Deep Learning-Based Histopathological Image Analysis for Automated Detection and Staging of Melanoma

Salah Alheejawi,Mrinal Mandal,Hongming Xu,Cheng Lu,Richard Berendt,Naresh Jha

DOI: https://doi.org/10.1016/b978-0-12-819061-6.00010-0

2020-01-01

Abstract:With recent advances in digital microscopy, quantitative analysis of pathology images has become important in disease diagnosis as well as in understanding the underlying reasons. Digital whole slide images (WSI) are typically obtained from a biopsy slide using a high-resolution slide scanner. The hematoxylin and eosin (H&E) stain is typically used to stain the slides. However, immunospecific stains such as MART-1 (for melanoma) and PD-L1 (for lung cancer) are also sometimes used to validate detection. In advanced stages, cancer cells spread to lymph nodes through the lymph vessels. Therefore, a lymph node biopsy is typically done by doctors to determine if the cancer cells have invaded the lymph node system. High-resolution WSIs typically have very large size, in the order of a billion pixels. Hence, image analysis and diagnosis (quantitative as well as qualitative) is expected to take significant time. In this chapter, we present the state of the art on computer-aided diagnosis and grading of skin cancer/melanoma based on histopathology. The techniques generally include traditional image analysis modules such as preprocessing, segmentation, feature extraction, and classifications. Recently, deep learning techniques such as SegNet and U-Net have been used successfully in histopathological image analysis. The detailed implementation of our works on melanoma detection and proliferation index calculation using deep learning networks is presented and compared with traditional feature-based techniques. Experimental results show superior performance by the deep learning networks.