Automated histopathological analyses at scale

Abstract

Histopathology is the microscopic examination of processed human tissues to diagnose conditions like cancer, tuberculosis, anemia and myocardial infractions. The diagnostic procedure is, however, very tedious, time-consuming and prone to misinterpretation. It also requires highly trained pathologists to operate, making it unsuitable for large-scale screening in resource-constrained settings, where experts are scarce and expensive. In this thesis, we present a software system for automated screening, backed by deep learning algorithms. This cost-effective, easily-scalable solution can be operated by minimally trained health workers and would extend the reach of histopathological analyses to settings such as rural villages, mass-screening camps and mobile health clinics. With metastatic breast cancer as our primary case study, we describe how the system could be used to test for the presence of a tumor, determine the precise location of a lesion, as well as the severity stage of a patient. We examine how the algorithms are combined into an end-to-end pipeline for utilization by hospitals, doctors and clinicians on a Software as a Service (SaaS) model. Finally, we discuss potential deployment strategies for the technology, as well an analysis of the market and distribution chain in the specific case of the current Indian healthcare ecosystem.