The performance of current machine learning methods to detect heterogeneous pathology is limited by the quantity and quality of pathology in medical images. A possible solution is anomaly detection; an approach that can detect all abnormalities by learning how ‘normal’ tissue looks like. In this work, we propose an anomaly detection method using a neural network architecture for the detection of chronic brain infarcts on brain MR images. The neural network was trained to learn the visual appearance of normal appearing brains of 697 patients. We evaluated its performance on the detection of chronic brain infarcts in 225 patients, which were previously labeled. Our proposed method detected 374 chronic brain infarcts (68% of the total amount of brain infarcts) which represented 97.5% of the total infarct volume. Additionally, 26 new brain infarcts were identified that were originally missed by the radiologist during radiological reading. Our proposed method also detected white matter hyperintensities, anomalous calcifications, and imaging artefacts. This work shows that anomaly detection is a powerful approach for the detection of multiple brain abnormalities, and can potentially be used to improve the radiological workflow efficiency by guiding radiologists to brain anomalies which otherwise remain unnoticed.

Abstract4D phase contrast magnetic resonance imaging (PC-MRI) allows for the visualization and quantification of the cerebral blood flow. A drawback of software that is used to quantify the cerebral blood flow is that it oftentimes assumes a static arterial luminal area over the cardiac cycle. Quantifying the lumen area pulsatility index (aPI), i.e. the change in lumen area due to an increase in distending pressure over the cardiac cycle, can provide insight in the stiffness of the arteries. Arterial stiffness has received increased attention as a predictor in the development of cerebrovascular disease. In this study, we introduce software that allows for measurement of the aPI as well as the blood flow velocity pulsatility index (vPI) from 4D PC-MRI. The internal carotid arteries of seven volunteers were imaged using 7 T MRI. The aPI and vPI measurements from 4D PC-MRI were validated against measurements from 2D PC-MRI at two levels of the internal carotid arteries (C3 and C7). The aPI and vPI computed from 4D PC-MRI were comparable to those measured from 2D PC-MRI (aPI: mean difference: 0.03 (limits of agreement: −0.14 – 0.23); vPI: 0.03 (−0.17–0.23)). The measured blood flow rate for the C3 and C7 segments was similar, indicating that our proposed software correctly captures the variation in arterial lumen area and blood flow velocity that exists along the distal end of the carotid artery. Our software may potentially aid in identifying changes in arterial stiffness of the intracranial arteries caused by pathological changes to the vessel wall.

The intracranial arteries play a major role in cerebrovascular disease, but arterial remodeling due to hypertension has not been well described in humans. We aimed to quantify this remodeling for: the basilar artery, the vertebral, internal carotid, middle/anterior (inferior)/posterior cerebral, posterior communicating, and superior cerebellar arteries of the circle of Willis. Ex vivo circle of Willis specimens, selected from individuals with (n=24) and without (n=25) a history of hypertension, were imaged at 7T magnetic resonance imaging using a 3-dimensional gradient-echo sequence. Subsequently, histological analysis was performed. We validated the vessel wall thickness and area measurements from magnetic resonance imaging against histology. Next, we investigated potential differences in vessel wall thickness and area between both groups using both techniques. Finally, using histological analysis, we investigated potential differences in arterial wall stiffness and atherosclerotic plaque severity and load. All analyses were unadjusted. Magnetic resonance imaging and histology showed comparable vessel wall thickness (mean difference: 0.04 mm (limits of agreement:−0.12 to 0.19 mm) and area (0.43 mm 2 [−0.97 to 1.8 mm 2 ]) measurements. We observed no statistically significant differences in vessel wall thickness and area between both groups using either technique. Histological analysis showed early and advanced atherosclerotic plaques in almost all arteries for both groups. The arterial wall stiffness was significantly higher for the internal carotid artery in the hypertensive group. Concluding, we did not observe vessel wall thickening in the circle of Willis arteries in individuals with a history of hypertension using either technique. Using histological analysis, we observed a difference in vessel wall composition for the internal carotid artery.

No abstract