ECR 2026: Poster Presentation
Pim Moeskops, Lotte Mulder, Edgar Wakelin, Annie Ng, Jiye Kim, Bryan Haslam
Abstract
Introduction
White matter hyperintensities (WMHs) are clinically actionable imaging markers of several neurological conditions, including cognitive impairment and multiple sclerosis (MS). Manual segmentation of WMHs on Fluid-Attenuated Inversion Recovery (FLAIR) MRI is time-consuming and impractical for routine clinical practice, limiting consistent use in risk stratification and longitudinal assessment. This study evaluates an AI algorithm for automated quantification and localization by identifying, segmenting, and labelling anatomical location of WMHs.
Methods
A total of 110 FLAIR MRI exams were acquired from 90 patients between 2011-2022 on 1.5T and 3T scanners from GE, Philips and Siemens. For each exam, ground truth WMH segmentation and anatomical location labelling (juxtacortical, periventricular, deep white matter and infratentorial) was produced by expert radiologists: two radiologists performed segmentation and labelling, and an independent radiologist adjudicated between the two results. A subset of 43 studies from 34 patients were segmented and labelled by an independent expert not involved in the ground truth to measure benchmark expert performance. The algorithm was compared to benchmark expert performance in terms of segmentation (Dice) and location labelling (Kappa) performance.
Results
The patient cohort was 65% female with clinical indication of cognitive impairment or MS, with an average age of 53.3±16.7 years.
The algorithm achieved a Dice coefficient of 0.58 (95%CI: 0.54–0.62) for the whole dataset, similar to benchmark expert performance: 0.54 (95%CI: 0.47–0.61). Furthermore, the algorithm achieved a Kappa coefficient for location labelling of 0.63 (95%CI: 0.58–0.68), similar to benchmark expert performance: 0.65 (95%CI: 0.57-0.72).
Conclusion
The AI algorithm demonstrated comparable performance to benchmark expert performance, indicating its ability to automatically provide accurate and consistent quantification of WMHs to support longitudinal assessment, risk stratification, and clinical decision making.
Disclaimers:
Disclaimers: Neuro Suite comprises multiple applications including DeepHealth Brain Health, DeepHealth Brain Age and DeepHealth Viewer. DeepHealth Brain Health is manufactured by Quantib B.V. and Distributed by DeepHealth in the U.S. DeepHealth Viewer is Manufactured by eRad, Inc and distributed by DeepHealth, Inc. Any claims made about Neuro Suite may reference claims associated with its individual components. Not all products and functionalities are commercially available in all countries.
Any claims made about Neuro Suite may reference claims associated with its individual components.
Not all products and functionalities are commercially available in all countries. For clearance and commercial availability in your geography of functionalities listed and compatibility with other systems, please contact your account manager.