Brain arteriolosclerosis, one of the main pathologies of cerebral small vessel disease, is common in older adults and has been linked to lower cognitive and motor function and higher odds of dementia. In spite of its frequency and associated morbidity, arteriolosclerosis can only be diagnosed at autopsy. Therefore, the purpose of this work was to develop an in-vivo classifier of arteriolosclerosis based on brain MRI. First, an ex-vivo classifier of arteriolosclerosis was developed based on features related to white matter hyperintensities, diffusion anisotropy and demographics by applying machine learning to ex-vivo MRI and pathology data from 119 participants of the Rush Memory and Aging Project (MAP) and Religious Orders Study (ROS), two longitudinal cohort studies of aging that recruit non-demented older adults. The ex-vivo classifier showed good performance in predicting the presence of arteriolosclerosis, with an average area under the receiver operating characteristic curve AUC = 0.78. The ex-vivo classifier was then translated to in-vivo based on available in-vivo and ex-vivo MRI data on the same participants. The in-vivo classifier was named ARTS (short for ARTerioloSclerosis), is fully automated, and provides a score linked to the likelihood a person suffers from arteriolosclerosis. The performance of ARTS in predicting the presence of arteriolosclerosis in-vivo was tested in a separate, 91% dementia-free group of 79 MAP/ROS participants and exhibited an AUC = 0.79 in persons with antemortem intervals shorter than 2.4 years. This level of performance in mostly non-demented older adults is notable considering that arteriolosclerosis can only be diagnosed at autopsy. The scan-rescan reproducibility of the ARTS score was excellent, with an intraclass correlation of 0.99, suggesting that application of ARTS in longitudinal studies may show high sensitivity in detecting small changes. Finally, higher ARTS scores in non-demented older adults were associated with greater decline in cognition two years after baseline MRI, especially in perceptual speed which has been linked to arteriolosclerosis and small vessel disease. This finding was shown in a separate group of 369 non-demented MAP/ROS participants and was validated in 72 non-demented Black participants of the Minority Aging Research Study (MARS) and also in 244 non-demented participants of the Alzheimer's Disease Neuroimaging Initiative 2 and 3. The results of this work suggest that ARTS may have broad implications in the advancement of diagnosis, prevention and treatment of arteriolosclerosis. ARTS is publicly available at https://www.nitrc.org/projects/arts/.

脑动脉硬化是脑小血管疾病的主要病理之一，在老年人中很常见，并且与认知和运动功能较低以及痴呆症发病率较高有关。尽管其频率和相关发病率很高，但动脉硬化只能通过尸检来诊断。因此，这项工作的目的是开发一种基于脑 MRI 的动脉硬化体内分类器。首先，通过将机器学习应用于来自 Rush Memory and Aging Project (MAP) 119 名参与者的离体 MRI 和病理数据，根据白质高信号、弥散各向异性和人口统计学相关特征，开发了动脉硬化的离体分类器宗教秩序研究（ROS），两项针对衰老的纵向队列研究，招募非痴呆老年人。离体分类器在预测动脉硬化的存在方面表现出良好的性能，受试者工作特征曲线下的平均面积 AUC = 0.78。然后，根据同一参与者可用的体内和体外 MRI 数据，将体外分类器转换为体内分类器。体内分类器被命名为 ARTS（ARTerioloSclerosis 的缩写），是完全自动化的，并提供与一个人患有动脉硬化的可能性相关的分数。ARTS 在预测体内动脉硬化存在方面的表现在一个由 79 名 MAP/ROS 参与者组成的独立组（91% 无痴呆）中进行了测试，在生前间隔短于 2.4 年的人中，AUC = 0.79。考虑到动脉硬化只能通过尸检才能诊断，大多数非痴呆老年人的这种表现水平值得注意。ARTS 评分的扫描重复性非常好，组内相关性为 0.99，表明在纵向研究中应用 ARTS 可能在检测微小变化方面表现出高灵敏度。最后，非痴呆老年人中较高的 ARTS 分数与基线 MRI 两年后认知能力的更大下降相关，尤其是与动脉硬化和小血管疾病相关的知觉速度。这一发现在由 369 名非痴呆 MAP/ROS 参与者组成的单独小组中得到了证实，并在少数族裔老龄化研究 (MARS) 的 72 名非痴呆黑人参与者以及阿尔茨海默病神经影像计划的 244 名非痴呆参与者中得到了验证2 和 3。这项工作的结果表明，ARTS 可能对动脉硬化的诊断、预防和治疗的进展具有广泛的影响。ARTS 可在 https://www.nitrc.org/projects/arts/ 上公开获取。