Active Noise Cancelling in Point-of-Care Testing

Abstract

Point-of-care testing (POCT) holds great potential to improve diagnostic accessibility because of its low cost, portability, user-friendliness, and rapid turnaround times. However, the limited analytical precision and high error rates of most POCT platforms usually caused by the assay noises at the POC settings frequently create diagnostic ambiguity and lead to further confirmatory testing. Consequently, these limitations restrict broader clinical translation. To address this issue, we introduce an active assay noise canceling approach, using the deep-learning-based Whole-Process Analysis (WPA), a universal, device-adaptable software paradigm without altering assay hardware or chemistry. Instead of relying on static endpoint readouts, WPA uses a deep-learning architecture to continuously capture the spatiotemporal evolution of the entire POCT assay process. The architecture decodes dynamic, high-dimensional "assay noise footprints" and algorithmically compensate for the signal variations caused by assay noises. We demonstrated the proof of concept, we chose the lateral flow assay (LFA) as a model, which is a leading POCT platform. We validated WPA for xylazine LFA across 251 patient urine samples, benchmarking against gold-standard liquid chromatography-tandem mass spectrometry (LC-MS/MS). WPA drastically decreased signal variations among different clinical samples and compressed the diagnostic gray zone, delivering 94.02% accuracy (95% CI: 90.33%-96.62%), 71.15% sensitivity (95% CI: 56.92%-82.87%), and 100% specificity (95% CI: 98.16%-100.00%), compared with human reading, which achieved 89.2% accuracy (95% CI: 84.74%-92.79%), 48.1% sensitivity (95% CI: 34.01%-62.37%), and 100% specificity (95% CI: 98.16%-100.00%). By computationally untangling true analyte signals from assay noises in POC settings, WPA provides a transformative, device-adaptable pathway to improve POCT accuracy and precision without altering assay hardware or chemistry.