Asr-Net:Brain Source Imaging Method Based on Deep Optimization Unfolding Network with Adaptive Sparse Regularization

EEG Source Imaging (ESI) is a technique for estimating cortical neural activity from EEG/MEG signals, with widespread applications, particularly in epilepsy research and Brain-Computer Interface (BCI) systems. While EEG/MEG technology provides high temporal resolution, its spatial resolution is limited and cannot directly measure signal sources in deeper brain regions. To address this issue, this paper proposes a novel brain source imaging method based on a deep optimization unfolding network. This method combines the advantages of classical optimization algorithms and deep learning by introducing adaptive sparse regularization and smoothness constraints, which improve the accuracy of source signal reconstruction while enhancing the model's interpretability and robustness. Our approach unfolds the traditional Iterative ShrinkageThresholding Algorithm (ISTA) into a deep neural network, where each layer corresponds to an iteration step of ISTA, and the network parameters are optimized through end-to-end training. Compared to traditional methods, our approach automatically adjusts the sparsity regularization parameters, avoiding the limitations of manual tuning, and shows better robustness in highnoise environments. Experimental results demonstrate that the proposed algorithm outperforms existing brain source imaging methods across multiple evaluation metrics, particularly showing higher reconstruction accuracy in low SNR conditions, with strong potential for practical applications.