An Efficient CNN-BiLSTM-Based Model for Multi-Class Intracranial Hemorrhage Classification

Intracranial hemorrhage (ICH) is a type of bleeding that occurs within the skull. ICH may be caused by a wide range of pathology, including, trauma, hypertension, cerebral amyloid angiopathy, and cerebral aneurysms. Depending on the location inside the brain, an ICH is typically classified into five different subtypes: epidural hemorrhage (EDH), subdural hemorrhage (SDH), subarachnoid hemorrhage (SAH), intraven-tricular hemorrhage (IVH), and intraparenchymal hemorrhage (IPH). ICH is a life-threatening medical emergency with high morbidity and mortality rates, especially when not detected and managed promptly. Therefore, the presence, type, and location of an ICH must be immediately diagnosed so that the patients can receive medical intervention. However, the detection of ICH in CT slices can be challenging due to the complex anatomy of the brain and the variability in the appearance of hemorrhages. Recently, deep learning models have shown great effectiveness in medical imaging interpretation tasks that are comparable to, and sometimes surpass, trained medical specialists. This research investigates an efficient deep learning approach that integrates a 2-D convolutional neural network (CNN) with a bidirectional long-short-term memory (Bi-LSTM) module to achieve accurate ICH detection and subtype classification. The proposed model achieves a weighted multi-label logarithmic loss score of 0.0501 on the benchmark RSNA brain hemorrhage CT scan dataset. It has the potential to be deployed as an intelligent assistive tool (IAT) for radiologists in the accurate early diagnosis of ICH.