Healthcare analytics on multimodality data by deep learning

Abstract

“What’s wrong with healthcare today is that it’s missing care”. Doctors have been criticized for paying too little time, attention, and sympathy to patients. At a glance, they may appear to be at the center of the problem. However, their burnout has drained away their capability of doing more energy-consuming “human work” and they themselves are also victims of the strain of medical resources, which has been a major issue in the healthcare system globally. This strain has more immediate complications, such as a long waiting time to get serviced and a high volume of misdiagnoses. Unfortunately, the increasingly aged global population only exacerbates the problem. The field of medicine needs to confront this dire situation, and the solution is unlikely to be humans. Recently, Artificial Intelligence (AI) has been demonstrated to achieve human-level performance in many tasks considered to require human intelligence. Its core technology, deep learning, has been proven to hold great promise in augmenting doctors’ decisions. This work contributes to technological development by proposing three state-of-the-art architectures in their specific task during the time they were developed. Additionally, complementary to proof-of-concept algorithms, this work also explores the model’s performance in a live environment and devises algorithms to boost model transparency, both are critical to the adoption of AI in medicine.