ARTS

ARTS quantifies the likelihood of cerebral arteriolosclerosis from brain MRI (3D T1-weighted, T2-weighted FLAIR, and Diffusion Tensor Imaging) and demographic data (age at MRI scan, sex) by generating an ARTS score that correlates with arteriolosclerosis and related cognitive decline.

Key Features:

• ARTS score: Produces a continuous score that quantifies the probability of arteriolosclerosis presence from MRI-derived features and demographics.
• Input modalities: Uses 3D T1-weighted, T2-weighted FLAIR, and Diffusion Tensor Imaging (DTI) together with age and sex as model inputs.
• Imaging-derived features: Incorporates features related to white matter hyperintensities and diffusion anisotropy extracted from MRI.
• Training data: Trained on MRI and pathology datasets from 119 participants drawn from the Rush Memory and Aging Project (MAP) and the Religious Orders Study (ROS).
• Model development: Developed via an ex-vivo machine learning classifier that was adapted for in-vivo application.
• Performance (ex-vivo): The ex-vivo classifier achieved an average area under the receiver operating characteristic curve (AUC) of 0.78.
• Performance (in-vivo): The in-vivo ARTS classifier achieved an AUC of 0.79 on 79 MAP/ROS participants with antemortem intervals shorter than 2.4 years.
• Reproducibility: Demonstrated scan-rescan intraclass correlation coefficient (ICC) of 0.99, indicating high longitudinal sensitivity.
• Cognitive associations: Higher ARTS scores were associated with greater cognitive decline, particularly in perceptual speed, observed two years post-baseline MRI.
• Validation cohorts: Cognitive and predictive associations were validated in 369 non-demented MAP/ROS participants, 72 non-demented Black individuals from the Minority Aging Research Study (MARS), and 244 non-demented participants from ADNI phases 2 and 3.

Scientific Applications:

• In-vivo assessment of arteriolosclerosis: Provides a non-invasive in-vivo surrogate for arteriolosclerosis that can be compared to autopsy-confirmed pathology.
• Longitudinal monitoring: Enables detection of subtle cerebrovascular changes over time in longitudinal MRI studies due to high scan-rescan reproducibility.
• Risk stratification for cognitive decline: Facilitates stratification of individuals by arteriolosclerosis-related risk for cognitive decline and dementia in aging research cohorts.
• Cross-cohort validation: Supports application across diverse cohorts (MAP/ROS, MARS, ADNI2/3) for studies linking cerebrovascular pathology and cognition.

Methodology:

An ex-vivo machine learning classifier was trained on MRI-derived features (white matter hyperintensities, diffusion anisotropy) and demographic and pathology labels from MAP and ROS (n=119), the model was adapted for in-vivo MRI application, and performance was evaluated using AUC and intraclass correlation coefficient (ICC).