Table 6 Comparison with some of the papers that used machine learning approaches for rTMS treatment outcome prediction against methods and results. R and NR stand for responders and non-responders.
study | Number of subjects | Methods | Performance |
|---|---|---|---|
Erguzel10—2015 | 147 MDD (90 R/57 NR) | Cordance feature + ANN | ACC = 89.09% |
Bailey12—2018 | 39 MDD (10 R/29 NR) | Working memory-related fronto-midline theta power and Theta connectivity + SVM | ACC = 91% |
Erguzel43—2016 | 147 MDD (90 R/57 NR) | Cordance + ANN, SVM, Decision Tree | AUC = 0.92 |
Shalbaf6—2018 | 51 MDD (30 R/21 NR) | Permutation entropy + ANOVA | AUC = 0.8 |
Hasanzadeh9—2019 | 46 MDD (23 R/23 NR) | Correlation dimension, fractal dimension, power and Bispectrum + KNN | ACC = 93.5% |
Corlier13—2019 | 109 MDD (45% R/55% NR) | Functional connectivity + ElasticNet | AUC = 0.91 |
Bailey8—2021 | 193 MDD (128 R/65 NR) | Theta connectivity + ANOVA | No statistically difference |
Our work | 46 MDD (23 R/23 NR) | CWT images of frontal electrodes + TL-BLSTM-attention | ACC = 97.1% |
