Deep learning-assisted sensitive detection of fentanyl using a bubbling-microchip

• DOI: 10.1039/d2lc00478j
• Publication Date: 2022-10-31T16:41:22Z
• Authors: Hui Chen; Sungwan Kim; Joseph Michael Ha...; Prudhvi Thirumala...; Supriya Gharpure; Sahar Rostamian; Srisruthi Udayakumar; Qingsong Lei; Giwon Cho; Manoj Kumar Kanak...; Hadi Shafiee
• Keywords: Fentanyl; 0301 basic medicine; 03 medical and health sciences; Deep Learning; Image Processing, Computer-Assisted; Humans; Metal Nanoparticles; Algorithms; Retrospective Studies; Platinum

Deep learning-enabled smartphone-based image processing has significant advantages in the development of point-of-care diagnostics. Conventionally, most deep-learning applications require task specific large scale expertly annotated datasets. Therefore, these algorithms are oftentimes limited only to that have retrospective datasets available for network development. Here, we report possibility utilizing adversarial neural networks overcome this challenge by expanding utility non-specific data deep learning models. As a clinical model, detection fentanyl, small molecular weight drug is type opioid, at using empowered smartphone assay. We used catalytic property platinum nanoparticles (PtNPs) smartphone-enabled microchip bubbling assay achieve high analytical sensitivity (detecting fentanyl concentrations as low 0.23 ng mL-1 phosphate buffered saline (PBS), 0.43 human serum and 0.64 artificial urine). Image-based inferences were made our adversarial-based SPyDERMAN was developed dataset 104 images microchips with bubble signals from tests performed known library 17 573 bubbling-microchip images. The accuracy (± standard error mean) system determining presence when cutoff concentration 1 mL-1, 93 ± 0% (n = 100) 95.3 1.5% urine 100).