Cerebral hemorrhage detection and localization with medical imaging for cerebrovascular disease diagnosis and treatment using explainable deep learning

Cerebral hemorrhages require rapid diagnosis and intensive treatment. This study aimed to detect cerebral hemorrhages and their locations in images using a deep learning model applying explainable deep learning. Normal brain images with no hemorrhages and images with subarachnoid, intraventricular, subdural, epidural, and intraparenchymal hemorrhages according to computed tomography (CT) (n = 200) were analyzed. A ResNet deep learning model, including image processing, was utilized. The visual explanation from a heatmap was made at the hemorrhage location using a gradient-class activation map (Grad-CAM). To evaluate the performance of the deep learning system, the accuracy, sensitivity, and specificity were determined. A hemorrhage prediction system for images of normal brains and brains with subarachnoid, intraventricular, subdural, epidural, and intraparenchymal hemorrhages was built. The Grad-CAM representation indicated the location of the hemorrhages in these images. In the prediction results, accurate predictions of the hemorrhage areas were made and visualizations of the corresponding locations overlapped in the images within (− 4, 1) pixel difference. The evaluation of the system performance showed an accuracy of 0.81 with a sensitivity of 0.67 and specificity of 0.86. These results constitue a proof of concept for the use of explainable artificial intelligence (XAI) to detect cerebral hemorrhages and visualize their locations in medical images, which will allow rapid diagnosis and treatment.