نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Interpretable Machine Learning for Functional Brain Mapping of Motor Imagery Using EEG

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Morteza Shahedifar

Author:
Name 1:	Morteza Shahedifar - Corresponding - Presenter
Affiliation 1:	Electrical and Computer Eng. University of Tabriz, Iran

Name 2:	Alireza Siadatan
Affiliation 2:	Associate Professor, Senior Member of IEEE, University of Toronto, Canada

Name 3:
Affiliation 3:

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Electroencephalography (EEG) has long been used as a practical tool for studying functional brain activity due to its high temporal resolution and noninvasive nature. In the context of motor imagery, EEG-based analysis has been widely applied in brain–computer interface (BCI) research, primarily with the goal of signal classification and performance optimization. However, many existing studies focus on improving classification accuracy while providing limited insight into the underlying neural patterns learned by the models. As a result, the potential of machine learning as a tool for functional brain mapping, rather than simple decision making, remains underexplored. In particular, there is a need for approaches that can extract spatially and spectrally meaningful information from EEG data and relate these patterns to known neurophysiological mechanisms involved in motor imagery tasks.

Methods:

This study proposes an interpretable machine learning framework for functional brain mapping based on EEG recordings from publicly available motor imagery datasets. EEG data from benchmark sources, including PhysioNet and BCI Competition datasets, are analyzed without additional data acquisition. Standard preprocessing steps are applied, including band-pass filtering in motor-related frequency bands and artifact reduction. Spatial feature extraction is performed using common spatial patterns to emphasize task-relevant channels. The extracted features are then used to train compact learning-based models, including convolutional neural networks designed for temporal EEG analysis. To enable interpretability, model outputs are analyzed using channel-wise and frequency-aware relevance measures, allowing the identification of spatial and spectral contributions associated with different motor imagery classes. All analyses are conducted in a simulation environment using Python based tools.

Results:

The trained models achieve stable performance in motor imagery classification while providing additional information beyond conventional accuracy metrics. Interpretable analysis reveals consistent spatial patterns across subjects, with higher relevance observed over sensorimotor regions commonly associated with motor imagery processing. Frequency-specific analysis indicates that μ and β bands contribute differently to class discrimination, depending on the imagined movement. These findings demonstrate that the proposed framework is able to capture physiologically meaningful EEG patterns while maintaining reliable classification performance. Importantly, the extracted relevance maps remain relatively consistent across datasets, suggesting robustness of the learned representations.

Conclusion:

The results indicate that interpretable machine learning can serve as an effective tool for functional brain mapping in EEG based motor imagery studies. By combining classical signal processing with learning-based models and post hoc interpretability analysis, the proposed approach provides insight into the spatial and spectral structure of motor-related brain activity. This framework supports the use of machine learning not only for classification, but also for exploratory analysis in brain mapping research, and can be readily extended using existing public EEG datasets.

Keywords:

Machine Learning; BCI; Brain Mapping; EEG; Motor Imagery