WaveFusion Squeeze-and-Excitation: Towards an Accurate and Explainable Deep Learning Framework in Neuroscience.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Neurological event classification and localization are critical for diagnosis and treatment.
• Existing deep learning (DL) frameworks often lack real-time capabilities or explainability.

Purpose of the Study:

• To introduce WaveFusion Squeeze-and-Excite (SE), a multi-modal deep fusion architecture for neurological event classification and localization.
• To demonstrate the framework's effectiveness in classifying anxiety levels and enabling real-time event localization.

Main Methods:

• Utilizing lightweight Convolutional Neural Networks (CNNs) for per-lead time-frequency analysis.
• Employing a squeeze-and-excitation (SE) network with a temperature factor for multi-modal integration and final prediction.
• Developing a deep fusion architecture for enhanced performance.

Main Results:

• Achieved a high accuracy of 97.53% in classifying subjects' anxiety levels, surpassing prior methods.
• Demonstrated real-time localization of neurological events during inference without post-processing.
• Validated the framework's practical effectiveness and efficiency.

Conclusions:

• WaveFusion SE provides a practical and effective solution for neurological event classification and localization.
• The architecture represents a significant step towards explainable DL frameworks in neuroscience.
• Real-time capabilities and high accuracy position this as a valuable tool for clinical applications.