Application Of Deep Learning Algorithm In Cell Segmentation And Faded Image Restoration Of Breast Pathology Image

Objective: Computer-aided diagnosis system based on deep learning is one of the research hotspots and difficulties in the field of artificial intelligence,and deep learning algorithms have made great progress in the analysis of medical images.But the efficient and accurate intelligent diagnosis system for breast cancer pathology is not complete,which is necessary to conduct deeper research.In this study,the practical application value of deep learning algorithm in breast pathological diagnosis was discussed by applying U-Net++ neural network algorithm to achieve cell segmentation of breast pathological images.And a color transfer algorithm based on digital pathological images was studied to realize the protective color repair of faded breast HE(hematoxylin-eosin)sections.Methods:(1)From 2019 to 2021,60 patients with invasive breast cancer and 60 patients with adenosis were screened from the Department of Pathology of Qingdao Central Hospital.The postoperative HE staining pathological sections were obtained and scanned with a digital scanner at 400× magnification to establish a digital pathological image dataset.Filter ROI region in the image,label malignant and benign breast cells in the image,obtain 28884 malignant labeled cells and 11814 benign labeled cells,and establish labeled image dataset.Then train the neural network framework,and use U-Net++ algorithm to realize the segmentation of benign and malignant cells.Accuracy,sensitivity,specificity,positive predictive value,negative predictive value,Dice coefficient and AUC(Area Under Curve)were used to evaluate the algorithm capability.(2)A total of 20 faded HE stained sections of breast cancer,which obtained from Qingdao Central Medical Group since 2014 to 2016,were selected as rendered images(providing shapes),and 5 HE stained paraffin sections of breast cancer in 2022 were selected to provide colors.After digital scanning,color transfer algorithm is applied to repair the color of faded image.SPSS 26.0 software was used for statistical analysis.The color of the restored image is evaluated according to the basic standard of HE film quality,and the quality of the restored image is evaluated using the image quality evaluation indicators NIQE(Natural Image Quality Evaluator),Entropy,and AG(Average Gradient).Results:(1)Use U-Net++ algorithm to realize the segmentation of benign and malignant cells.Malignant cell set training results: accuracy,sensitivity,specificity,positive predictive value,negative predictive value,DICE and AUC were respectively 0.92,0.73,0.95,0.76,0.95,0.75,0.95,0.95,0.95,0.95.Benign cell set training results: accuracy,sensitivity,specificity,positive predictive value,negative predictive value,DICE and AUC were respectively 0.96,0.73,0.98,0.80,0.97,0.77,and 0.98.(2)The repair of the faded breast cancer HE stained sections was carried out by using the sub-regional color transfer algorithm.The image after repairing met the basic standards of HE film production quality;the tissue structure of the tumor area and the interstitial region under low magnification was outstanding;the structure of cells and nuclei under high magnification was clear;and the score was higher than that before repair(P<0.01),and the difference was statistically significant.After repairing,the NIQE value decreased(P<0.01),Entropy value increased(P<0.01),and AG value increased(P<0.01),and the differences were statistically significant.Conclusion: In this study,by establishing the breast digital pathological section dataset and the annotation dataset of benign and malignant cells of the breast,the U-net++ neural network model is trained to realize the segmentation of benign and malignant cells of the breast,which is helpful to reduce the reading work of pathologists and assist pathologists to draw objective and accurate conclusions,so as to achieve efficient and accurate diagnosis of breast cancer.By applying the sub-regional color transfer algorithm to repair the color of the faded breast cancer HE stained sections,the contrast between nucleus and cytoplasmic staining was significantly restored;the color of important structures such as nuclei was effectively preserved;the quality of images was improved;and its diagnostic and teaching value was restored.