Deep Learning based Segmentation and Volume Estimation for Prognostication of Intracerebral Hemorrhage Advancing Neuro Critical Care

Intracerebral hemorrhage (ICH) is a severe and often fatal type of stroke, necessitating accurate and timely diagnosis for improved patient outcomes. Traditionally, radiol-ogists manually assess hemorrhages in computed tomography (CT) imaging, a laborintensive process prone to variability. In addition, ICH volume calculation is crucial for determining hemorrhage severity (normal, moderate, severe). This study introduces an automated deep learning (DL) technique for segmenting and calculating ICH volume in using publicly available datasets BHSD and HemSeg-200. We evaluated advanced 2D medical image segmentation techniques and found that YOLOv5 outperformed other models in speed and accuracy. The proposed model achieved IoU score of 73.2%, significantly outperforming the previous SOTA results of 50.8% and 66.9%. This approach aims to detect the severity of hemorrhage at the early stage of stroke. This helps for better treatment and fast recovery of stroke patients. It also enhance patient care and reduce radiologist's workload by improving the precision and effectiveness of ICH diagnosis and treatment planning.