A Multi-Method Fusion Approach for Precise Nuclei Segmentation in Prostate Histopathological Images for Digital Pathology-Assisted Diagnosis

This study designs a processing method for nuclear segmentation to provide a reliable basis for Gleason classification and quantitative analysis in digital pathology-assisted diagnosis. Aiming at the complex tissue structure, obvious nuclear overlap and adhesion, and difficult boundary identification in prostate histopathological images, an unsupervised segmentation framework based on multi-method fusion is constructed. First, fuzzy c-means clustering was used for preliminary region division to alleviate the influence of blurred nuclear boundaries. Then, Lab color space K-means and adaptive thresholding were introduced to further separate dark-stained areas and improve nuclear overlap and adhesion. Finally, Canny edge detection and morphological operations refined the segmentation results and located the centroid. Experiments on the public SICAPv2 dataset show this method can stably extract the nuclear region in different Gleason patterns. The positive predictive value was 89.94%, the intersection over union (IoU) was 75.13%, and the Dice similarity coefficient was 81.02%. The proposed method maintains high segmentation accuracy while controlling computational complexity and annotation cost, serving as a feasible solution for a basic image analysis module in digital pathology systems.