FIGNet: A Robust and Interpretable Fuzzy-Irreversible Gated Network for Auditory Brainstem Response Classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Auditory brainstem response (ABR) is crucial for newborn hearing screening and neurological assessment.
• ABR signal analysis is challenged by noise interference, weak waveforms, and time-consuming data acquisition.
• Accurate and robust automatic classification models are needed for efficient ABR analysis.

Purpose of the Study:

• To develop an advanced deep learning model, FIGNet, for accurate and interpretable ABR signal classification.
• To address signal uncertainty and temporal direction in ABR data using novel deep learning techniques.
• To improve the efficiency and reliability of ABR analysis in clinical settings.

Main Methods:

• Developed FIGNet, a deep learning model integrating type-2 fuzzy logic with a time-irreversible attention mechanism.
• Applied fuzzy attention to mitigate noise impact and irreversible attention to model neural response directionality.
• Validated FIGNet on real-world ABR datasets for binary and five-class classification tasks.

Main Results:

• FIGNet achieved 93.72% accuracy in binary classification and 84.42% in five-class classification.
• The model demonstrated superior performance compared to existing methods on ABR datasets.
• Visualizations confirmed FIGNet's ability to focus on key waveform areas and maintain reliability under varying noise levels.

Conclusions:

• FIGNet provides a fast, interpretable, and robust solution for clinical ABR analysis.
• The model achieves high classification accuracy in both clean and noisy conditions, outperforming current approaches.
• FIGNet offers a promising advancement for automated ABR interpretation and diagnostic applications.