Deep learning-enhanced microwell array biochip for rapid and precise quantification of Cryptococcus subtypes

Abstract

Cryptococcus is a family of strongly infectious pathogens that results in a wide variety of symptoms, particularly threatening the patients undergoing the immune-deficiency or medical treatment. Rapidly identifying Cryptococcus subtypes and accurately quantifying their contents remain urgent needs for infection control and timely therapy. However, traditional detection techniques heavily rely on expensive, specialized instruments, significantly compromising their applicability for large-scale population screening. In this work, we report a portable microwell array chip platform integrated with a deep learning-based image recognition program, which enables rapid, precise quantification of the specific subtypes of Cryptococcus. The platform features four zones of microwell arrays preloaded with the subtype-targeted CRISPR–Cas12a system that avoid dependence on slow, instrumental-mediated target amplification, achieving rapid (10 min), high specificity for identifying the sequence of Cryptococcus. The deep learning-based image recognition program utilizing segment anything model (SAM) significantly enhances automation and accuracy in identifying target concentrations, which eventually achieves ultra-low limit of detection (0.5 pM) by personal smartphones. This platform can be further customized to adapt to various scenarios in clinical settings.