Rapid, point-of-care bacterial bioburden detection on diabetic foot ulcers, burn wounds, and surgical site wounds, and subsequent gram-type classification is a challenge that cannot be addressed in low-resource settings. This is because current gold-standard practices based on cell culture take 2-3 days for a pathogen detection and classification. Further, these methods are cumbersome, require sophisticated microbiological facilities, and trained personnel. To address this issue, we demonstrate a rapid (<2 min), point-of-care platform, Illuminate®, that combines autofluorescence intensity imaging and deep learning to detect bacterial bioburden. We use a UNET deep learning model with a DenseNet 201 backbone to detect bioburden spatially and, to further classify the gram-type. Our initial results show a >80% accuracy in detecting bioburden and a >70% accuracy in gram-type classification. We further use this platform for identifying clinically relevant bacteria such as pseudomonas aeruginosa that cause antimicrobial resistance.
|