Research And Implementation Of Renal Staining Transfer System Based On Cycle-Consistent Adversarial Networks

With the increasing number of patients suffering from end-stage renal disease,the demand for kidney transplantation as a treatment method has risen.However,there is always a shortage of qualified donor kidneys,which poses a challenge to the transplantation process.To effectively alleviate the shortage of donor kidneys,various transplant institutions have started to use expanded criteria donor(ECD).However,the use of ECDs comes with problems such as acute and chronic damage to kidney tissue,and their survival rate is lower compared to standard criteria donors(SCD).Therefore,it is crucial to have a fast and accurate donor kidney health scoring method to judge whether an extended ECD is healthy.The health of a standard donor kidney is usually judged using the Maryland scoring method,which requires scoring images of frozen sections stained and paraffin sections Masson-stained kidney sections.Although HE-stained sections were obtained quickly,the degree of kidney damage was inaccurate,while Masson-stained sections were more accurate,but the acquisition time was more than24 hours.For health judgment of the expanded standard donor kidney,the use of the Maryland scoring method results in prolonged pathological judgment and impaired ECD while awaiting the scoring result,preventing kidney transplantation as a qualified donor kidney.However,HE HE-staineddney sections can be quickly converted to Masson-stained kidney sections using kidney staining migration technology,making it possible to quickly and accurately determine ECD health using the Maryland scoring method.However,obtaining paired datasets of HE-stained and Masson-stained kidney images can be challenging,and obtaining qualified Masson-stained kidney images after migration using traditional deep-learning methods can also be difficult.The emergence of cycle-consistent adversarial networks has made it possible to transfer differently stained pathological images without the need for paired datasets.However,due to the lack of effective supervision,cycle-consistent adversarial networks still have the drawback of low staining migration accuracy in pathological images.As a result,the image quality of Masson-stained kidney sections after migration using cyclic generative adversarial networks can be low.And the generator does not consider the color correlation between images,resulting in some color differences between the small images after migration.It further affects the staining accuracy of the Masson-stained large image after migration obtained by splicing,and makes doctors make mistakes in judging the health of the ECD.Given the above shortcomings,this thesis studies and realizes the kidney image staining migration system based on cycle-consistent adversarial networks.It mainly includes the following aspects of work:1.An image preprocessing method for kidney images was developed.The method includes four steps: large image cropping,image enhancement using the HER staining enhancement method,acquisition of semantic segmentation images,and format modification of small images in the test set before slicing.First,the method crops the large kidney image into smaller images.Next,HE-stained kidney images are enhanced using the HER staining enhancement method,and the Masson-stained small images are subjected to image rotation transformation.Then,the method obtains the semantic segmentation map of each small image.Lastly,the format of the test set images is modified before splicing.2.A kidney image staining migration algorithm(SPCGan)is presented based on cycle-consistent adversarial networks.The algorithm modifies the normalization layer of the generator in the cyclic generation adversarial network into the spatial adaptive normalization layer and makes the image features not lose the semantic information of the image during the normalization process through the small graph semantic segmentation map obtained by data preprocessing.At the same time,the color correlation between input images is strengthened.To verify the effectiveness of the SPCGan algorithm,comparative experiments were performed on three datasets from different hospitals.Experimental results show that the SPCGan algorithm performs better than the classical algorithm.After migration,the FID scores of Masson-stained small plots were reduced by 1.87,2.14,and 0.42 compared with the classical algorithm Stain Gan in the same field.After migration,the absolute difference between the area of collagen fibers in Masson’s stained large map and the real large map was the smallest,reaching 0.045,which was 0.024 lower than that of the classical algorithm Stain Gan in the same field.It shows the staining accuracy and effectiveness of the kidney image staining transfer algorithm based on the recurrent generative adversarial network in the kidney image transfer task.3.The methods of stitching small images and visualization of large images after migration are given.Pathologists cannot judge the health of ECD through the Masson-stained small map after migration,so it is necessary to give a large image visualization method to make it more convenient for pathologists to observe the large picture of the Masson-stained kidney after migration and make correct judgments about the health of the ECD.The method of stitching and visualization of small images after migration is to stitch the small pictures of the migrated kidneys into a large image in Deep Zoom format,and then visualize them through the Openseadragon framework on the front end.4.Implement and run a test kidney dye migration system.Implement and run tests on the functions of each part of the system.This system’s running test results show that it can more quickly and accurately complete the kidney image staining migration task,obtain the Masson stained kidney slice image after migration,and visualize it on the web page,helping pathologists better judge the health of the kidney.Experiments show that the system can improve kidney image migration tasks’ staining accuracy and effectiveness.Through the method research and systematic implementation of this thesis,Masson-stained kidney images can be quickly obtained,which can help pathologists quickly judge the health of the kidney,reduce the discarding rate of expanded standard donors,and allow more patients with end-stage renal disease to undergo kidney transplantation.