A comprehensive review of the deep learning-based tumor analysis approaches in histopathological images: segmentation, classification and multi-learning tasks

Medical Imaging has become a vital technique that has been embraced in the diagnosis and treatment process of cancer. Histopathological slides, which microscopically examine the suspicious tissue, are considered the golden standard for tumor prognosis and diagnosis. This excellent performance caused a sudden and growing interest in digitizing these slides to generate Whole Slide Images (WSI). However, analyzing WSI is a very challenging task due to the multiple-resolution, large-scale nature of these images. Therefore, WSI-based Computer-Aided Diagnosis (CAD) analysis gains increasing attention as a secondary decision support tool to enhance healthcare by alleviating pathologists' workload and reducing misdiagnosis rates. Recent revolutionized deep learning techniques are promising and have the potential to achieve efficient automatic representation of WSI features in a data-driven manner. Thus, in this survey, we focus mainly on deep learning-based CAD systems in the context of tumor analysis in histopathological images, i.e., segmentation and classification of tumor regions. We present a visual taxonomy of deep learning approaches that provides a systematic structure to the vast number of diverse models proposed until now. We sought to identify challenges that face the automation of histopathological analysis, the commonly used public datasets, and evaluation metrics and discuss recent methodologies for addressing them through a systematic examination of presented deep solutions. The survey aims to highlight the existing gaps and limitations of the recent deep learning-based WSI approaches to explore the possible avenues for potential enhancements.