INFLUENCE OF NETWORK CONSTRUCTION METHODS ON PATH LENGTH VALUES IN ALZHEIMER’S DISEASE: A MULTI-STUDY ANALYSIS OF MRI CONNECTIVITY STUDIES
Ellen Dicks,Yong He,Faisal Beg,Geert Jan Biessels,Cristian Carmeli,Guangyu Chen,Koray Çiftçi,Zhengjia Dai,Edwin Van Dellen,Martin Dyrba,Eric J. Friedman,Willem de Haan,Mahdi Jalili,Jaeseung Jeong,Yong Jeong,Tianzi Jiang,Maria G. Knyazeva,Dong Young Lee,Shi-Jiang Li,Yong Liu
DOI: https://doi.org/10.1016/j.jalz.2018.06.1703
2018-01-01
Alzheimer s & Dementia
Abstract:Brain connectivity is disrupted in Alzheimer's disease (AD). However, diverging results have been reported for the directionality of network measure differences between patients and controls (CN), especially for path length values. Here, we investigated the influence of network construction choices on path length differences between controls and AD subjects in an ongoing subject-level multi-study analysis of MRI connectivity studies on AD. We collected data from 35 studies (sMRI, DTI, fMRI, PET, MEG/EEG). Here, we present data for 8 studies for which preprocessing so far has been completed (2 sMRI, 6 rs-fMRI; see Fig.1 for study details), including 355 CN (mean age 73.6±6.8 years, 55.8% female) and 383 AD patients (mean age 73.4±7.7 years, 50.4 % female). We tested the influence of two types of thresholds on path length values: a commonly used connectivity density of 15% and the minimum connectivity density to keep the network fully connected. For each modality and type of threshold, we tested differences in path length values between CN and AD with linear mixed models including study where the data originated as random effect. When networks had the same connectivity density of 15%, AD subjects tended to show higher path length values in rs-fMRI (β±SE: 0.19±0.11; p=0.07; Fig.2A). Both sMRI studies showed lower path length values for AD (Fig.2A). However, group differences were not significant when combining these studies (β±SE: -0.18±0.12; p>0.05). When using the minimum connectivity density, the majority of the studies (6 of 8; Fig.2B) tended to show increased path length values in AD, but these differences did not reach significance (rs-fMRI: β±SE: 0.17±0.11, p>0.05; sMRI: β±SE: 0.06±0.11, p>0.05). Adding the number of nodes and/or age and sex as covariates did not change the results . Overview of included studies and network analysis. The path length was calculated after binarization of the respective network. Abbreviations: sMRI, structural MRI; rs-fMRI, resting state functional MRI. Characteristic path lengths per study after applying different connectivity density thresholds. (A) Characteristic path length of networks after enforcing a connectivity density of 15%. (B) Characteristic path length of fully connected networks. Path length values were transformed into z scores using the mean and standard deviation of the CN group of the respective study. Path length differences between AD and CN showed variability between studies, also within the same modality. Group differences were more consistent when applying a threshold that ensures a fully connected network. These results suggest that it may be possible to identify a unified approach to construct networks with more consistent group differences between studies and improve understanding how brain connectivity changes in AD.