Table 3.
Mean plus or minus standard deviation for eloquent class true positive rate (TPR), specificity, F1 and AUC for the HCP cohort (100 subjects). The final column shows the FDR corrected p-values for the associated t-scores where we compare AUC between our method against each baseline.
Task | Method | Eloquent TPR | Specificity | F1 | AUC | t-score | p-value |
---|---|---|---|---|---|---|---|
Language | MTGNN | 0.67 ± 0.013 | 0.62 ± 0.012 | 0.63 ± 0.014 | 0.68 ± 0.01 | ||
FCNN | 0.59 ± 0.022 | 0.56 ± 0.021 | 0.58 ± 0.019 | 0.62 ± 0.018 | 14.08 | 3.5 × 10−44 | |
RF | 0.32 ± 0.036 | 0.61 ± 0.026 | 0.45 ± 0.013 | 0.52 ± 0.034 | 17.02 | 1.8 × 10−64 | |
SVM | 0.36 ± 0.026 | 0.49 ± 0.024 | 0.39 ± 0.018 | 0.51 ± 0.016 | 34.68 | 2.7 × 10−262 | |
Finger | MTGNN | 0.78 ± 0.011 | 0.75 ± 0.013 | 0.77 ± 0.014 | 0.82 ± 0.008 | ||
FCNN | 0.75 ± 0.014 | 0.69 ± 0.016 | 0.71 ± 0.015 | 0.73 ± 0.011 | 17.84 | 3.1 × 10−70 | |
RF | 0.41 ± 0.026 | 0.71 ± 0.022 | 0.54 ± 0.023 | 0.58 ± 0.028 | 27.61 | 1.2 × 10−166 | |
SVM | 0.41 ± 0.024 | 0.55 ± 0.028 | 0.42 ± 0.025 | 0.52 ± 0.015 | 52.66 | ≈ 0 | |
Foot | MTGNN | 0.83 ± 0.009 | 0.82 ± 0.008 | 0.8 ± 0.011 | 0.79 ± 0.009 | ||
FC-NN | 0.73 ± 0.016 | 0.65 ± 0.017 | 0.66 ± 0.013 | 0.71 ± 0.015 | 21.45 | 4.9 × 10−101 | |
RF | 0.42 ± 0.025 | 0.74 ± 0.026 | 0.46 ± 0.021 | 0.58 ± 0.029 | 14.32 | 1.2 × 10−45 | |
SVM | 0.50 ± 0.031 | 0.53 ± 0.028 | 0.48 ± 0.027 | 0.51 ± 0.013 | 65.72 | ≈ 0 | |
Tongue | MTGNN | 0.80 ± 0.01 | 0.78 ± 0.009 | 0.77 ± 0.011 | 0.78 ± 0.009 | ||
FC-NN | 0.76 ± 0.012 | 0.72 ± 0.014 | 0.73 ± 0.016 | 0.73 ± 0.015 | 7.63 | 9.1 × 10−14 | |
RF | 0.44 ± 0.03 | 0.69 ± 0.032 | 0.50 ± 0.026 | 0.57 ± 0.032 | 23.73 | 2.1 × 10−123 | |
SVM | 0.55 ± 0.023 | 0.52 ± 0.025 | 0.48 ± 0.024 | 0.53 ± 0.014 | 49.61 | ≈ 0 |