An explainable deep learning framework for biosensing data interpretation in biomedical engineering and real-time...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Biosignal analysis often relies on complex models lacking interpretability.
• There is a need for AI frameworks that provide transparent health assessments from physiological data.

Purpose of the Study:

• To develop an explainable deep learning framework for transforming biosignal dynamics into interpretable health assessments.
• To enhance adaptability and diagnostic granularity using an Adaptive Health State Inference Mechanism (AHSIM).

Main Methods:

• The PhysioGraph Inference Network (PGIN) combines temporal graph reasoning with probabilistic modeling.
• An Adaptive Health State Inference Mechanism (AHSIM) adjusts diagnostic granularity based on uncertainty and signal entropy.

Main Results:

• The framework achieved superior diagnostic accuracy (up to 92.48%) and AUC (up to 93.65%) on four biosensing datasets.
• Outperformed transformer-based baselines like RoBERTa and T5.
• Provided transparent uncertainty estimates, demonstrating suitability for clinical and wearable applications.

Conclusions:

• The framework successfully integrates physiological semantics and model interpretability.
• It bridges the gap between black-box AI and trustworthy biomedical intelligence.
• Offers a reliable solution for interpretable health assessments using biosignals.