eduzhai > Applied Sciences > Engineering >

Surface-based 3D Deep Learning Framework for Segmentation of Intracranial Aneurysms from TOF-MRA Images

  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Segmentation of intracranial aneurysms is an important task in medicaldiagnosis and surgical planning. Volume-based deep learning frameworks havebeen proposed for this task; however, they are not effective. In this study, wepropose a surface-based deep learning framework that achieves higherperformance by leveraging human intervention. First, the usersemi-automatically generates a surface representation of the principal brainarteries model from time-of-flight magnetic resonance angiography images. Thesystem then samples 3D vessel surface fragments from the entire brain arterymodel and classifies the surface fragments into those with and withoutaneurysms using the point-based deep learning network (PointNet++). Next, thesystem applies surface segmentation (SO-Net) to the surface fragmentscontaining aneurysms. We conduct a head-to-head comparison of segmentationperformance by counting voxels between the proposed surface-based framework andexisting pixel-based framework, and our framework achieved a much higher dicesimilarity coefficient score (72 ) than the existing one (46 ).

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...