A cost-effective system for automated early antimicrobial susceptibility testing using deep learning

Abstract

We demonstrate an automated, cost-effective system that delivers early antimicrobial-susceptibility-testing results, minimizing incubation time and eliminating human errors, while remaining compatible with standard clinical workflow. A neural network processes the time-lapse intensity information from a fiber-optic array to detect growth in each well of a 96-wellplate. Our blind testing on clinical Staphylococcus aureus infections reveals that 95.03% of all the wells containing growth were correctly identified, with an average incubation time of 5.72-h. This deep learning-based optical system met the FDA-defined essential and categorical agreement criteria for all 14 antibiotics tested, after an average of <7-h of incubation time.