ICH Segmentation Accuracy Improvement Using Gaussian Filtering

Abstract

Intracranial Hemorrhage (ICH) refers to bleeding inside the skull, which can have various causes such as head trauma, arterial blockage, and blood clotting disorders. Timely detection of ICH in Head Computed Tomography (CT) scans is crucial for improved patient outcomes. However, limited expertise in interpreting CT scans can lead to missed haemorrhage diagnoses. To address this, we propose a computer-aided diagnosis model that employs deep learning techniques and Gaussian filtering for pre-processing. Gaussian filtering is employed as a pre-processing step to enhance CT scan images, effectively reducing noise and enhancing image quality. The filtered images are then subjected to an advanced and highly developed segmentation algorithm customized to identify ICH regions with precision. A deep learning architecture called U net is utilized to segment the haemorrhage region from the CT scan of the head. By incorporating Gaussian filtering into the workflow, we have significantly improved the accuracy of ICH segmentation. Our empirical findings underscore substantial progress in segmentation accuracy. Through attentive testing, our approach achieves remarkable precision and recall rates, for the particular problem. This heightened accuracy has great remarks for patient care, enabling clinicians to identify and treat intracranial haemorrhages more effectively and quickly. Our experimental results demonstrate a significant improvement in testing accuracy, achieving 99.82%.