Research And Implementation Of Nuclear Segmentation Algorithm For H&E Stained Cancer Tissue Pathological Image

Cancer is currently one of the leading killers of human health,with more than five million people dying from cancer every year worldwide.Pathological images of cancer tissues based on H&E(hematoxylin-eosin staining)staining are the main basis for pathologists to diagnose cancer.The main reference indicators are the size,morphology,and distribution of cancer cells.Helps pathologists diagnose early and late stages of cancer and their severity.Because H&E stained pathological images often have problems such as blurred staining,uneven staining,border adhesion between nuclei,and large differences in nuclear morphology of different cancer species,efficient nuclear segmentation algorithms can significantly help pathologists reduce the diagnostic process Subjectivity,reducing the error rate.This paper mainly researches and designs a set of nuclear segmentation algorithm for H&E staining pathological images of cancer tissues.This paper is based on a full convolutional neural network.After analyzing some mainstream methods,this paper deeply studies the image colorization normalization technology,image segmentation deep learning method,and image denoising technology.The dyeing normalization preprocessing method is used to reduce the subjective problems introduced in the dyeing process.The pre-processed pathological pictures are input to a full convolutional neural network with an encoder-decoder structure called Refined Residual feature network(RRFNet).By introducing a residual network into the encoder,the underlying features of the extracted images(nucleus edges,nuclear morphology,etc.)are refined,and a connection layer with an attention mechanism called Pass feature after squeeze and excited(PSF)is designed to selectively pass the image information extracted by the encoder to the decoder.In addition,the overall information is refined at the decoder called Feature Refine(FR)to obtain a more accurate nuclear edge segmentation.The post-processing part,combined with watershed algorithm,image morphology operation,etc.,further denoises the segmentation result,and finally obtains accurate nuclear segmentation result.The experimental results of the paper are analyzed experimentally on two internationally publicly available datasets,Monuseg and breast cancer cell nuclear segmentation datasets.The algorithm of the paper has obtained more advanced experimental results.In addition,this paper use MFC and Opencv to complete the software development,convenient algorithm display and application of the algorithm.