Yongdong Fan

Deep neural networks are driving the iterative advancement of magnetoencephalography (MEG) decoding models. While explainable artificial intelligence, particularly traditional post‐hoc feature attribution approaches, has made significant progress in interpreting the prediction mechanisms of individual models, a critical gap remains in understanding the differences in decision logic between various models, known as model differencing. By facilitating model selection, optimization updates, and practical applications such as error pattern analysis and decision fusion, this approach demonstrates significant research and application potential. However, existing approaches face fundamental limitations: insufficient accuracy in differencing measurements, often relying on binary simplification, and deficient localization of differencing decision boundaries, typically constrained to low‐dimensional spaces. To address these challenges, we propose a rule‐based MEG model differencing approach called boundary‐optimized rules via predict‐probability difference (BO‐RPPD). Key innovations include (1) a novel measurement based on predict‐probability differences between dual models, enabling the direct learning of differencing rules, and (2) integrated counterfactual generation and feature reduction to guide the exploration of optimal predict‐probability difference boundaries, especially within low‐sample, high‐dimensional MEG feature spaces. Experiments on two MEG datasets demonstrate the overall superiority of our proposed approach: prediction performance significantly outperforms all benchmarks, achieving up to a 24% improvement in F1‐score and effectively covering broader samples. The number of generated rules matches the optimal benchmark, ensuring strong explainability. The approach shows practical value in error pattern analysis and decision fusion. Model‐agnostic, our approach generalizes effectively to electroencephalography (EEG) and structured datasets.

When decoding neuroelectrophysiological signals represented by Magnetoencephalography (MEG), deep learning models generally achieve high predictive performance but lack the ability to interpret their predicted results. This limitation prevents them from meeting the essential requirements of reliability and ethical-legal considerations in practical applications. In contrast, intrinsically interpretable models, such as decision trees, possess self-evident interpretability while typically sacrificing accuracy. To effectively combine the respective advantages of both deep learning and intrinsically interpretable models, an MEG transfer approach through feature attribution-based knowledge distillation is pioneered, which transforms deep models (teacher) into highly accurate intrinsically interpretable models (student). The resulting models provide not only intrinsic interpretability but also high predictive performance, besides serving as an excellent approximate proxy to understand the inner workings of deep models. In the proposed approach, post-hoc feature knowledge derived from post-hoc interpretable algorithms, specifically feature attribution maps, is introduced into knowledge distillation for the first time. By guiding intrinsically interpretable models to assimilate this knowledge, the transfer of MEG decoding information from deep models to intrinsically interpretable models is implemented. Experimental results demonstrate that the proposed approach outperforms the benchmark knowledge distillation algorithms. This approach successfully improves the prediction accuracy of Soft Decision Tree by a maximum of 8.28%, reaching almost equivalent or even superior performance to deep teacher models. Furthermore, the model-agnostic nature of this approach offers broad application potential.

Deep learning has greatly enhanced the predictive performance of magnetoencephalography (MEG) decoding. However, the lack of interpretability has become a major obstacle to the practical application of deep learning-based MEG decoding algorithms, which may lead to non-compliance with legal requirements and distrust among end-users. To address this issue, this paper proposes a feature attribution approach, which can provide interpretative support for each individual MEG prediction for the first time. The approach first transforms a MEG sample into a feature set, then assigns contribution weights to each feature using modified Shapley values, which are optimized by filtering reference samples and generating antithetic sample pairs. Experimental results show that the Area Under the Deletion test Curve (AUDC) of the approach is as low as 0.005, which means a better attribution accuracy compared to typical computer vision algorithms. Visualization analysis reveals that the key features of the model decisions are consistent with neurophysiological theories. Based on these key features, the input signal can be compressed to one-sixteenth of its original size with only a 0.19% loss in classification performance. Another benefit of our approach is that it is model-agnostic, enabling its utilization for various decoding models and brain-computer interface (BCI) applications.

As a biometric characteristic, electroencephalography (EEG) signals have the advantages of being hard to steal and easy to detect liveness, which attract researchers to study EEG‐based personal identification technique. Among different EEG protocols, resting state signals are the most practical option since it is more convenient to operate than the other protocols. In this paper, a personal identification system based on resting state EEG is proposed, in which data augmentation and convolutional neural network are combined. The cross‐validation is performed on a public database of 109 subjects. The experimental results show that when only 14 EEG channels and 0.5 seconds data are employed, the average accuracy and average equal error rate of the system can reach 99.32% and 0.18%, respectively. Compared with some existing representative works, the proposed system has the advantages of short acquisition time, low computational complexity, and rapid deployment using market available low‐cost EEG sensors, which further advances the implementation of practical EEG‐based identification systems.