The State of the Art and Potential Challenges in AI Health

Abstract

Health has not been far behind in the fever for Artificial intelligence (AI) technologies that have raged a whirlwind across multiple sectors these last few years. Alongside with growing computer hardware and software applications in medicine, also the digitization of health-related data accelerates both development and application of AI techniques for medicine. This progress, however, brings us new opportunities and challenges as a movement of the future for AI in health. This survey aims to shed light on the state-of-the-art in AI-based health improvements, opportunities, challenges and practical implications. In this review, we provide an overview of progress made over the past five years and future directions. Pertinent publications over the past five years of AI use in health reported by clinical and biomedical informatics journals, as well computer science conferences identified through Google Scholar citation. Next, the publication were segmented into five categories of data types studied. The most common data types identified in the results were multiomics, clinical, behavioural health research/data and R&D ecosystem environmental/systems approach to pharmaceutical breakthrough treatments. We then articulate the current state of AI with respect to each data type, describe related challenges and practical learnings that have come up in past few years. Discussions are provided on opportunities and where this may lead in the future with these advances. AI in healthcare is based on the technologies that have come with AI-assisted forms of care. Still, there are challenges that lie ahead. Current work is on multi-modal data integration, quant vs qual trade-off of an algorithm performance (how much a traffic controller intervenes given the prediction),