Statistical Feature Analysis and Machine Learning based Classification of Lower Back Pain Using Thermal Images

Abstract

Nearly one-quarter of the world's population suffers from low back pain (LBP). LBP can originate from several different areas of the body, including the nerves, spinal cord, discs, bones, and tendons of the lumbar spine. Getting a proper diagnosis of LBP at an early stage is the initial step toward a speedy and complete recovery. Though much effort and money have been invested in LBP research approaches, successful diagnosis remains an important objective, and LBP remains to be a major reason for concern in primary healthcare. The inability of conventional medical images to identify the LBP was one of the reasons for the above problem. This research intends to provide a five-step process for automatic LBP detection using thermal images. At first, the thermal images of both healthy and LBP individuals are taken. Second, the images are analysed by employing Grey Level Co-occurrence Matrix (GLCM) and Grey Level Run-Length Matrix (GRLM) techniques to get a total of 18 features. Third, using the Particle Swarm Optimisation (PSO) method, the most crucial of the 18 features is determined. Fourth, ML model training involves using both the raw feature data and the optimized feature data. Finally, the performance of the ML model is assessed using both data before and after feature selection to determine the optimal method for LBP detection.