Web application for ICH subtype classification from CT head scans

Abstract

Traumatic brain injury (TBI) causes intracranial hemorrhage (ICH) that requires urgent diagnosis and treatment to improve patient outcome. Machine learning techniques can help clinicians to classify brain lesions and assist clinicians diagnose TBI from radiological scans. The project objective was to build a CAD system which assists in the detection, screening, and diagnosis of ICH in routine clinical practice. The models are trained and created using different CNN models developed on Tensorflow, Keras, and OpenCV using sliced CT scanned images from the 2019-RSNA Brain CT Hemorrhage Challenge dataset. The results from these models were evaluated and the MobileNetV1 architecture model is determined to give the best performance analysis. The CAD system, which was constructed using the Django and ReactJS frameworks, was able to extract medical picture analysis for use in a deep learning solution.