Table 3.
Classification performance scores obtained using data, respectively, from the DTI, GT fMRI, and DTI + GT data sets.
| ACC (%) | SENS (%) | SPEC (%) | PREC (%) | NPV (%) | |
|---|---|---|---|---|---|
| DTI DATASET | |||||
| SVMRBF | 79.75 (66,89) | 68.00 (54,79) | 91.50 (80,97) | 88.89 (76,95) | 88.89 (77,95) |
| SVMMLP | 73.00 (59,84) | 63.00 (49,75) | 83.00 (70,92) | 78.75 (65,88) | 79.00 (65,88) |
| MLP | 75.00 (61,85) | 74.00 (60,85) | 76.00 (62,86) | 75.51 (62,86) | 74.50 (60,85) |
| RBFN | 60.25 (46,73) | 65.00 (51,77) | 55.50 (42,69) | 59.36 (45,72) | 38.34 (25,52) |
| ANFIS | 83.50 (71,92) | 81.00 (68,90) | 86.00 (74,94) | 85.26 (73,93) | 79.90 (66,89) |
| GT fMRI DATASET | |||||
| SVMRBF | 81.00 (68-90) | 93.50 (83,99) | 68.50 (55,80) | 74.80 (61,85) | 74.80 (61,85) |
| SVMMLP | 78.25 (64-88) | 81.00 (68,90) | 75.50 (62,86) | 76.78 (63,86) | 76.78 (63,87) |
| MLP | 58.25 (44,71) | 55.50 (42,69) | 61.00 (47,74) | 58.73 (45,72) | 57.81 (43,70) |
| RBFN | 55.75 (42,69) | 55.50 (42,69) | 56.00 (42,69) | 55.78 (42,69) | 17.06 (8,29) |
| ANFIS | 82.75 (69,91) | 73.50 (64,88) | 92.00 (81,97) | 90.18 (78,96) | 73.60 (59,84) |
| DTI+GT DATASET | |||||
| SVMRBF | 84.75 (72,93) | 84.00 (71,92) | 85.50 (73,93) | 85.28 (73,93) | 85.28 (72,93) |
| SVMMLP | 74.75 (61,85 | 65.00 (51,77) | 84.50 (72,93) | 80.75 (68,90) | 80.75 (67,89) |
| MLP | 76.75 (63,87) | 74.00 (60,85) | 79.50 (66,89) | 78.31 (65,88) | 75.35 (61.85) |
| RBFN | 62.75 (49,75) | 86.50 (74,94) | 39.00 (27,53) | 58.64 (45,71) | 74.28 (60,84) |
| ANFIS | 85.25 (73,93) | 82.00 (69,91) | 88.50 (77,96) | 87.70 (76,95) | 81.69 (68,90) |
Each performance score is expressed as a percentage (%). For each score, the 95% IC is also reported in brackets. Classification performance scores (accuracy, sensitivity, specificity and AUC) for pairwise classifiers are expressed as percentage (%) with 95% confidence intervals in brackets. AUC, Area under receiver operating characteristic curve. ANFIS emerged as the most performant method in discriminating at baseline AD from VD independently of the data set (scores highlighted in bold).