Detection of intracranial hemorrhage on CT scan images using convolutional neural network

Abstract

Intracranial hemorrhage is an acute bleeding within the skull which can damage the brain tissue and can eventually lead to disability or even death. It is a serious medical condition that occurs when blood is built up within the skull after a blood vessel is ruptured. Brain damage can be minimized if intracranial hemorrhage is diagnosed immediately, and the patient may regain mobility. Deploying applications of Artificial Intelligence (AI) in clinical medicine to accelerate the accuracy of intracranial hemorrhage diagnosis aims to minimize the severity of the condition, therefore, enhancing medical care. Adequate analysis of the Computed Tomography (CT) scan imaging is integral for diagnosis and management. Deep Learning, which is a subset of AI, is widely used in interpreting medical images and has shown promising advancements in diagnosing brain hemorrhage. With time playing a crucial factor, automatic lesion identification is one of the most important factors in precision medicine dealing with huge datasets of neuroimaging compared to manual lesion segmentation. This paper proposes a Deep Learning method called Convolutional Neural Network (CNN) on neuroimaging with transfer learning techniques to assist in the diagnosis of intracranial hemorrhage on CT scan images. We used six pretrained CNN models (EfficientNetB6, DenseNet121, ResNet50, InceptionRes- NetV2, InceptionV3, VGG16) and also present a traditional 11-layer CNN model for binary classification and detection of intracranial hemorrhage using brain CT scan images. The paper depicts a comparative analysis on the performance between the proposed traditional and pre-trained CNN models in terms of accuracy, precision, recall, F1 score, and AUC curve on the existing dataset. The EfficientNetB6 model yields an accuracy of 95.99%, which is higher than any of the experimental results of the CNN models used in this dataset.