M2DCNN is a repository of codes and experiment results for A Multi-channel 2D Convolutional Neural Networks Model for Task-evoked fMRI Data Classification, using Keras (version 2.2.4) with TensorFlow (version 1.12.0) as backend.
See the following publications for examples of this code in use:
- A Multi-channel 2D Convolutional Neural Networks Model for Task-evoked fMRI Data Classification. Jinlong Hu, Yuezhen Kuang, Bin Liao, Lijie Cao, Shoubin Dong, Ping Li, Computational Intelligence and Neuroscience, 2019.
M2D_CNN_model.py is the Python code of M2D CNN model.
cnn3d_model.py is the Python code of 3D CNN model.
sep3d_model.py is the Python code of 3D SepConv model. To run this model, you should import SeparableConv3D from sepconv3D.
s2D_CNN_model.py is the Python code of s2D CNN model.
mv2D_CNN_model.py is the Python code of mv2D CNN model.
cnn1d_model.py is the Python code of 1D CNN model.
svm_model.py is the Python code of SVM model.
9950 samples from 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 |
2000 samples from 200 subjects (mean±std):
| Model | Training time (S) | Total number of epochs |
|---|---|---|
| mv2D CNN | 909±134 | 54±8 |
| 3D CNN | 1156±185 | 39±6 |
| 3D SepConv | 1601±196 | 41±5 |
| 1D CNN | 834±157 | 39±7 |
| s2D CNN | 565±102 | 31±6 |
| M2D CNN | 1074±348 | 39±13 |
2000 samples from 200 subjects:
5000 samples from 500 subjects：
