A hybrid CNN-LSTM approach for detecting intracerebral hemorrhage in CT images

A Stroke is a life-risk condition caused due to lack in blood flow to the part of the brain. Accurate diagnosis using computed tomographic (CT) scans is crucial for immediate care in the emergency department. A multidisciplinary approach is essential to manage this health challenge. Medical imaging using computed tomography (CT) and MRI data from stroke patients plays a crucial role in supporting clinical decision-making. Locating lesions in image is a challenging and time-consuming task for radiologists, hence automated techniques powered by Artificial Intelligence (AI) is vital for detecting and quantifying Intracerebral Hemorrhage (ICH). In this study, we proposed deep learning approach based on CNN along with long short-term memory (CNN-LSTM) for accurate prediction of ICH on CT scans. Non-contrast whole-head CT scan images were collected from 72 hemorrhagic stroke patients at KIMS, Bhubaneswar. A total of 2,394 head CT images were used to train the models with both CNN and CNN-LSTM architectures with a split of $80: 20$. The performance of the two models was then compared to evaluate their effectiveness. The CNN alone achieved an accuracy of 0.95, a recall of 0.89, and an $F 1$-score of 0.94, whereas the hybrid CNN-LSTM model outperformed it with an accuracy of 0.97, a recall of 0.97, and an $\mathbf{F 1}$-score of 0.97. The findings indicate that the hybrid CNN-LSTM model performs better in detecting ICH lesions in CT images. Further research is essential to enhance the detection of hemorrhagic stroke in clinical settings for effective management of stroke.