Deep Learning-Based Automated Landmark Localization for Evan’s Index Computation

Abstract: Hydrocephalus is a neurological disorder characterized by the accumulation of cerebrospinal fluid in the brain, leading to an enlargement of the ventricular system. Among its subtypes is idiopathic normal pressure hydrocephalus (iNPH), characterized by normal cerebrospinal fluid pressure. Accurately diagnosing iNPH presents considerable difficulties due to its non-specific clinical manifestations. This thesis presents an innovative approach for calculating Evan’s Index by accurately estimating the Anterior Commisure (AC), Posterior Commisure (PC), and Vertex of the Superior Pontine Sulcus (VSPS) landmarks thereby aiding the iNPH diagnosis process. The primary emphasis of this study lies in harnessing dedicated frameworks tailored for medical image segmentation. The project constructs a streamlined pipeline that precisely segments the AC, PC, and VSPS regions, and also performs MRI scan alignment to calculate the Evan’s Index. The study investigates the effectiveness of this approach in providing an automated and efficient estimation of the landmark points. The methodology includes network training, and evaluation, followed by the analysis of the results. The outcomes of this study highlight the potential of deep learning techniques in assisting clinicians with iNPH diagnosis.