Deep learning-assisted portable sensing platform based on a fluorescent MOF-polymer mixed-matrix membrane for rapid detection of nitrazepam in biological samples

Nitrazepam (NZP) is a benzodiazepine sedative-hypnotic widely used in the treatment of anxiety and sleep disorders. However, non-medical use of NZP as a new psychoactive substance poses serious health risks. Therefore, rapid and accurate detection of NZP in biological samples is important for point-of-care testing. Herein, a deep learning-assisted fluorescent sensing platform was developed based on a mixed-matrix membrane (1@PMMA), fabricated by incorporating a Cd-MOF, {[Cd(L)(H₂O)₂]·3H₂O}_n (1) (H₂L = 5-(3-(pyridin-4-yl)ureido)isophthalic acid), into a poly(methyl methacrylate) (PMMA) matrix for NZP detection. The resulting 1@PMMA exhibited good mechanical stability, processability, and fluorescence stability in biological samples. Upon exposure to NZP, 1@PMMA showed efficient fluorescence quenching, with a quenching efficiency of 96.19%, a detection limit of 0.21 μM, and a response time of approximately 2 s. The membrane maintained high selectivity in the presence of common biological components and stable sensing performance over five quenching-regeneration cycles. A smartphone-based fluorescence sensing method enabled portable visual detection of NZP through extraction of RGB values from fluorescence images of 1@PMMA. Additionally, a deep learning-based photo-to-result platform employing a ResNet-18 model directly predicted NZP concentrations from fluorescence images without manual data processing, showing a strong linear relationship between predicted and experimental concentrations (R² = 0.9949). For spiked urine and serum samples, recoveries of 95.18-100.46% were obtained, with relative standard deviations of 0.18-3.10%. This work integrates a fluorescent MOF-based mixed-matrix membrane with a deep learning-assisted photo-to-result strategy for on-site quantitative determination of NZP in biological samples.