Lacunes are small cerebrospinal fluid-filled lesions that are generated by the occlusion of penetrating deep branches of cerebral arteries. Early detection of lacunes could decrease the possible clinical implications such as dementia, gait impairment, and lacunar stroke. In this study, we propose a deep learning 3D multi-scale residual network for lacunes identification using FLAIR and T1-MPRAGE MR images. We redesign the proposed network via applying multiple parallel paths using different input scales. This enables to extract more robust contextual global features and hence achieve better detection performance. The proposed work exhibits its ability to distinguish true lacunes from non-lacunes.
This abstract and the presentation materials are available to members only; a login is required.