Automated analysis and diagnosis of skin melanoma on whole slide histopathological images

Melanoma is the most aggressive type of skin cancer, and the pathological examination remains the gold standard for the final diagnosis. Traditionally, the histopathology slides are examined under a microscope by pathologists which typically leads to inter- and intra-observer variations. In addition, it is time consuming and tedious to analyze a whole glass slide manually. In this paper, we propose an efficient technique for automated analysis and diagnosis of the skin whole slide image. The proposed technique consists of five modules: epidermis segmentation, keratinocytes segmentation, melanocytes detection, feature construction and classification. Since the epidermis, keratinocytes and melanocytes are important cues for the pathologists, these regions are first segmented. Based on the segmented regions of interest, the spatial distribution and morphological features are constructed. These features, representing a skin tissue, are classified by a multi-class support vector machine classifier. Experimental results show that the proposed technique is able to provide a satisfactory performance (with about 90% classification accuracy) and is able to assist the pathologist for the skin tissue analysis and diagnosis.