The diagnosis of Autism Spectrum Disorder (ASD) using resting-state functional Magnetic Resonance Imaging (rs-fMRI) is typically conducted in the time domain. However, frequency-based approaches (e.g., Fast Fourier Transform, FFT) for fMRI diagnosis offer advantages, such as capturing oscillatory patterns in brain connectivity. Despite these advantages, most studies focus only on the magnitude or real component and rely on FFT-based methods, which are sensitive to low signal-to-noise ratios.
To address these limitations, we propose BrainWaveNet, a wavelet-based Transformer that leverages the frequency domain and learns spatio-temporal information for rs-fMRI diagnosis. Specifically, BrainWaveNet captures intra- and inter-relationships between two different frequency-based features (real and imaginary parts) through self- and cross-attention mechanisms, enabling deeper exploration of ASD. Experiments on the ABIDE dataset demonstrate the superiority of BrainWaveNet compared to other deep learning methods, with further analysis identifying significant brain regions for neurological interpretation.
It is an honor to present my paper at MICCAI 2024, particularly as it will be my first conference experience. I am grateful for the opportunity to share my research on such a prestigious stage and engage with researchers in the field.
While I am particularly interested in frequency-domain research due to the focus of my paper, I am eager to explore a broad range of topics and learn about advancements across various fields represented at the conference.
I am also looking forward to engaging with esteemed researchers who share a passion for this field. It will be a valuable experience to gain insights from their work, witness the depth and breadth of ongoing research, and expand my knowledge through thoughtful discussions.