Urinary Sediments Classification Using Image Processing

Abstract

Urinary sediment analysis is a very basic diagnostic test that helps to identify and assess the presence of red blood cells, white blood cells, epithelial cells, crystals, and bacteria. Microscopic examination using human eyes is subjective and labor-intensive and would be susceptible to artificially induced inter-observer variation undermining diagnostic reliability. The proposed project suggests an automated image-processing-based Convolutional Neural Network (CNN) to use in MATLAB to classify urinary sediments. The system uses a hierarchical chain of preprocessing, segmentation, featurehood and classification of microscopic urine images. Preprocessing methods are used to improve image, whereas morphological and texture characteristics are retrieved on segmented particles. The CNN classifier then classifies the sediments. It is an automated process that offers more objective and accurate data of urinary 
sediment analysis by helping clinical personnel and minimizing human error. The experiment shows that CNNs are effective in 
high classification accuracy achieving high validity showing the promise of standardizing urinalysis and aiding the consistency 
of diagnostic findings.