Automatic arteriovenous separation of brain via TOF-MRA and MR-T1

Time-of-flight (TOF) Magnetic Resonance Angiography (MRA) and MR-T1 are the most commonly used imaging modalities in clinical diagnosis. Vascular extraction is of great significance for the computer-assisted diagnosis and treatment, where completeness, accuracy, and computation speed are the focused issues of vascular segmentation. Besides, it is very concerned that the lack of prior knowledge impacts the effectiveness of cerebral artery-vein (CA/CV) separation. In this paper, a high-efficient statistical method is proposed to cope with the challenges, where the aforementioned image modalities provide important functional and structural information. Our work contributes on following aspects: (1) For cerebrovascular segmentation, the Markov random field (MRF) of vascular direction vector is first employed for the statistical model to segment TOF-MRA, which greatly improves traditional MRF model with optimal space restraints and well avoids structure confusion in region of low contrast and weak vessel illuminance. (2) The cerebral priori knowledge (w.r.t. vessel and tissues) is explored via TOF-MRA and MR-T1 data, then a morphological algorithm is proposed to automate CA/CV separation. We employ public datasets from MIDAS for qualitative and quantitative assessment. The Dice similarity coefficient of cerebrovascular segmentation is 0.928, as well as the agreement of CA/CV separation results is 0.972.