BrainWaveNet: Wavelet-based Transformer for Autism Spectrum Disorder Diagnosis
Ah-Yeong Jeong*, Da-Woon Heo*, Eunsong Kang, and Heung-Il Suk†
*Equally contributed, †Corresponding authorAbstract. The diagnosis of Autism Spectrum Disorder (ASD) using resting-state functional Magnetic Resonance Imaging (rs-fMRI) is commonly analyzed through functional connectivity (FC) between Regions of Interest (ROIs) in the time domain. However, the time domain has limitations in capturing global information. To overcome this problem, we propose a wavelet-based Transformer, BrainWaveNet, that leverages the frequency domain and learns spatial-temporal information for rs-fMRI brain diagnosis. Specifically, BrainWaveNet learns inter-relations between two different frequency-based features (real and imaginary parts) by crossattention mechanisms, which allows for a deeper exploration of ASD. In our experiments using the ABIDE dataset, we validated the superiority of BrainWaveNet by comparing it with competing deep learning methods. Furthermore, we analyzed significant regions of ASD for neurological interpretation.
Dataset: ABIDE-I
Method | AUC | ACC (%) | SEN (%) | SPC (%) |
---|---|---|---|---|
BrainNetCNN | 0.6907 ± 0.01 | 64.36 ± 1.19 | 53.40 ± 3.79 | 75.32 ± 2.31 |
BrainNetTF | 0.7147 ± 0.02 | 66.80 ± 1.78 | 61.06 ± 4.96 | 72.55 ± 8.08 |
STAGIN (GARO) | 0.5886 ± 0.07 | 57.23 ± 5.69 | 57.02 ± 11.62 | 57.45 ± 7.41 |
STAGIN (SERO) | 0.5839 ± 0.05 | 62.66 ± 3.54 | 51.92 ± 2.18 | 61.28 ± 4.02 |
BoIT | 0.6989 ± 0.02 | 62.66 ± 3.54 | 55.32 ± 5.53 | 70.00 ± 3.31 |
BrainWaveNet (Ours) | 0.7388 ± 0.02 | 67.55 ± 2.04 | 66.49 ± 9.17 | 68.35 ± 9.80 |
Dataset: ADHD-200
Method | AUC | ACC (%) | SEN (%) | SPC (%) |
---|---|---|---|---|
BrainNetCNN | 0.6428 ± 0.01 | 62.06 ± 1.12 | 35.48 ± 4.56 | 82.36 ± 3.25 |
BrainNetTF | 0.6886 ± 0.03 | 63.71 ± 2.43 | 43.10 ± 13.71 | 79.45 ± 10.65 |
STAGIN (GARO) | 0.5304 ± 0.04 | 51.04 ± 5.20 | 46.35 ± 32.98 | 54.77 ± 30.53 |
STAGIN (SERO) | 0.5520 ± 0.04 | 57.18 ± 1.62 | 34.91 ± 20.96 | 74.28 ± 13.70 |
BoIT | 0.6748 ± 0.03 | 64.22 ± 3.04 | 49.52 ± 5.62 | 75.45 ± 1.11 |
BrainWaveNet (Ours) | 0.7330 ± 0.01 | 65.98 ± 2.28 | 59.52 ± 12.49 | 70.91 ± 7.47 |
- python 3.10.14
- pytorch 2.2.2
- torchvision 0.17.2
- torchaudio 2.2.2
- scikit-learn 1.4.1
- wandb 0.17.2
- hydra-core 1.3.2
@InProceedings{Jeo_BrainWaveNet_MICCAI2024,
author = { Jeong, Ah-Yeong and Heo, Da-Woon and Kang, Eunsong and Suk, Heung-Il},
title = { { BrainWaveNet: Wavelet-based Transformer for Autism Spectrum Disorder Diagnosis } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {pending}
}
This work was supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. RS-2019-II190079, Artificial Intelligence Graduate School Program (Korea University)) and the National Research Foundation of Korea (NRF) grant funded by the Korea government (No. 2022R1A4A1033856).