Detection of Intracranial Hemorrhage by Artificial Intelligence and Deep Learning Methods

This paper addresses the task of intracranial hemorrhage detection and classification based on the RSNA 2019 brain computed tomography (CT) dataset. Several popular neural network (NN) architectures based on convolutional neural networks (CNNs) have been considered and interpretation of the neural network classifier by GradCAM method has been shown. To efficiently process the dataset, various radiology information processing methods were considered to create the resulting training sample. In addition, the training sample was reduced to 10
^%
of the total dataset to avoid technical limitations and improve performance. The main contribution of the work is the creation and testing of deep learning algorithms for the detection of acute intracranial hemorrhage and its five subtypes. Various steps related to data cleaning, augmentation and preprocessing have also been applied in the presented work.