2019/05/13 by Yu-Min Chung, Chung, Yu-Min, Chuan-Shen Hu +5
Computer Science · Medicine · #Topological and Geometric Data Analysis #Clusterin in disease pathology #Leprosy Research and Treatment
paper · pdf · doi:10.48550/arxiv.1905.08607
Skin cancer is one of the most common cancers in the United States. As\ntechnological advancements are made, algorithmic diagnosis of skin lesions is\nbecoming more important. In this paper, we develop algorithms for segmenting\nthe actual diseased area of skin in a given image of a skin lesion, and for\nclassifying different types of skin lesions pictured in a given image. The\ncores of the algorithms used were based in persistent homology, an algebraic\ntopology technique that is part of the rising field of Topological Data\nAnalysis (TDA). The segmentation algorithm utilizes a similar concept to\npersistent homology that captures the robustness of segmented regions. For\nclassification, we design two families of topological features from persistence\ndiagrams---which we refer to as em persistence statistics (PS) and em\npersistence curves (PC), and use linear support vector machine as classifiers.\nWe also combined those topological features, PS and PC, into ResNet-101 model,\nwhich we call em TopoResNet-101, the results show that PS and PC are\neffective in two folds---improving classification performances and stabilizing\nthe training process. Although convolutional features are the most important\nlearning targets in CNN models, global information of images may be lost in the\ntraining process. Because topological features were extracted globally, our\nresults show that the global property of topological features provide\nadditional information to machine learning models.\n