AI-Ready Multimodal Wearable Biosensors Beyond Glucose: Biofluid Sampling, Sensor Fusion, and Clinical Translation
TL;DR
This review proposes an artificial intelligence (AI)-ready framework for multimodal wearable biochemical monitoring and highlights recurring gaps in biofluid validity, real-world robustness, reference-comparator alignment, subgroup evidence, and algorithmic governance.
Abstract
Continuous glucose monitoring has established that a molecular signal can be repeatedly measured in daily life and translated into clinically meaningful action. The next frontier is broader: wearable biosensors that monitor metabolites, electrolytes, hormones, drugs, nutrients, inflammatory markers, and tissue-state indicators beyond glucose. This review proposes an artificial intelligence (AI)-ready framework for multimodal wearable biochemical monitoring. We organize the field as a complete measurement chain that links biofluid access, sampling chronology, flexible biointerfaces, molecular recognition, signal conditioning, metadata capture, sensor fusion, clinical validation, and lifecycle governance. The central argument is that the clinically useful variable is rarely a raw current, potential, optical intensity, or spectrum; it is a quality-controlled, context-aware, and uncertainty-aware digital biomarker. We compare sweat, interstitial fluid, saliva, tears, wound exudate, and breath condensate; evaluate enzymatic, ion-selective, affinity, transistor, optical, and spectroscopic sensing strategies; synthesize representative high-impact studies; and define minimum metadata, validation metrics, and translation gates. The review highlights recurring gaps in biofluid validity, real-world robustness, reference-comparator alignment, subgroup evidence, and algorithmic governance. We conclude with practical design rules for converting flexible biochemical wearables from attractive prototypes into clinically credible intelligent biosensing systems.