Fully Automated Deep Learning Enabled Miniature Mass Spectrometry System for Psychoactive Therapeutic Drug Monitoring

Advancing precision medicine requires efficient small molecule biomarker detection in biofluids, yet existing methods encounter challenges in complexity, portability, and throughput. This study presents an integrated miniature blood processing and mass spectrometry (MS) analysis system, which incorporates automated magnetic solid-phase extraction, self-aspiration sampling miniature mass spectrometer, and deep learning algorithms for automated quantitative analysis. It achieves full automation from sample preparation to detection, demonstrating the capability to analyze serum psychoactive drugs with a 15-second/MS acquisition and 8-sample parallel processing within 30 minutes (including pretreatment). This has significantly increased detection throughput and facilitated the establishment of the standard curve. The novel dual-target ion parallel tandem MS analysis technique, combined with a U-net peak area recognition algorithm, achieved over 98% identification accuracy with less than 0.2% area prediction deviation. Quantitative analysis showed high correlation coefficients >0.99 in medically relevant ranges, supported by relative standard deviation < 10% and average back-calculated accuracy deviation < 3.5%. Clinical validation revealed strong concordance with LC-MS/MS. The system's integration of automated sample processing, miniature MS hardware, and AI-driven data analysis establishes a paradigm for high-throughput clinical detection. The advantages of accuracy, automation, intelligence, miniaturization, and high throughput suggest significant potential for this system in clinical detection and personalized medicine.