nnDetection for Intracranial Aneurysms Detection and Localization
This work addresses a critical medical problem for patients at risk of life-threatening aneurysms, but it is incremental as it applies an existing framework to a specific dataset.
The study tackled the detection and localization of intracranial aneurysms in medical images using the nnDetection framework, achieving effective 3D coordinate prediction as evaluated with free-response receiver operative characteristics, with model weights and predictions made publicly available.
Intracranial aneurysms are a commonly occurring and life-threatening condition, affecting approximately 3.2% of the general population. Consequently, detecting these aneurysms plays a crucial role in their management. Lesion detection involves the simultaneous localization and categorization of abnormalities within medical images. In this study, we employed the nnDetection framework, a self-configuring framework specifically designed for 3D medical object detection, to detect and localize the 3D coordinates of aneurysms effectively. To capture and extract diverse features associated with aneurysms, we utilized TOF-MRA and structural MRI, both obtained from the ADAM dataset. The performance of our proposed deep learning model was assessed through the utilization of free-response receiver operative characteristics for evaluation purposes. The model's weights and 3D prediction of the bounding box of TOF-MRA are publicly available at https://github.com/orouskhani/AneurysmDetection.