Portable urine analysis device based on optical spectroscopy and machine learning

Optical spectroscopy combined with machine learning offers a promising route to affordable, portable urinalysis. Here, we report a low-cost, noninvasive device that leverages a machine-learning model and visible and near-infrared transmission spectroscopy. The handheld system records spectra of LED light transmitted through a standard urine collection container using an 18-channel spectrometer covering 410–940 nm. From the optical spectra, we trained 19 parameter-specific binary classifiers for specific gravity, pH, protein, glucose, ketones, leukocyte esterase, nitrite, urobilinogen, blood, red and white blood cells, squamous and transitional epithelial cells, hyaline and pathological casts, bacteria, crystals, yeast, and mucus. In cross-validated experiments, the method achieved parameter-dependent performance, with balanced sensitivity and specificity for several analytes, and identified others as candidates for further optimization. These results demonstrate that compact visible and near-infrared spectroscopy paired with machine learning can support multiparametric urinalysis using inexpensive hardware and no reagents. Together, these results demonstrate the feasibility and translational potential of compact, reagent-free multiparametric urinalysis and support its further development toward point-of-care and home-use screening.