White matter hyperintensity features differentially contribute to dementia subtypes as shown by machine learning
Year: 2023
Type: article
Study design: Observational Study
Abstract: Background White matter hyperintensities (WMH) are frequently observed on MRI scans of older people. Usually interpreted as a sign of cerebrovascular disease (CVD), they are also associated with an increased risk of cognitive impairment and dementia. WMH are prevalent in Alzheimer’s disease (AD) and in behavioral variant frontotemporal dementia (bvFTD). However, the total and spatial differential burden of WMH between these two pathologies is not well understood. Furthermore, the differential contributions of WMH across both conditions is unknown. Method Here we assessed the total and tract‐specific WMH burden features to classify the two dementia subtypes. We used a gradient boosting (GB) machine learning algorithm with a multi‐feature approach (demographic, neuropsychological, and magnetic resonance imaging (MRI) data). A total of 66 patients (43 AD, 23 bvFTD) and 66 controls (CN) from the ReDLat consortium underwent high‐resolution brain MRI and neuropsychological examination. WMH were extracted from Flair images using automated segmentation algorithms (LST) to perform analyses of total volume and white matter (WM) tract spatial distribution using the JHU atlas. As input data, three different approaches were tested: cognitive data only (MoCA and IFS scores), structural data only (total and tract‐specific WMH, from 20 different WM tracts) and mixed data (cognitive, structural, and demographic data (age, gender and years of education)). Then, we employed the GB algorithm to classify between AD vs CN, bvFTD vs CN, and AD vs bvFTD. Result The model using only total and tract‐specific WMH data yielded the highest area under the curve classification values between AD and bvFTD (0.92 (±0.8)), in comparison with only cognitive (0.86 (±0.1)) and mixed inputs (0.81 (±0.12)). Feature importance analysis revealed specific tract affectation for each dementia subtype. Demographic features were not relevant for the classification, discarding possible sample biases. Conclusion The model using only WMH data accurately identified dementia subtypes using readily available measures in patients with dementia. We gained biological insights into WMH pathophysiological profiles differentiating AD and bvFTD. If confirmed and replicated in future studies, this approach could be of complementary use for diagnosing and phenotyping dementias.
Source: Alzheimer s & Dementia
DOI: 10.1002/alz.075320
Authors: Cecilia Gonzalez Campo, Camila Sol Guardia, María Antonia Saracco, Matias Fraile‐Vazquez, Florencia Altschuler, Agustin Ibanez, Sebastian Moguilner
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4390200775