Chronic kidney disease stage identification using texture analysis of ultrasound images

Abstract Early detection of chronic kidney disease (CKD) stages is an important aspect of human healthcare. Clinical approach for identification of CKD stage is expensive, time consuming and mostly not accessible to common people living in remote locations. An automatic method based on ultrasound (US) imaging is more appropriate to receive the first-hand information of kidney health in quick time. Extraction of local features from US image and selection of classifier plays a major role to improve the CKD stage detection accuracy. In this paper, gray level co-occurrence matrix (GLCM) used to extract local features of US kidney images in terms of a fourteen-point feature-vectors. The classifier processes the extracted feature-vectors to identify CKD stage. Extensive simulation performed in MATLAB using different multiclass classifier on 300 sample US images collected by the radiologist to select the multiclass classifier suitable to identify CKD stages with higher accuracy. Simulation result shows that linear discriminant analysis (LDA) classifier performs better than other available multiclass classifiers to classify CKD stages with higher accuracy. Proposed scheme identifies, CKD stage-1, stage-2 and stage-3 with accuracies nearly 96.82%, 100% and 98.38%, respectively. The proposed scheme using the fourteen-point feature vector identify CKD stages with 16% higher accuracy over the exisitng schemes. However, the proposed scheme able to identify CKD stages with 24% higher accuracy than the existing scheme when both the fourteen-point feature-vector and the LDA classifier are used. Therefore, the proposed scheme could be an alternative to the existing scheme for automatic CKD stage identification.