention merely refer to different application stages.
[0126] Other embodiments of the present disclosure will be readily apparent to those skilled in the art after considering the specification and practicing the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0127] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A high-resolution reconstruction method for pathological images based on deep learning, characterized in that: include: Acquiring a pathological image, and preprocessing the pathological image to obtain a first processed pathological image; The first processed pathology image includes: a first processed pathology training image and a first processed pathology target image; Build a deep learning model to obtain a deep learning model; Using the first processed pathological training image to train the deep learning model to obtain deep learning model training data; Performing training analysis based on the first processed pathological training image and the deep learning model training data to obtain a model training analysis result, and performing model optimization on the deep learning model according to the model training analysis result to obtain a deep learning optimization model; The first processed pathological target image is input into the deep learning optimization model for high-resolution reconstruction to obtain a reconstructed image of the pathological target.
2. The high-resolution reconstruction method of pathological images according to claim 1, characterized in that: There are multiple pathological images. After preprocessing the pathological images, the first processed pathological images are screened, the first processed pathological training images are screened into the training image set, and the first processed pathological target images are screened into the to-be-processed image set.
3. The high-resolution reconstruction method of pathological images according to claim 2, characterized in that: When there are a plurality of first pathological target images in the set of images to be processed, the first pathological target images are arranged in the set of images to be processed in the order of their acquisition time to form a queue of images to be processed.
4. The high-resolution reconstruction method of pathological images according to claim 1, characterized in that: The deep learning model is used to learn the features and texture information of pathological images, and reconstruct the pathological images based on the features and texture information of the pathological images to obtain high-resolution pathological images. The deep learning model includes: a generator network and a discriminator network. When determining the generator network and the discriminator network, the number of layers and the number of nodes of the generator network and the discriminator network are determined according to the requirement standards of the pathological images, thereby determining the deep learning model.
5. The high-resolution reconstruction method of pathological images according to claim 1, characterized in that: Perform training analysis based on the first processed pathological training image and the deep learning model training data, including: Determine loss analysis parameters, and select parameter loss functions based on the loss analysis parameters; Performing model loss analysis based on the parameter loss function in combination with the first processed pathological training image and the deep learning model training data to obtain parameter loss data; Analyze the parameter loss data to determine the model training analysis results.
6. The method for high-resolution reconstruction of pathological images according to claim 4, characterized in that: Optimize the deep learning model based on the model training analysis results, including: Adjust the model parameters based on the parameter loss data in the model training analysis results to obtain an adjusted deep learning model; Retrain the adjusted deep learning model, obtain the adjusted deep learning model training data, and obtain the adjusted parameter loss data; Perform performance analysis on the deep learning model using cross-validation based on the parameter loss data and the adjusted parameter loss data to determine the optimal model parameter data; The deep learning model is optimized according to the optimal data of model parameters to obtain a deep learning optimization model.
7. The method for high-resolution reconstruction of pathological images according to claim 6, characterized in that: The number of times the model parameters are adjusted is not unique. When adjusting the model parameters, the parameter adjustment interval is determined, and the single variable control method is used to adjust the model parameters of the deep learning model in a positive or negative direction according to the parameter adjustment interval to obtain multiple adjusted deep learning models, and then the multiple adjusted deep learning models are retrained to obtain multiple groups of adjusted deep learning model training data.
8. The method for high-resolution reconstruction of pathological images according to claim 1, characterized in that: After obtaining the pathological target reconstructed image, the pathological target reconstructed image is also evaluated, including: Acquiring image attribute data for the pathological target reconstructed image to obtain the pathological target reconstructed image attribute data; Perform image quality index analysis based on the image attribute data reconstructed according to the pathological target to obtain image quality index data; The image quality index data is integrated to obtain the pathological target reconstructed image evaluation data, and the pathological target reconstructed image evaluation result is obtained according to the pathological target reconstructed image evaluation data.
9. The method for high-resolution reconstruction of pathological images according to claim 8, characterized in that: When performing image quality index analysis based on the attribute data of the pathological target reconstructed image, the pathological target reconstructed image is compared with the original high-resolution image, including: Determine the original high-resolution image and use the original high-resolution image as a reference standard for comparative analysis; Obtaining original high-resolution image attribute data; According to the pathological target reconstructed image attribute data and the original high-resolution image attribute data, effective data extraction is performed according to the image quality index, and the image quality index is calculated based on the extraction results to obtain the image quality index data.
10. The method for high-resolution reconstruction of pathological images according to claim 8, characterized in that: After obtaining the pathological target reconstructed image evaluation result, when the pathological target reconstructed image evaluation result is unqualified, the pathological target reconstructed image is corrected according to the pathological target reconstructed image evaluation result, including: In the process of high-resolution reconstruction of the first processed pathological target image corresponding to the pathological target reconstructed image by the deep learning optimization model, features and texture information of the pathological image are retrieved; The pathological target reconstructed image is disassembled according to the image information to obtain a plurality of pathological target reconstructed image area blocks; Screening a plurality of pathological target reconstructed image region blocks to be corrected to determine a pathological target reconstructed image region block to be corrected; Correction and restoration are performed on the reconstructed image region block of the pathological target to be corrected according to the characteristics and texture information of the pathological image, so as to obtain the corrected reconstructed image region block of the pathological target; A corrected pathological target reconstructed image is obtained based on the corrected pathological target reconstructed image region block.