Cloud-Enabled Healthcare: A Scalable Approach To Disease Research And Cure Using Artificial Intelligence

Abstract

Since the inception of the COVID-19 pandemic, artificial intelligence has led to biomarker discovery and causal inferences for clinical outcomes from unlabeled and structured healthcare datasets. Biomedical data are produced in vast amounts and at high speeds, but remain mostly untapped in the absence of knowledge about their origins. A novel, comprehensive, cloud-enabled, collaborative, and automated artificial-intelligence-assisted platform and methodology for early and scalable research, development, and clinical adoption of novel biomarkers, disease subtyping, or targeted treatments from big data is presented. The combination of federated algorithms, standard ontological data representations, and a health data exchange will enable safe, scalable, and ethical research endeavours on sensitive data without transfer of original data. A rigorous and transparent evaluation of biomarker performance in spatial disease locations and through novel notions of clinical validation are designed. This platform bridges instrumented biomedical data production and artificial-intelligence-driven insights, enabling human-scale automated science. Artificial intelligence (AI) has a history of success in domains such as image recognition, where with the advent of sufficient data and compute, some tasks reached human-level performance. Healthcare is often thought naive of such success, both in simple reasoning health tasks and automation of unguided exploration of patient data. Three key, non-exclusive healthcare properties have prevented previous successes of AI. Drawn from molecular medicine, mammographic screening, and the COVID-19 pandemic, an evidence basis is presented, together with a novel clinical AI platform expected to enable scalable, early retrieval of biomedical knowledge. The properties of precision (benign/uninformative) populations, wide-grained, prognostic and inexpensive data, and opaque nuisance variables facilitate cost and complexity reduction in imaging and population biomedicine, while presenting unique challenges in molecular medicine and for a decade-spanning infectious disease. These properties imply a reversible change in the perceived relevance of standalone AI white-box systems. Standard open and fully facilitated app-based systems could now transparently and ethically interrogate semi-rigid inter-instrument discoverable data on fast timescales. An elegant methodology to inform both collaborative human-scale discovery of knowledge and a healthy AI economy is outlined.