Purpose
Alzheimer's disease (AD) is a progressive neurodegenerative disorder associated with damage of brain cells and brain shrinkage [1]. Automatic classification of AD or mild cognitive impairment (MCI) from cognitively normal (CN) subjects using magnetic resonance imaging (MRI) requires a good representation of brain features and trained classifiers.
Brain shrinkage features relating to AD can be assessed with MRI by using an atlas providing prior knowledge of the structure of the targeted brain regions.
It has been demonstrated that using a multi-atlas approach achieves better feature...
Methods and materials
We used T1w brain scans from collected by the Alzheimer's Disease Neuroimaging Initiative (ADNI) to construct and evaluate our classifiers (SVMs). Three classification tasks were considered: AD (n=195 subjects) vs. CN (n= 195 subjects); AD (n=195 subjects) vs. MCI (n=195 subjects); and CN (n=215 subjects) vs. MCI (n=215 subjects). The age range was 56-95 yrs. (mean 75.79) for the AD subjects, 55-93 yrs. (mean 75.09) for the MCI subjects, and 55-95 yrs. (mean 76.98) for the CN subjects. The MALF method provided ROIs at five...
Results
The overall highest performing classification model regarding accuracy, sensitivity, and specificity was achieved at granularity level 5 for the combined ROI and ROI-C features. This model had an accuracy/sensitivity/specificity of 86.6/85.7/87.4% for AD vs. CN, 71.9/73.1/70.7% for AD vs. MCI, and 67.3/63.3/71.3% for CN vs. MCI (Fig. 3). Fig. 3(d) shows the ranking of feature importance for the combined (ROI+ROI-C) regions in the AD vs. CN classification task.
Fig. 4 depicts the ROC curves and corresponding AUCs for the single set of ROIs (in blue),...
Conclusion
In summary, our framework showed good performance in distinguishing AD/MCI from CN. Moreover, the finer the granularity level, the better the classification performance. A major finding was that the correlative ROI-C features outperformed the single set ROI features across all granularity levels. Also, the ROC-AUC metric value was higher (marginally) for the ROI-C than for the ROI+ROI-C features, indicating the importance of the correlative features. Interestingly, the top three most important correlative features included (i) the correlation between CSF volume in the sulci across the...
Personal information and conflict of interest
S. Alam; Bergen, Hordaland, NORWAY/NO - Author at University of Bergen A. Lundervold; Bergen/NO - Author at University of Bergen A. S. Lundervold; Bergen/NO - Author at Western Norway University of Applied Sciences
References
[1] A. Collie and P. Maruff, “The neuropsychology of preclinical Alzheimer’s disease and mild cognitive impairment,” Neurosci. Biobehav. Rev., vol. 24, no. 3, pp. 365–374, May 2000.
[2] R. Min, G. Wu, J. Cheng, Q. Wang, and D. Shen, “Multi-Atlas Based Representations for Alzheimer’s Disease Diagnosis,” Hum. Brain Mapp., vol. 35, no. 10, pp. 5052–5070, Oct. 2014.
[3] X. Tang et al., “Segmentation of brain magnetic resonance images based on multi-atlas likelihood fusion: testing using data with a broad range of anatomical and photometric profiles,”...