NeuroBOLT: Resting-state EEG-to-fMRI Synthesis with Multi-dimensional Feature Mapping

Authors: Yamin Li, Ange Lou, Ziyuan Xu, Shengchao Zhang, Shiyu Wang, Dario Englot, Soheil Kolouri, Daniel Moyer, Roza Bayrak, Catie Chang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments demonstrate that Neuro BOLT effectively reconstructs unseen resting-state f MRI signals from primary sensory, high-level cognitive areas, and deep subcortical brain regions, achieving state-of-the-art accuracy with the potential to generalize across varying conditions and sites, which significantly advances the integration of these two modalities.3 Experiments
Researcher Affiliation Academia Yamin Li1 Ange Lou1 Ziyuan Xu1 Shengchao Zhang1 Shiyu Wang1 Dario J. Englot2 Soheil Kolouri1 Daniel Moyer1 Roza G. Bayrak1 Catie Chang1 1Vanderbilt University 2Vanderbilt University Medical Center yamin.li@vanderbilt.edu
Pseudocode No The paper describes its model architecture and process through text and diagrams (Figure 1, 2, 3), but it does not include a formally labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code Yes Project page: https://soupeeli.github.io/Neuro BOLT
Open Datasets No The paper states in Appendix D that the data 'can be made public upon request', indicating it is not openly and readily accessible via a specific link, DOI, or repository at the time of publication.
Dataset Splits Yes For subject-specific prediction, where training and testing occur on the same scan, we split the scan in an 8:1:1 ratio for training, validation, and testing, i.e., training on the first 80% of the data and testing on the last 10%. ... For the inter-subject analysis in resting-state data, we randomly divided the datasets into training/validation/testing sets by approximately 3:1:1 (18 scans : 5 scans : 6 scans).
Hardware Specification Yes Experiments are conducted on a single RTX A5000 GPU using Python 3.9.12, Pytorch 2.0.1, and CUDA 11.7.
Software Dependencies Yes Experiments are conducted on a single RTX A5000 GPU using Python 3.9.12, Pytorch 2.0.1, and CUDA 11.7.
Experiment Setup Yes The batch sizes are set at 16 and 64 for intra-subject and inter-subject analyses, respectively. Adam W is utilized as our optimizer, and MSE as our training objective. The initial learning rate is set at 3e-4 with a weight decay of 0.05, and a minimal learning rate of 1e-6. Table 5: Hyperparameters for Neuro BOLT