摘要
Biliary atresia (BA) is a rare but life-threatening neonatal liver disease and remains one of the leading causes of pediatric liver transplantation worldwide. If untreated, the 2-year survival rate is below 10%, and delayed diagnosis markedly reduces the success rate of Kasai portoenterostomy. Current diagnostic methods rely primarily on dried blood spots (DBS) and ELISA, which require relatively large blood volumes, are time-consuming, and are poorly suited for routine newborn screening, creating an urgent need for rapid and neonate-friendly diagnostic tools. Here, we present an AI-assisted biosensing platform for rapid and quantitative detection of matrix metalloproteinase-7 (MMP7), a clinically validated biomarker for BA. The platform integrates an electric double-layer extended-gate field-effect transistor (EDL-EGFET) with an AI-based calibration strategy that compensates for chip-to-chip variation and enables consistent quantification across sensors. By incorporating process parameters into model training, the AI system effectively reduces chip-to-chip variability, significantly improving the robustness of the sensing platform. Clinical validation using 20 patient samples demonstrated a strong correlation with ELISA-based quantification. The Pearson correlation coefficient improved from 0.80 using the conventional method to 0.96 with the AI-enhanced model. The system achieved a limit of detection of 0.32 ng/mL, covered the clinically relevant MMP7 range, and reduced the total diagnostic time from 8 h to 10 min. Only 10 μL (one drop) of blood was required per test. This innovative AI-driven EDL-EGFET platform demonstrates strong potential for advancing precise clinical diagnostics, enhancing neonatal care, and addressing the diagnostic challenges associated with this neglected disease.
[Display omitted]
•Investigated and compared four prediction models, with the AI-enhanced calibration model outperforming the others.•Leveraged AI-driven process tolerance to reduce chip-to-chip variability in biosensors.•Developed an AI-trained buffer-to-blood correlation model, minimizing reliance on blood samples and significantly lowering diagnostic costs.•Achieved superior predictive accuracy for clinical samples with the AI-enhanced calibration model, increasing the coefficient of determination by 41.5%.•Revolutionized infant diagnostics by reducing testing time from 8 hours to just 10 minutes and requiring only a single drop of blood.