Table 4 Performance of TL, TL-BLSTM, and TL-BLSTM-attention models based on EfficientNetB0 for different brain regions on unseen test data. The mean and standard deviation of each performance measure is presented.

From: A convolutional recurrent neural network with attention for response prediction to repetitive transcranial magnetic stimulation in major depressive disorder

  

Frontal

Central

Temporal

Parietal

Occipital

EfficientNetB0

ACC (%)

91.2 ± 3.7

87.6 ± 3.5

88.8 ± 4.0

89.3 ± 3.2

91.2 ± 4.1

SEN (%)

91.2 ± 3.5

89.1 ± 2.8

90.9 ± 3.4

90.7 ± 3.7

91.9 ± 3.1

SPE (%)

89.2 ± 3.4

86.5 ± 2.9

86.0 ± 3.0

88.4 ± 4.0

89.7 ± 3.2

AUC

0.90 ± 0.05

0.88 ± 0.05

0.88 ± 0.04

0.88 ± 0.04

0.90 ± 0.02

EfficientNetB0-BLSTM

ACC (%)

96.2 ± 2.3

94.2 ± 1.8

94.5 ± 2.1

94.7 ± 2.6

95.8 ± 2.8

SEN (%)

96.7 ± 2.2

94.5 ± 2.1

94.8 ± 2.0

95.5 ± 2.7

96.3 ± 3.4

SPE (%)

95.8 ± 3.1

93.5 ± 2.8

93.2 ± 2.9

94.3 ± 2.1

95.0 ± 2.3

AUC

0.95 ± 0.03

0.93 ± 0.04

0.94 ± 0.04

0.95 ± 0.03

0.96 ± 0.03

EfficientNetB0-BLSTM-Attention

ACC (%)

97.1 ± 1.1

95.6 ± 0.9

95.3 ± 1.0

95.7 ± 1.4

96.3 ± 0.7

SEN (%)

97.3 ± 0.8

96.0 ± 1.2

95.9 ± 1.2

96.5 ± 1.8

97.1 ± 1.0

SPE (%)

97.0 ± 1.2

95.5 ± 2.2

94.9 ± 1.5

95.0 ± 1.6

96.8 ± 0.9

AUC

0.96 ± 0.01

0.95 ± 0.02

0.95 ± 0.03

0.96 ± 0.03

0.96 ± 0.02

  1. Significant values are in bold.