A Multichannel 2D Convolutional Neural Network Model for Task-Evoked fMRI Data Classification

Table 1

Classification results for all deep learning models on data of 995 subjects (mean ± std).

Model	Accuracy (%)	Precision (%)	F1-score
PCA + SVM	48.94 ± 2.36	48.17 ± 2.48	0.4779 ± 0.0232
mv2D CNN	63.36 ± 2.19	63.59 ± 2.27	0.6306 ± 0.0222
3D CNN	82.34 ± 1.27	82.68 ± 1.39	0.8239 ± 0.0130
3D SepConv	80.44 ± 1.16	80.88 ± 1.24	0.8043 ± 0.0116
1D CNN	80.76 ± 1.69	80.94 ± 1.73	0.8068 ± 0.0178
s2D CNN	81.80 ± 0.89	81.95 ± 0.97	0.8179 ± 0.0094
M2D CNN	83.20 ± 2.29	83.63 ± 1.87	0.8321 ± 0.0223

Note: accuracy by chance is 20% (i.e., given 5 types of movement behavior).