Colorimetry-based and smartphone-assisted machine-learning model for quantification of urinary albumin

Abstract

Presence of albumin in urine is indicative of kidney damage and can happen due to several underlying conditions like diabetes. The concentration of albumin in urine is used for diagnosis and staging of chronic kidney disease. In clinical samples, detection of albumin at lower concentrations is very crucial for early diagnosis and monitoring of chronic kidney disease (CKD). Current urine analyzers precisely quantify albumin but are expensive and difficult to use in point-of-care (PoC) settings. Here, we demonstrate the quantification of albumin concentration in urine sample using colorimetry. This model presents an accessory-free urine analyzer that uses a smartphone and customized machine learning algorithms. Here, urine sample is introduced onto the chemically impregnated dipstick that exhibits the change in its color with the amount of albumin. The images of the urine dipsticks are captured using the camera of the smartphone under different illumination/experimental conditions and are processed to extract the change in the color values arising due to change in the concentration of the urinary albumin. Concentrations of the albumin are estimated from the change in color values. We have used customized machine learning algorithms for the classification of albumin concentrations and to mitigate the effect of ambient light conditions. The k-Nearest Neighbor (kNN) algorithm yielded an average classification accuracy of 96% with detection limit of 4 mg/L. Proposed scheme can be extensively used to monitor albumin concentration in PoC settings.