Microinterventional in-sensor computing system for real-time metabolic health assessment

Microneedle biosensors enable dynamic monitoring of interstitial fluid biomarkers, but remain constrained by sensing interface susceptibility to motion artifacts and the prohibitive energy consumption of wireless cloud-based processing. Here, we report a bio-inspired, self-anchoring microinterventional in-sensor computing system. By leveraging a starfish-inspired suction cup-microneedle self-anchoring mechanism, the system effectively counteracts microneedle extrusion, attenuating signal fluctuations by 38-fold and enhancing signal intensity by up to 5.49-fold compared to conventional planar devices. Crucially, the high-fidelity data acquisition reduces the computational burden, enabling the deployment of a lightweight algorithm (43 KB) on a coin-sized embedded circuit, achieving 98.68% diagnostic accuracy and a 120-h battery life via local closed-loop feedback. Validation in a porcine model confirmed the system’s capability to capture continuous biochemical dynamics. This co-design of a robust biomimetic interface and lightweight deep learning paves the way for next-generation wearables capable of performing high-fidelity, on-chip metabolic risk stratification in dynamic daily settings. Continuous metabolic monitoring is hindered by motion artifacts. Here, the authors present a self-anchoring in-sensor computing system that provides high-fidelity data acquisition and edge processing for real-time ketogenic health assessment