H2OPulse: Smartphone-assisted Vein Evaluation for Early Recognition of Dehydration

Maintaining human health relies on adequate water intake for digestion, waste elimination, and temperature regulation. Dehydration occurs when the body loses more fluids than it takes in, leading to electrolyte imbalances and serious complications, especially in vulnerable populations. Traditional dehydration detection methods rely on laboratory tests and physical exams, which can be invasive, expensive, time-consuming, and often inaccessible in rural areas. Recent research has proposed several solutions, but these are often either unreliable or not widely accessible. We introduce H2OPulse, the first smartphone-based solution to use vein visibility patterns for dehydration detection, analyzing changes in dorsal and wrist veins to distinguish between hydrated and dehydrated states. Our detection pipeline uses optimized image segmentation and enhancement with transfer learning and Siamese networks to accurately learn and differentiate vein visibility features. Our dataset comprises 3,440 hand vein images from 86 healthy volunteers, captured in both hydrated and dehydrated conditions. Experimental results demonstrate that H2OPulse can accurately detect mild dehydration, making it a practical and accessible solution, with an accuracy of 83.1% and an F1-score of 80.8%. This system empowers early dehydration detection anywhere, enabling timely interventions to prevent further complications.