Artificial intelligence supported computer vision for identification of antibiotic response of vancomycin towards gram-positive bacteria

Abstract

The spread of antibiotic resistance is a major public health issue on the global scale that requires the development of new techniques that are both rapid and reliable for antimicrobial susceptibility testing (AST). This research introduces a low-cost, AI-supported computer vision system that, is currently still virtual but will make a real impact, to look into whether Staphylococcus carnosus absorbed the antibiotic vancomycin. On the one hand, thanks to phase-contrast microscopy and the YOLOv5 framework, the system has managed to unleash 1,200 grayscale images of bacterial cells allowing the classification of the effects of vancomycin, among others, with accuracy of 97% precision, 92% recall, and 94.5% F1 score (mAP: 0.977). The construction of a sophisticated microfluidic system for incubating bacteria has resulted in unprecedented levels of environmental control, and yet, even traditional systems have not been this successful. This method is a non-visualised, strain-free step, with the promise of scalability and efficiency.