BrainNetMLP

   

---

🏆 BrainNetMLP is honored with the -Best Paper Award- at the 1st EMA4MICCAI workshop.

📢 BrainNetMLP is accepted for oral presentation at the 1st MICCAI Workshop on Efficient Medical AI (EMA4MICCAI).

In this project, we build a very simple yet efficient and effective pure MLP-based model for functional brain network classification, which achieves comparative performance for Autism Spectral Disorder (ASD) and gender classification.

Pipeline

We hope our approach can act as a simple yet efficient and effective baseline for the task of functional brain network classification.

Pseudo Code

"""
BrainNetMLP: An efficient and effective baseline

Args:
    X: the brain connectivity matrix (NxN)
    T: the processed BOLD signal (NxT)
    k: the number of low-frequency components to keep
"""
def BrainNetMLP(X, T, k):
  F = torch.fft.rfft(T)
  F = torch.abs(F)[:, 1:k] # apply low-pass filter
  X = uptriu(X).flatten() # get valid values 
  X, F = MLP_1(X), MLP_2(F) 
  y = MLP_3([X,F]) # make prediction based on spectral and spatial features
  return y

Usage

Here, we provide the guideline for running the training and evaluation code on ABIDE dataset:

1. Prepare the data and save it at path: full_data.
2. The format of dataset is .npy by default (or you can download ABIDE dataset here).
3. Run the following command for training and evaluation.

python train.py -d abide

For training and inference on other datasets, you can prepare the dataset like the format of ABIDE dataset.

Acknowledgement

Thanks for the work of BrainNetTF, ComBrainTF, STGCN, GBT and so on, for the open-source spirit!

Citation

If you find this project is helpful, please star the repository and cite our paper:

@article{hou2025brainnetmlp,
  title={BrainNetMLP: An Efficient and Effective Baseline for Functional Brain Network Classification},
  author={Hou, Jiacheng and Song, Zhenjie and Kuruoglu, Ercan Engin},
  journal={arXiv preprint arXiv:2505.11538},
  year={2025}
}