Age-related changes of the brain’s arterial network assessed with machine learning segmentation

Purpose To provide a tool for the automatic segmentation of an arteriogram of the brain from MRA images and the estimation of arterial tortuosity as a summary marker. Methods A deep learning model was trained and validated on a previously published set of semi-automatically segmented brain arteriograms. We tested whether arterial tortuosity estimated from a large number of age-representative subjects (N = 478) would reproduce previously published statistics of increasing tortuosity with age. Results The tool provides a segmentation of the arteriograms from MRA images of varying resolution and quality. The arterial tortuosity estimated from the automatically segmented brain arteriograms approximately matched their previously published statistics. Further, a highly significant increase of tortuosity with age was observed in the large dataset with 478 subjects (p = 9 ×10 -8 ). Discussion and conclusion The proposed ASN (Angiogram Segmentation Network) algorithm can provide the radiologist with a clean arteriogram of the brain, without the need for manual segmentation by the MR operator. Moreover, it offers arterial tortuosity as an instant quantitative metric that can augment the qualitative visual reading of the MRA.