# Modularly-Assembled Smart Microneedle Platform for Machine Learning-Driven Personalized Health Monitoring

## TL;DR

A smart microneedle patch continuously monitors six metabolic biomarkers and uses machine learning to provide personalized health insights.

## Contribution

A modular, flexible microneedle biosensor patch with machine learning for real-time, personalized health monitoring.

## Key findings

- The eMPatch achieved classification accuracy of 0.996 in distinguishing normal and diet-induced metabolic disorders.
- The system demonstrated an R2 score of 0.977 for evaluating the degree of metabolic disorder.
- The patch enables real-time physiological tracking across diverse daily activities in animal models.

## Abstract

A skin-interfaced microneedle patch simultaneously and continuously measures six metabolic biomarkers from dermal interstitial fluid—glucose, uric acid, cholesterol, sodium, potassium, and pH.Modular microneedle units assembled on a compliant polystyrene-isoprene-polystyrene substrate offer mechanical robustness and excellent flexibility, enabling seamless adhesion, stable skin-sensor coupling, and user-specific configuration, which delivers durable, conformal wear with high signal fidelity in daily use.An end-to-end personalized health evaluation system: high-dimensional multiplexed signals are processed by an optimized machine-learning pipeline to quantify and predict metabolic responses to daily behaviors, supporting personalized guidance (e.g., postprandial control, electrolyte balance).

A skin-interfaced microneedle patch simultaneously and continuously measures six metabolic biomarkers from dermal interstitial fluid—glucose, uric acid, cholesterol, sodium, potassium, and pH.

Modular microneedle units assembled on a compliant polystyrene-isoprene-polystyrene substrate offer mechanical robustness and excellent flexibility, enabling seamless adhesion, stable skin-sensor coupling, and user-specific configuration, which delivers durable, conformal wear with high signal fidelity in daily use.

An end-to-end personalized health evaluation system: high-dimensional multiplexed signals are processed by an optimized machine-learning pipeline to quantify and predict metabolic responses to daily behaviors, supporting personalized guidance (e.g., postprandial control, electrolyte balance).

The online version contains supplementary material available at 10.1007/s40820-026-02095-x.

Given the inherent complexity of metabolic pathways and disease-associated agents, next-generation healthcare necessitates wearable, non-invasive, and customized approaches to continuously monitor a broad spectrum of physiologically relevant biomarkers for personalized health management. Moreover, existing data-based analytical strategies remain inadequate for delivering quantitative and predictive evaluations of health status in real-life settings. Here, we report an electronic multiplexed microneedle-based biosensor patch (eMPatch) that enables real-time, minimally invasive monitoring of key metabolic biomarkers in interstitial fluid, including glucose, uric acid, cholesterol, sodium, potassium, and pH. By integrating modular microneedle (MN) sensors into a skin-interfaced flexible platform, the eMPatch achieves robust mechanical stability and seamless skin conformity, thereby ensuring reliable and continuous sensing within the dermal space. In vivo validation in animal models under metabolic intervention highlights the strong capability of the eMPatch for real-time physiological tracking across diverse daily activities. Implemented with a machine learning algorithm, the eMPatch enables automatic feature extraction and multi-task health assessment, achieving a classification accuracy of 0.996 in distinguishing normal and diet-induced metabolic disorder for health condition identification and an R2 score of 0.977 for the corresponding degree evaluation. This study highlights the potential of the MN-integrated, machine learning-enhanced biosensing platform toward personalized health management.

The online version contains supplementary material available at 10.1007/s40820-026-02095-x.