Object Detection for MRI Image Based Segmentation

Abstract

Low Back Pain (LBP) is a worldwide disorder that affects almost all people during their life span. The dominant factors that influence the onset and course of back pain include smoking, educational status, obesity, female, older age, physically strenuous work, sedentary works, job dissatisfaction and stressful job. The physiological factors that lead to low back pain are weight, anxiety or depression, and structural defects of the spinal column. The most common cause of LBP is Intervertebral disc (IVD) degeneration. The clinical routine consists of a physical examination followed by the analysis of spine images. newlineIVD lies in between two adjacent vertebrae and acts as cushions between the bones. The water contents of discs are about 80%. As age progresses, the water content gets decreases and the disc will degenerates. Each disc consists of two parts; the hard, tough, outer layer called the Annulus Fibrosus (AF), surrounds a mushy, moist center termed the Nucleus Pulposus (NP). The inner region appears as a bright ellipse surrounded by the AF. In the analysis of clinical images, detection of Disc Degeneration (DD) is the major challenging one. The main challenges facing the segmentation of IVD include (i) Partial volume effect (ii) Bias field distortion. The analysis uses different imaging modalities such as radiography, MRI (Magnetic Resonance Imaging) and Positron Emission Tomography (PET), and Computed Tomography (CT). newlineMRI is an indispensable image modality in the prognosis of DD since discs are more visible in these images. Manual segmentation is a labor-intensive process. Thus, arises the need for a Computer Aided Design (CAD) to analyze spine MRI. The computer-assisted method involves quantitative analysis of DD, disease progression, and surgical planning. The segmentation of Spine MRI is an indispensable step in the diagnosis of various spinal pathological scenarios such as DD, herniation, scoliosis, etc. newlineThis thesis proposes region-based algorithm, texture-based algorithm, and deep learning techniques for t