Table 2.
Rank order list with relative SVM weights for MRI, ApoE, Age, Sex and BMI in AD and MCI classification. Hippocampal volumes were the most influential feature for differentiating AD from controls, closely followed by ApoE genotype, which outperformed all the other MRI-derived markers. For classifying subjects as either MCI or controls, the exact same features were useful, in the same order of priority. This is somewhat in line with expectation, as hippocampal volume is so widely used and is perhaps the most well-validated MRI measure in AD studies. This rank order refers to a situation in which all measures are used jointly for classification. Also, the gray highlighted measures are the ones that, when used jointly, gave the best classification accuracy in our independent test datasets (see Figure 1 for ROC curves)
Biomarker |
||
---|---|---|
Rank | AD vs. control (weight / wi2) |
MCI vs. control (weight / wi2) |
1 | MRI Hipa 0.1664 |
MRI Hip 0.1045 |
2 | ApoE 0.1063 |
ApoE 0.0938 |
3 | Age 0.0369 |
Age 0.0188 |
4 | MRI Ventb 0.0349 |
MRI Vent 0.0103 |
5 | MRI Tempc 0.0210 |
MRI Temp 0.0045 |
6 | BMI 0.0147 |
BMI 0.0019 |
7 | Sex 0.0013 |
Sex 0.0009 |
Hippocampal volume summary
Ventricular volume summary
Temporal lobe summary from tensor-based morphometry (TBM)
Groups of biomarkers yielding the highest leave-one-out accuracy are highlighted.