A Smartphone and Web‐Based Automated Platform for Segmenting Urinary Tract Infection Using a Deep Learning‐Based Approach

Abstract

ABSTRACT Urinary tract infections (UTIs) affect millions of people annually, with early and accurate diagnosis being essential for effective treatment. Traditional diagnostic methods such as urine culture and dipstick testing are often slow, costly, and sometimes unreliable, particularly in low‐resource settings where access to advanced laboratory facilities is limited. This study presents a deep learning–powered urinary sediment analysis system designed to operate on both web and mobile platforms. At its core is a UNet++ multi‐class segmentation model trained to identify various sediment types, including background, rod, red blood cells/white blood cells, yeast, miscellaneous, single epithelial cell, small epithelial cell sheet, and large epithelial cell sheet. The model was trained for 50 epochs using the Adam optimizer with a learning rate of 0.001 and evaluated using the Dice coefficient, Intersection over Union, precision, recall, and area under the curve. The background class achieved the highest accuracy (Dice coefficient = 0.9963, Intersection over Union = 0.9926, area under the curve = 0.9587), while rare categories such as yeast (Dice coefficient = 0.0092) and miscellaneous (Dice coefficient = 0.0234) were more difficult to detect due to class imbalance and visual similarity. In experimental performance tests, the system processed complex samples in about 11 s and simpler ones in 1–2 s, instantly displaying results on the web or mobile interface. This integrated approach offers a faster, more consistent alternative to traditional methods, with the potential to improve access to automated urine sediment image analysis in resource‐constrained healthcare environments.