SBIR Phase I: MedSwarm, an Artificial Intelligence System for Medical Decision Making

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to demonstrate the feasibility of MedSwarm, a software product for radiology that will significantly improve patient outcomes and reduce medical costs. MedSwarm is an AI-based collaborative decision-making system that will significantly amplify diagnostic accuracy across a wide range diagnoses, not by replacing radiologists with AI, but by amplifying their human expertise. MedSwarm employs a technology called Artificial Swarm Intelligence that collects input from small groups of doctors and uses AI to optimize their combined accuracy. This hybrid human-AI system will achieve diagnostic accuracies that significantly outperform (a) individual radiologists using standard techniques, and (b) current state-of-the-art in software-only AI. As a product, the MedSwarm system will be deployed as downloadable software compatible with all existing radiological workstations. Distribution to diagnostic teams will be as easy as opening a web-browser and connecting to a cloud-based server. The market opportunity is best modeled on the Computer Aided Detection (CAD) market, which is projected at $1.9 Billion by 2022. MedSwarm will not compete with current CAD products, as it will provide additional accuracy on all radiology workstations, whether or not other CAD tools are in use. This Small Business Innovation Research (SBIR) Phase I project aims to develop, test, and validate MedSwarm, the first commercially viable system for amplifying diagnostic accuracy of radiologists by combining human and machine intelligence. The technology, called Swarm AI, does not replace radiologists with AI, but connects small groups of radiologists together and optimizes their combined insights, enabling accuracy levels that greatly exceed baseline abilities. The company anticipates diagnostic error reductions of at least 30% using the system developed in Phase I, with the majority of the improvement being in the reduction of false-positives. The proposed research requires the development of new software modules, new optimization algorithms, and new interface protocols for enabling radiologists to connect online, view the same images at the same time, and work together as a real-time system to converge on unified diagnoses. Development also requires machine learning optimization across multiple experimental parameters, to reduce the time required for each diagnosis, increasing throughput and reducing costs. The company's goals for Phase I include a proof-of-concept system that (i) significantly outperforms individual radiologists, (ii) significantly outperforms state-of-the-art machine learning systems, and (iii) can identify which cases are most appropriate for this system, as compared to alternate methods. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.