Recognition of Intracranial Hemorrhage with its Subtypes from CT Images using Deep Learning Approach

Abstract

Intracranial Hemorrhage (ICH) or Intracranial Bleeding is a severe problem for a person who rushes towards death unless it is recognized accurately within a short period. Commonly, it occurs within the skull or brain. Since the brain is the most crucial part of the human body and consists of soft tissue, its lesion detection, and treatment are so sensitive and also challenging within a faster time. The radiologist analyzes brain CT images to diagnose diseases consuming more time that affects a severe patient negatively. Being the process slow, non-automated, and costly, an automated deep learning approach is proposed to diagnose the problems accurately within the shortest time. For our research, the Radiological Society of North America (RSNA) dataset is considered, which consists of 752,803 training images and 121,232 test images. Normalization, windowing, and augmentation are used as preprocessing techniques on the training data, and during the model training, 30% of data are treated as validation data for hyper-tuning. After the successful completion of learning, this study has applied the test data to evaluate the performance of the proposed model, and it was observed that the proposed model performs better in the recognition of ICH with 98% accuracy.