Acute kidney injury silently progresses before standard blood or urine tests catch it — a window during which irreversible damage accumulates. A wearable, skin-penetrating sensor that could flag kidney stress within the first day of onset would fundamentally change the monitoring calculus for critically ill patients, post-surgical cases, and those on nephrotoxic medications.
Researchers engineered a microneedle patch that harvests interstitial fluid transdermally and detects neutrophil gelatinase-associated lipocalin (NGAL) — a validated early-warning protein for acute kidney injury — using a multi-layered nanosensor architecture. The patch employs aptamer-guided DNA strand assembly to link carbon dots (fluorescent donors) with gold nanoparticles (energy-absorbing acceptors) on each microneedle tip. In the absence of NGAL, energy transfer between these nanoparticles quenches fluorescence. When NGAL binds its aptamer, gold nanoparticles detach, restoring a measurable fluorescence signal. Because raw fluorescent images from microneedle arrays are spatially noisy, the team integrated a U-Net deep learning model to isolate true signal from background interference — raising the correlation between fluorescence recovery and NGAL concentration from R² = 0.567 to R² = 0.994. Applied to an AKI mouse model, the patch reliably detected elevated NGAL (~80.7 ng/mL) within 24 hours of injury onset.
This convergence of aptamer chemistry, nanoparticle energy transfer, and AI-assisted image analysis addresses three longstanding bottlenecks in continuous kidney monitoring simultaneously. Current NGAL assays rely on blood draws or urine collection — neither is suitable for real-time, continuous, minimally invasive surveillance. Interstitial fluid is increasingly recognized as an accessible proxy for blood biomarker dynamics, and microneedle-based ISF sampling avoids the pain and infection risks of venipuncture. The critical caveat is that this work remains entirely in a rodent model; translating nanomaterial-based microneedles to human skin — with its variable thickness, hydration, and immune reactivity — introduces substantial engineering and regulatory hurdles. Nonetheless, the deep learning signal-extraction framework is transferable and the design modular, making this an architecturally significant platform that warrants accelerated preclinical development.