Brain Hemorrhage Detection Using Improved AlexNet with Inception-v4

Injury in the human brain is complex and delicate to study. Following a traumatic brain injury (TBI), there is a risk of intracranial hemorrhage (ICH), which can have severe consequences, including fatality or lifelong disabilities, if not promptly diagnosed and treated. The manual diagnosis of ICH is a time-consuming process and is also prone to errors. This paper presents an advanced transfer learning-based mechanism using AlexNet combined with Inception-V4 to automatically detect a brain hemorrhage. Furthermore, it compares the performance with individual deep learning models. Experimental results on the Computed Tomography scans dataset show that, in terms of accuracy and F1 score, the proposed approach outperforms other machine learning methods, including a two-dimensional convolution neural network (2D CNN), bidirectional long short-Term memory (BLSTM), and support vector machine (SVM). The proposed approach obtains the highest accuracy of 94.54%, much better than 85.07%, 79.2%, and 71.53% for 2D CNN, BLSTM, and SVM, respectively. Also, the highest F1 score for the proposed approach is 0.938, much better than 0.846, 0.781, and 0.693 for 2D CNN, BLSTM, and SVM, respectively. The performance in terms of accuracy, time consumption, and F1 score and the non-data-hungry nature indicate the potential usefulness of the proposed approach for brain hemorrhage detection.