VSSI(2p)-Net: Physics-guided deep unfolding with L(2p)-norm and variation sparsity for EEG source imaging

Comprehensive Summary

Wang et al. created a model called VSSI(2p) Net to improve EEG source imaging, which identifies where brain activity begins using scalp EEG recordings. This is a difficult task because there are many areas in the brain that can generate similar activity on the scalp. To combat this, the authors have designed a model that uses traditional mathematical imaging techniques with a learning based approach that adjusts its parameters automatically. The model was tested on various levels of noise, source size, and three other datasets that included epilepsy and nerve stimulation recordings. In testing VSSI(2p) Net against other models, it was discovered that VSSI(2p) Net is more accurate, less blurry, and computationally efficient while also producing interpretable results.

Outcomes and Implications

This is a significant problem because accurate localization of brain activity is necessary in the diagnosis of epilepsy, particularly in the localization of the seizure focus for surgical and brain mapping procedures. The previous models have been inaccurate and difficult to interpret. Through the use of source modeling and adaptive learning, VSSI(2p) Net is an accurate and trustworthy model that produces interpretable results. After further testing, this model could potentially enhance noninvasive seizure localization.