Toward Generalizing Visual Brain Decoding to Unseen Subjects

Abstract

Visual brain decoding aims to decode visual information from human brain activities. Despite the great progress, one critical limitation of current brain decoding research lies in the lack of generalization capability to unseen subjects. Prior works typically focus on decoding brain activity of individuals based on the observation that different subjects exhibit different brain activities, while it remains unclear whether brain decoding can be generalized to unseen subjects. This study aims to answer this question. We first consolidate an image-fMRI dataset consisting of stimulus-image and fMRI-response pairs, involving 177 subjects in the movie-viewing task of the Human Connectome Project (HCP). This dataset allows us to investigate the brain decoding performance with the increase of participants. We then present a learning paradigm that applies uniform processing across all subjects, instead of employing different network heads or tokenizers for individuals as in previous methods, which can accommodate a large number of subjects to explore the generalization capability across different subjects. A series of experiments are conducted and we have the following findings. First, the network exhibits clear generalization capabilities with the increase of training subjects. Second, the generalization capability is common to popular network architectures (MLP, CNN and Transformer). Third, the generalization performance is affected by the similarity between subjects. Our findings reveal the inherent similarities in brain activities across individuals. With the emerging of larger and more comprehensive datasets, it is possible to train a brain decoding foundation model in the future. Codes and models can be found at https://github.com/Xiangtaokong/TGBD.

Figures (5)

• Figure 1: The performance on unseen subjects with the increase of the number of training subjects.
• Figure 2: Dataset reconstruction from the HCP data. We extract the last frame $i$ in each second of the movie clip as the stimulus image, and average the fMRI voxels in the subsequent 4 seconds (due to hemodynamic delay) as the corresponding neural response $v$ to obtain image-fMRI pairs.
• Figure 3: The visualization of scanned brain data in NSD dataset. The highlighted regions indicate the manually labeled NSDGeneral data. Compared to the whole brain, the NSDGeneral regions show significant variations across different subjects.
• Figure 4: The overview of our learning pipeline (left) and visual brain decoding network (right).
• Figure A.1: The visualized image retrieval results. 'Target images' refer to those viewed by Subj 1 from the HCP dataset, while 'Retrieved images' represent the corresponding retrieval outputs from the visual brain decoding model trained on 167 subjects. The retrieved images are ranked from left to right based on retrieval similarity.