Enhanced Architecture of Convolutional Neural Network (CNN) for Intracranial Hemorrhage Prediction from CT-Scan Images

Head injury is the leading cause of mortality and morbidity and often results in pathological bleeding inside the skull, known as Intracranial Hemorrhage (ICH). Currently, the diagnosis of ICH and its type is still manually performed by taking brain images, commonly known as CT scans, which are then analyzed by experienced radiology specialists. In this study, a Convolutional Neural Network (CNN) modeling was conducted as a tool for doctors to detect ICH based on CT scans images. The CNN architecture used to build the classification model underwent four variations of adjustments until the recommended CNN architecture was obtained. With enhanced architecture, the classification achieved an average accuracy of 80% for each type of brain hemorrhage. The study shows that modifying the CNN architecture leads to obtaining a recommended architecture, which ultimately yields a more precise model compared to the initial architecture.