Microscopic Urinary Sediments Detection Using Deep Learning

Urine particles and sediments are important requirements in clinical urinalysis, particularly in the diagnosis and monitoring of patients suspected of renal diseases and urinary tract infections. As a result, it is critical to identify urinary particles accurately in the clinical area. Also, the outcome is hugely affected by the lab technician ’s experience. However, because the traditional manual microscopic analysis relies on human operators who read the samples visually and identify them, this method is slow, time-consuming, and labor-intensive. This research presents a deep learning method for analyzing urinary particles from microscopic images. It investigates the design, implementation, and evaluation of models for classifying sediments in urine samples. Using three datasets from different sources, multiple classes were distinguished. Several deep learning models of CNN like DenseNet, VGG16, and MobileNet were implemented and compared for their classification performance. Additionally, this study explores data fusion techniques that combine information from multiple data sources to enhance classification. Results showed that the VGG-16 model achieved the highest accuracy of 91% whereas DenseNet and MobileNet demonstrated accuracy results of 89% and 84% respectively. This is greatly beneficial for clinicians to assist, analyze, and quantify urine sediment images.