CBIR-DSN: integrating clustering and retrieval platforms for disk space narrowing degradation assessment

Abstract

A system that is capable of assessing spine osteoarthritis conditions which affect a significant portion of the elderly population could be very valuable to radiologists, researchers of arthritis and musculoskeletal diseases, and educators. To this end, there is very limited research published in the literature concerning the degradation assessment of spinal intervertebral disc space narrowing (DSN). Thus, this paper intends to develop a system that focuses on assessing the degradation of disk space narrowing (DSN) to assist in radiologist’s decision-making in the characterization of cervical and lumbar images. A novel experiment based on our previous research (Aouache et al. 2009; Aouache et al. Biomed Eng Online 14(1):6, 2015) was conducted by integrating clustering and retrieval platforms to achieve this objective. Two shape boundary, 9-points, and B-spline have been used as the foundation for DSN model construction using active shape model. The segmented DSNs have then indexed via region and contour-based features descriptor. For better efficiency, clustering using a vocabulary tree model (VTM) is constructed to identify correct DSN cluster and build multi-clusters subsets for faster and robust retrieval research process. Our system achieved an accuracy of average retrieval rate (ARR) more than 90% and 88% for cervical and lumbar data set accordingly. We expect the proposed system will assist in decision-making and uses by radiologists or researchers for further investigation.