**MindBridge: A Cross-Subject Brain Decoding Framework** **Shizun Wang, Songhua Liu, Zhenxiong Tan, Xinchao Wang** **National University of Singapore** **{shizun.wang, songhua.liu, zhenxiong}@u.nus.edu, xinchao@nus.edu.sg** **Abstract:** Brain decoding, a critical field in neuroscience, aims to reconstruct stimuli from brain signals, primarily using functional magnetic resonance imaging (fMRI). Current methods are confined to a per-subject-per-model paradigm, limiting their applicability. This paper introduces MindBridge, a novel framework that achieves cross-subject brain decoding using a single model. MindBridge addresses challenges such as input dimension variability, unique neural patterns, and limited data for new subjects. It employs an adaptive aggregation function and a cyclic fMRI reconstruction mechanism for subject-invariant representation learning. The framework also includes a reset-tuning method for adapting to new subjects. Experimental results demonstrate MindBridge's ability to reconstruct images for multiple subjects, outperforming dedicated subject-specific models, especially with limited data. This advancement suggests promising directions for broader applications in neuroscience and more efficient utilization of fMRI data. **Introduction:** Brain decoding aims to reconstruct image stimuli from brain signals, a field with significant challenges, including input dimension variability, diverse neural responses, and data scarcity for new subjects. MindBridge addresses these challenges by employing an adaptive aggregation function, a cyclic fMRI reconstruction mechanism, and a reset-tuning strategy. The framework enables cross-subject brain decoding and novel fMRI synthesis, demonstrating its effectiveness and adaptability in experiments with the NSD dataset. **Methods:** MindBridge uses an adaptive max pooling function to unify fMRI signal dimensions, a cyclic fMRI reconstruction mechanism for subject-invariant representation learning, and a reset-tuning strategy for new subject adaptation. The framework is trained using a combination of SoftCLIP and MSE losses to ensure accurate CLIP embeddings. **Results:** MindBridge achieves comparable performance to state-of-the-art methods with a single model, demonstrating its effectiveness in cross-subject brain decoding. It also shows superior performance in new subject adaptation with limited data, highlighting the benefits of transferable knowledge from cross-subject pretraining. **Discussion:** MindBridge's success in cross-subject brain decoding and novel fMRI synthesis opens new avenues for practical applications in neuroscience. However, limitations include the need for a larger and more diverse dataset for generalizability and the challenge of preserving spatial relationships in serialized fMRI signals. **Conclusion:** MindBridge is a novel cross-subject brain decoding framework that overcomes the per-subject-per-model paradigm, enhancing decoding accuracy and enabling novel fMRI synthesis. Its effectiveness and adaptability make it a promising tool for advancing brain decoding technology.**MindBridge: A Cross-Subject Brain Decoding Framework** **Shizun Wang, Songhua Liu, Zhenxiong Tan, Xinchao Wang** **National University of Singapore** **{shizun.wang, songhua.liu, zhenxiong}@u.nus.edu, xinchao@nus.edu.sg** **Abstract:** Brain decoding, a critical field in neuroscience, aims to reconstruct stimuli from brain signals, primarily using functional magnetic resonance imaging (fMRI). Current methods are confined to a per-subject-per-model paradigm, limiting their applicability. This paper introduces MindBridge, a novel framework that achieves cross-subject brain decoding using a single model. MindBridge addresses challenges such as input dimension variability, unique neural patterns, and limited data for new subjects. It employs an adaptive aggregation function and a cyclic fMRI reconstruction mechanism for subject-invariant representation learning. The framework also includes a reset-tuning method for adapting to new subjects. Experimental results demonstrate MindBridge's ability to reconstruct images for multiple subjects, outperforming dedicated subject-specific models, especially with limited data. This advancement suggests promising directions for broader applications in neuroscience and more efficient utilization of fMRI data. **Introduction:** Brain decoding aims to reconstruct image stimuli from brain signals, a field with significant challenges, including input dimension variability, diverse neural responses, and data scarcity for new subjects. MindBridge addresses these challenges by employing an adaptive aggregation function, a cyclic fMRI reconstruction mechanism, and a reset-tuning strategy. The framework enables cross-subject brain decoding and novel fMRI synthesis, demonstrating its effectiveness and adaptability in experiments with the NSD dataset. **Methods:** MindBridge uses an adaptive max pooling function to unify fMRI signal dimensions, a cyclic fMRI reconstruction mechanism for subject-invariant representation learning, and a reset-tuning strategy for new subject adaptation. The framework is trained using a combination of SoftCLIP and MSE losses to ensure accurate CLIP embeddings. **Results:** MindBridge achieves comparable performance to state-of-the-art methods with a single model, demonstrating its effectiveness in cross-subject brain decoding. It also shows superior performance in new subject adaptation with limited data, highlighting the benefits of transferable knowledge from cross-subject pretraining. **Discussion:** MindBridge's success in cross-subject brain decoding and novel fMRI synthesis opens new avenues for practical applications in neuroscience. However, limitations include the need for a larger and more diverse dataset for generalizability and the challenge of preserving spatial relationships in serialized fMRI signals. **Conclusion:** MindBridge is a novel cross-subject brain decoding framework that overcomes the per-subject-per-model paradigm, enhancing decoding accuracy and enabling novel fMRI synthesis. Its effectiveness and adaptability make it a promising tool for advancing brain decoding technology.