Wi-Sun Ryu, Jae W. Song, Jae-Sung Lim, Ju Hyung Lee, Leonard Sunwoo, Dongmin Kim, Dong-Eog Kim, Hee-Joon Bae, Myungjae Lee, Beom Joon Kim
{"title":"Segmentation of Leukoaraiosis on Noncontrast Head CT Using CT-MRI Paired Data Without Human Annotation","authors":"Wi-Sun Ryu, Jae W. Song, Jae-Sung Lim, Ju Hyung Lee, Leonard Sunwoo, Dongmin Kim, Dong-Eog Kim, Hee-Joon Bae, Myungjae Lee, Beom Joon Kim","doi":"10.1002/brb3.70602","DOIUrl":null,"url":null,"abstract":"<div>\n \n \n <section>\n \n <h3> Objective</h3>\n \n <p>Evaluating leukoaraiosis (LA) on CT is challenging due to its low contrast and similarity to parenchymal gliosis. We developed and validated a deep learning algorithm for LA segmentation using CT-MRIFLAIR paired data from a multicenter Korean registry and tested it in a US dataset.</p>\n </section>\n \n <section>\n \n <h3> Methods</h3>\n \n <p>We constructed a large multicenter dataset of CT–FLAIR MRI pairs. Using validated software to segment white matter hyperintensity (WMH) on FLAIR, we generated pseudo-ground-truth LA labels on CT through deformable image registration. A 2D nnU-Net architecture was trained solely on CT images and registered masks. Performance was evaluated using the Dice similarity coefficient (DSC), concordance correlation coefficient (CCC), and Pearson correlation across internal, external, and US validation cohorts. Clinical associations of predicted LA volume with age, risk factors, and poststroke outcomes were also analyzed.</p>\n </section>\n \n <section>\n \n <h3> Results</h3>\n \n <p>The external test set yielded a DSC of 0.527, with high volume correlations against registered LA (<i>r</i> = 0.953) and WMH (<i>r</i> = 0.951). In the external testing and US datasets, predicted LA volumes correlated with Fazekas grade (<i>r</i> = 0.832–0.891) and the correlations were consistent across CT vendors and infarct volumes. In an independent clinical cohort (<i>n</i> = 867), LA volume was independently associated with age, vascular risk factors, and 3-month functional outcomes.</p>\n </section>\n \n <section>\n \n <h3> Interpretation</h3>\n \n <p>Our deep learning algorithm offers a reproducible method for LA segmentation on CT, bridging the gap between CT and MRI assessments in patients with ischemic stroke.</p>\n </section>\n </div>","PeriodicalId":9081,"journal":{"name":"Brain and Behavior","volume":"15 6","pages":""},"PeriodicalIF":2.6000,"publicationDate":"2025-06-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/brb3.70602","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Brain and Behavior","FirstCategoryId":"102","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1002/brb3.70602","RegionNum":3,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"BEHAVIORAL SCIENCES","Score":null,"Total":0}
引用次数: 0
Abstract
Objective
Evaluating leukoaraiosis (LA) on CT is challenging due to its low contrast and similarity to parenchymal gliosis. We developed and validated a deep learning algorithm for LA segmentation using CT-MRIFLAIR paired data from a multicenter Korean registry and tested it in a US dataset.
Methods
We constructed a large multicenter dataset of CT–FLAIR MRI pairs. Using validated software to segment white matter hyperintensity (WMH) on FLAIR, we generated pseudo-ground-truth LA labels on CT through deformable image registration. A 2D nnU-Net architecture was trained solely on CT images and registered masks. Performance was evaluated using the Dice similarity coefficient (DSC), concordance correlation coefficient (CCC), and Pearson correlation across internal, external, and US validation cohorts. Clinical associations of predicted LA volume with age, risk factors, and poststroke outcomes were also analyzed.
Results
The external test set yielded a DSC of 0.527, with high volume correlations against registered LA (r = 0.953) and WMH (r = 0.951). In the external testing and US datasets, predicted LA volumes correlated with Fazekas grade (r = 0.832–0.891) and the correlations were consistent across CT vendors and infarct volumes. In an independent clinical cohort (n = 867), LA volume was independently associated with age, vascular risk factors, and 3-month functional outcomes.
Interpretation
Our deep learning algorithm offers a reproducible method for LA segmentation on CT, bridging the gap between CT and MRI assessments in patients with ischemic stroke.
期刊介绍:
Brain and Behavior is supported by other journals published by Wiley, including a number of society-owned journals. The journals listed below support Brain and Behavior and participate in the Manuscript Transfer Program by referring articles of suitable quality and offering authors the option to have their paper, with any peer review reports, automatically transferred to Brain and Behavior.
* [Acta Psychiatrica Scandinavica](https://publons.com/journal/1366/acta-psychiatrica-scandinavica)
* [Addiction Biology](https://publons.com/journal/1523/addiction-biology)
* [Aggressive Behavior](https://publons.com/journal/3611/aggressive-behavior)
* [Brain Pathology](https://publons.com/journal/1787/brain-pathology)
* [Child: Care, Health and Development](https://publons.com/journal/6111/child-care-health-and-development)
* [Criminal Behaviour and Mental Health](https://publons.com/journal/3839/criminal-behaviour-and-mental-health)
* [Depression and Anxiety](https://publons.com/journal/1528/depression-and-anxiety)
* Developmental Neurobiology
* [Developmental Science](https://publons.com/journal/1069/developmental-science)
* [European Journal of Neuroscience](https://publons.com/journal/1441/european-journal-of-neuroscience)
* [Genes, Brain and Behavior](https://publons.com/journal/1635/genes-brain-and-behavior)
* [GLIA](https://publons.com/journal/1287/glia)
* [Hippocampus](https://publons.com/journal/1056/hippocampus)
* [Human Brain Mapping](https://publons.com/journal/500/human-brain-mapping)
* [Journal for the Theory of Social Behaviour](https://publons.com/journal/7330/journal-for-the-theory-of-social-behaviour)
* [Journal of Comparative Neurology](https://publons.com/journal/1306/journal-of-comparative-neurology)
* [Journal of Neuroimaging](https://publons.com/journal/6379/journal-of-neuroimaging)
* [Journal of Neuroscience Research](https://publons.com/journal/2778/journal-of-neuroscience-research)
* [Journal of Organizational Behavior](https://publons.com/journal/1123/journal-of-organizational-behavior)
* [Journal of the Peripheral Nervous System](https://publons.com/journal/3929/journal-of-the-peripheral-nervous-system)
* [Muscle & Nerve](https://publons.com/journal/4448/muscle-and-nerve)
* [Neural Pathology and Applied Neurobiology](https://publons.com/journal/2401/neuropathology-and-applied-neurobiology)