Wim Van Hecke1, Diana Sima1, Arno Liseune1, Ricardo Magalhães1, Thanh Vân Phan1, Simon Van Eyndhoven1, Arne Brys1, Rafay Khan1, Melissa Wittens2, Gert-Jan Allemeersch3, Tim Vanderhasselt3, Ann Vanbinst3, Johan de Mey3, Sebastiaan Engelborghs2, Dirk Smeets1
1icometrix, 2Center for Neurosciences (C4N) and Department of Neurology / Reference Center for Biological Markers of Dementia, Laboratory of Neurochemistry and Behavior, Vrije Universiteit Brussel / University of Antwerp, 3Radiology, UZ Brussel
Objective:
Develop a robust automated deep learning-based method for assessment of hyperintensities on T2 FLAIR images.
Background:
Cerebrovascular lesions often occur in Alzheimer’s Disease (AD) and may be an important modifier of AD progression. Automated MRI volumetry plays an important role in assessing the presence and extent of vascular lesions. icobrain dm is an established regulatory-cleared brain MRI quantification tool used in clinics worldwide to evaluate neurological markers pertaining to dementia care.
Design/Methods:
icobrain dm segments brain tissues and substructures on T1-weighted images and hyperintensities on fluid attenuated inversion recovery (FLAIR) images. Recent versions of icobrain dm employ fully-deep-learning-based segmentation models, trained on a variety of brain scans over the whole lifespan2. Reproducibility of icobrain dm 5.16 was assessed in test-retest studies acquired in four MRI scanners (GE 3T, Siemens 3T, Philips 3T and Philips 1.5T) from 5 patients in different stages of the AD continuum (aged 69.0 ± 10.5) and 5 controls (aged 52.2 ± 17.6)3.
Results:
All controls had less than 3ml FLAIR hyperintensity volumes, while the AD patients had volumes ranging from 4ml to 21ml. Overall, the absolute test-retest error was 0.04 ± 0.06 ml (mean±std; 0.02ml median, 0.08ml p90) for same-scanner comparisons (n=40) and 0.17 ± 0.18 ml (mean±std; 0.09ml median, 0.45ml p90) for inter-scanner comparisons (n=240).
Conclusions:
icobrain dm 5.16 produces robust and reproducible segmentations of vascular lesions in the elderly population. These findings underscore the value of icobrain dm for early and accurate disease detection and monitoring to help optimize patient monitoring and care.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.