基于全脑影像组学的阿尔茨海默病诊断研究

Diagnosis of Alzheimer's disease based on whole-brain radiomics

摘要:

目的： 探讨构建影像组学、临床和联合模型，对正常认知（cognitively normal，CN）组、轻度认知障碍（mild cognitive impairment，MCI）组和阿尔茨海默病（Alzheimer’s disease，AD）组的分类价值。方法： 选取阿尔茨海默病神经影像学倡议（Alzheimer’s Disease Neuroimaging Initiative，ADNI）数据库中139例CN、162 例MCI和128 例AD患者基线的临床和影像资料。以7∶3的比例随机分为训练集和验证集。基于3D-T1WI磁共振成像（magnetic resonance imaging，MRI）提取影像组学特征。在训练集中，使用套索回归算法（least absolute shrinkage and selection operator，LASSO）筛选组学特征，并通过多因素逻辑回归建立基于全脑皮层及皮层下核团的影像组学模型。使用单因素逻辑回归和多因素逻辑回归获得与分类相关的临床指标，并通过多因素逻辑回归模型建立临床模型和基于影像组学特征和临床指标的联合模型。用受试者工作特征（receiver operating characteristics，ROC）曲线评价分类模型的效能。结果： 影像组学模型在CN vs. AD、MCI vs. AD和CN vs. MCI的曲线下面积（area under the curve，AUC）分别为0.975、0.957和0.929，临床模型的AUC分别为0.839、0.675和0.685。将临床和影像组学特征相结合的联合模型的AUC分别为0.983、0.959和0.950。结论： 基于全脑皮质和皮质下核团的影像组学模型和联合模型可以准确地对CN、MCI和AD进行分类，联合模型是最佳分类模型。

Abstract:

Objective: To analyze the value of radiomics，clinical and combined models for the classification of the cognitively normal（CN）group，mild cognitive impairment（MCI） group and Alzheimer’s disease（AD） group. Methods: Data of 139 CN cases，162 MCI patients and 128 AD patients with the clinical and imaging information were collected from Alzheimer’s Disease Neuroimaging Initiative（ADNI） database. Patients of every group were randomly divided into the training cohorts and validation cohorts with a rate of 7∶3. Radiomics features were extracted from the regions covering the cortex and subcortical nuclei based on 3D-T1WI MRI. Least absolute shrinkage and selection operator（LASSO） was used to select features and multivariate logistic regression was used to develop radiomics models based on the whole-brain cortex and subcortical nuclei. Clinical features were identified by univariate and multivariate logistic regression. The clinical model and the combined model fusing the radiomics features with the clinical risk factors were developed by multivariate logistic regression model. The receiver operating characteristics（ROC） curve was used to evaluate the performance of the classification models. Results: The area under the curve（AUC） of radiomics model in CN vs. AD，MCI vs. AD and CN vs. MCI were 0.975，0.957 and 0.929，respectively. The AUC of clinical model in the three groups were 0.839，0.675 and 0.685 respectively. The AUC of combined model fusing radiomics features and clinical variables in the three groups were 0.983，0.959 and 0.950，respectively. Conclusion: The classification and diagnosis of AD could be accurately conducted using the radiomics model based on the whole-brain cortex and subcortical nuclei and combined model. The combined model is the best recommended model to classify CN，MCI and AD.