Reproducible Abnormalities and Diagnostic Generalizability of White Matter in Alzheimer's Disease

Abstract

Alzheimer's disease (AD) is associated with the impairment of white matter (WM) tracts. The current study aimed to verify the utility of WM as the neuroimaging marker of AD with multisite diffusion tensor imaging datasets [321 patients with AD, 265 patients with mild cognitive impairment (MCI), 279 normal controls (NC)], a unified pipeline, and independent site cross-validation. Automated fiber quantification was used to extract diffusion profiles along tracts. Random-effects meta-analyses showed a reproducible degeneration pattern in which fractional anisotropy significantly decreased in the AD and MCI groups compared with NC. Machine learning models using tract-based features showed good generalizability among independent site cross-validation. The diffusion metrics of the altered regions and the AD probability predicted by the models were highly correlated with cognitive ability in the AD and MCI groups. We highlighted the reproducibility and generalizability of the degeneration pattern of WM tracts in AD.

Keywords: Alzheimer’s disease; Cross-validation; Diffusion tensor imaging; White matter tracts.

Fig. 1

Schematic of the data analysis pipeline. A The DTI images process pipeline. All images from seven sites are first preprocessed by the dtiInit pipeline, including eddy current and motion correction. Then the AFQ toolbox is used to track 20 fiber tracts and to extract pointwise profiles (FA, MD, RD, and AxD of 100 points along each tract) for each subject. B Pointwise meta-analysis is applied to each fiber tract and diffusion metric to detect the difference between the AD and NC groups. The random-effect model and Cohen's d are used to measure the effect size. C The machine learning algorithm is applied to conduct prediction tasks of individual diagnostic status using the statistically significant tract-based features (AD vs NC binary classification). The leave-one-site-out cross-validation strategy is used to evaluate the generalizability of classifiers. Briefly, one site is selected as the testing set for the outer loop, and the other six sites are used as the training set to perform feature selection and model training for the inner loop. D Correlation analyses are applied to the diffusion metrics and MMSE scores, as well as classifier output and MMSE scores. Brain regions that are affected by the altered tracts are also investigated.

Fig. 2

The pointwise alteration pattern of FA in the AD group. FA values along fibers and the correlation between the FA of the regions with significant differences and MMSE scores are determined. Forest plots show the meta-analysis of the average FA values of essential locations along the fiber tract in the AD in comparison to NC groups (Pearson correlation, n = 549 (294 AD + 255 MCI), **P <0.0001, *P <0.005). The left thalamic radiation is also significantly different, but these data are not shown because of the instability and insignificant correlation between FA and MMSE scores.

Fig. 3

The reproducible abnormal pattern of white matter tracts in AD. We applied Pearson correlations among the effect sizes of points along 18 fiber tracts of 7 times leave-one-site-out meta-analyses for each metric. The colors show the values of the correlation coefficients (r values). All of the correlation coefficients are >0.55 (P <0.0001). A high correlation indicates high reproducibility of the abnormal pattern of white matter fiber tracts in AD.

Fig. 4

Brain regions that are affected by the impaired tracts in the AD (A) and MCI (B) groups. A Meta-analyses show the brain regions affected by the fiber tracts with significantly decreased FA in the AD group compared to the NC group. B Meta-analyses showing the brain regions affected by the fiber tracts with significantly reduced FA in the MCI group compared to the NC group. Colorbar: the ratio of subjects whose regions were associated with impaired WM fibers within the AD or MCI group.

Fig. 5

Individual diagnostic status prediction. Based on FA features, the ROC curves of the seven cross-validation loops are shown in A. The correlation between the AD-status probability (output of the classifiers) and MMSE in the MCI and AD groups are shown in B. **P <0.0001, *P <0.005.