Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK.
Oxford Brain Diagnostics, Oxford Centre for Innovation, New Road, Oxford, OX1 1BY, UK.
Sci Rep. 2020 Jul 8;10(1):11237. doi: 10.1038/s41598-020-68118-8.
Fronto-temporal dementia (FTD) is a common type of presenile dementia, characterized by a heterogeneous clinical presentation that includes three main subtypes: behavioural-variant FTD, non-fluent/agrammatic variant primary progressive aphasia and semantic variant PPA. To better understand the FTD subtypes and develop more specific treatments, correct diagnosis is essential. This study aimed to test the discrimination power of a novel set of cortical Diffusion Tensor Imaging measures (DTI), on FTD subtypes. A total of 96 subjects with FTD and 84 healthy subjects (HS) were included in the study. A "selection cohort" was used to determine the set of features (measurements) and to use them to select the "best" machine learning classifier from a range of seven main models. The selected classifier was trained on a "training cohort" and tested on a third cohort ("test cohort"). The classifier was used to assess the classification power for binary (HS vs. FTD), and multiclass (HS and FTD subtypes) classification problems. In the binary classification, one of the new DTI features obtained the highest accuracy (85%) as a single feature, and when it was combined with other DTI features and two other common clinical measures (grey matter fraction and MMSE), obtained an accuracy of 88%. The new DTI features can distinguish between HS and FTD subgroups with an accuracy of 76%. These results suggest that DTI measures could support differential diagnosis in a clinical setting, potentially improve efficacy of new innovative drug treatments through effective patient selection, stratification and measurement of outcomes.
额颞叶痴呆（FTD）是一种常见的早发性痴呆症，其临床特征表现多样，包括三种主要亚型：行为变异型额颞叶痴呆、非流利/语法障碍型原发性进行性失语和语义变异型进行性失语。为了更好地理解 FTD 亚型并开发更具针对性的治疗方法，正确的诊断至关重要。本研究旨在测试一组新的皮质弥散张量成像（DTI）测量指标在 FTD 亚型中的区分能力。共有 96 名 FTD 患者和 84 名健康对照者（HS）纳入本研究。使用“选择队列”来确定特征（测量）集，并使用它们从七种主要模型中选择“最佳”机器学习分类器。选择的分类器在“训练队列”上进行训练，并在第三个队列（“测试队列”）上进行测试。该分类器用于评估二进制（HS 与 FTD）和多类（HS 和 FTD 亚型）分类问题的分类能力。在二进制分类中，一种新的 DTI 特征的准确性最高（85%），作为单一特征，当与其他 DTI 特征和另外两个常见的临床指标（灰质分数和 MMSE）结合使用时，准确性达到 88%。新的 DTI 特征可以区分 HS 和 FTD 亚组，准确率为 76%。这些结果表明，DTI 测量值可能支持临床环境中的鉴别诊断，通过有效选择患者、分层和测量结果，可能会提高新型创新药物治疗的疗效。
