2353

Prediction of early stage of AD based on functional connectivity network characteristics: an fMRI-based study
Zhizheng Zhuo1,2, Zhuqing Long1, Bin Jing1, Xiangyu Ma1, Han Liu1, Jianxin Dong1, Xiao Mo1, Qi Yan1, and Haiyun Li1

1Bio-medical Engineering, Capital Medical University, Beijing, People's Republic of China, 2Clinical Science, Philips Healthcare, Beijing, People's Republic of China

Synopsis

MCI (Mild Cognitive Impairment) is a pre-stage of Alzheimer’s Disease (AD). And the early detection of MCI is important for early treatment of AD patients. In this work, prediction efficiency of early stage of AD based on the functional connectivity network characteristics was evaluated by using a couple of classifiers with AAL_90 and AAL_1024 templates. The results showed that brain functional characteristics were effective in the prediction of MCI with a SVM-based classifier. And a more fine template could improve prediction accuracy.

Purpose

MCI (Mild Cognitive Impairment) is a pre-stage of Alzheimer’s Disease (AD). And the early detection of MCI is important for early treatment of AD patients. MR techniques such as structural MRI and functional MRI (fMRI) have been proved to be available for the differentiation of AD, MCI and normal situation1. In this work, prediction efficiency of early stage of AD based on the functional connectivity network characteristics was evaluated by using a couple of classifiers with AAL_90 and AAL_1024 templates.

Methods

Bold-based fMRI and 3D T1w structural MRI images of 32 MCI patients and 35 normal controls were obtained from the ADNI (website: http://adni.loni.usc.edu/ ). All these subjects are age and sex matched. Pre-processing procedures including slice timing, head motion correction and nuisance covariates regression (realign parameter, white matter and CSF signals), filtering (0.01-0.1Hz), normalization by using 3D T1w structure MRI images and DARTEL, smooth by DARTEL were carried out based on DPARSFA software. And then the time series within each region based on AAL_90 and AAL_1024 templates were extracted and the corresponding connectivity matrix were calculated. All the matrices were threshold by a PSW method (PSW=0.25) and converted into binary matrices for network parameters calculation. For both AAL_90 and AAL_1024 template-based methods, the network parameters (shown in table 1) including 5 global and 9 local parameters were extracted from the binary connectivity matrix for each subject2. And there are totally thousands of features for each subject. So before the classification, a fisher-score method was performed for features reduction and finally the first leading 30 features with high fisher-scores were selected for the next classification. Classifies including Support Vector Machine (SVM), Artificial Neural Network (ANN), Naïve Bayes, Linear Discriminate Analysis (LDA) and Random forest were evaluated and compared.

Results

The extracted functional connectivity network parameters were shown in Table 1 and the classification results were summarized and compared in Table 2.

Discussion

Different template-based methods showed different classification efficiencies for a specific classifier. The AAL_1024 template-based technique will build a more fine connectivity network and extract more features to character the brain functional topological structure than the AAL_90 template-based one and so the classification by AAL_1024 template method is better. Different classifiers also showed different classification efficiencies for a specific ALL template-based technique. For AAL_1024 template-based method, SVM showed the best classification efficiency with a high accuracy (95.52%), precision (96.77% and 94.44% for MCI and NC) and recall (93.75% and 97.14% for MCI and NC), which indicate that the SVM was a proper classifier for predicting MCI from NC. It may assist the doctors to early diagnose the MCI patients in addition to clinical evaluation.

Conclusion

Brain functional characteristics were effective in the prediction of MCI with a SVM-based classifier. And a more fine template could improve prediction accuracy. For both doctors and patients, this will be very important for the early detection of MCI and corresponding treatment strategy.

Acknowledgements

No acknowledgement found.

References

1. Yu S, Qinhua Y, Rong F, et al. Disrupted functional brain connectivity and its association to structural connectivity in amnestic mild cognitive impairment and Alzheimer's disease. Plos one, 2014; 9(5). Doi: 10.1371/journal.pone.0096505.

2. Mikail R, Olaf S. Complex network measures of brain connectivity: Uses and interpretations. NeuroImage, 2010; 52: 1059-1069.

Figures

Table1. The network parameters extracted by using BCT (Brain Connectivity Toolbox)

Table 2. The classification results of MCI and NC based on different classifiers

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)
2353