Automatic detection of intracranial aneurysm from digital subtraction angiography with cascade networks

Abstract

Automatic detection of intracranial aneurysm based on Digital Subtraction Angiography (DSA) images is a challenging task for the following reasons: 1) effectively leverage the temporal information of the DSA sequence; 2) effectively extract features by avoiding unnecessary interference in the raw DSA images of large resolution; 3) effectively distinguish the vascular overlap from intracranial aneurysm in DSA images. To better identify intracranial aneurysm from DSA images, this paper proposed an automatic detection framework with cascade networks. This framework is consisted of a region localization stage (RLS) and an intracranial aneurysm detection stage (IADS). The RLS stage can significantly reduce the interference from unrelated regions and determine the coarse effective region. The IADS stage fully employed the spatial and temporal features to accurately detect aneurysm from DSA sequence. This method was verified in the posterior communicating artery (PCoA) region of internal carotid artery (ICA). In clinical trials, the accuracy of the baseline method was 62.5% with area under curve (AUC) of 0.650, and the time cost of the detection was approximately 62.546s. However, the accuracy of this method was 85.5% with AUC of 0.918, and the time cost of detection was about 3.664s. The experimental results showed that the proposed method significantly improved the accuracy and speed of intracranial aneurysm automatic detection.