Automated Urine Color Analysis using MTCNN for Rapid and Non-Invasive UTI Detection

One of the most common diseases, urinary tract infections (UTIs) are typically detected by laboratory urine analysis. This work introduces an automated UTI detection system that uses Multi-Task Cascaded Convolutional Neural Networks (MTCNN) for analyzing the urine color images provided by the users. The severity of the infection is classified into three categories, namely, normal, mild and severe according to color intensity and visual characteristics. Preprocessing methods including color correction and histogram normalization were employed to ensure that the data are normalized. The MTCNN framework is able to extract texture, hue and spatial features efficiently with the help of its multi-stage cascaded design and outperforms traditional color-threshold and rule-based classification methods in accuracy and stability. Experimental results show that MTCNN can learn the correlation between urine color characteristics and the level of infection well. The proposed approach provides a fast, cheap and non-invasive diagnostic solution for remote monitoring of patients and mobile health applications. Future work has the purpose of improving clinical reliability by combining multimodal parameters such as odor, pH, and chemical indicators in order to advance intelligent and accessible healthcare diagnostics through deep learning-based color images analysis.