Three simple steps to improve the interpretability of EEG-SVM studies

Area of Science:

• Neuroscience
• Machine Learning
• Medical Informatics

Background:

• Machine learning (ML) applied to electroencephalography (EEG) shows promise for neurological and psychiatric condition diagnosis and prognosis.
• Low clinical adoption of EEG-ML systems is a significant barrier to their real-world impact.

Purpose of the Study:

• To identify key reasons for the low clinical adoption of EEG-ML research.
• To propose actionable steps to improve the interpretability and clinical translation of EEG-ML studies.

Main Methods:

• Analysis of technical reporting standards in EEG-ML research, focusing on support-vector machine (SVM) algorithms.
• Identification of critical but often undocumented model development aspects: normalization, hyperparameter optimization, and cross-validation.

Main Results:

• A vast majority of EEG-SVM research literature fails to document crucial model development steps.
• Inconsistent reporting of normalization, hyperparameter optimization, and cross-validation significantly impairs performance interpretability.
• These undocumented aspects are critical determinants of system performance.

Conclusions:

• Inaccurate technical reporting is a primary obstacle to the clinical adoption of EEG-ML systems.
• Systematic documentation of normalization, hyperparameter optimization, and cross-validation is essential for improving EEG-SVM research interpretability.
• Enhanced reporting can facilitate the clinical translation and adoption of EEG-ML technologies.