From biosensing to perception: Collaborative few-shot learning for explainable digital biomarker identification in...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cancer Diagnostics

Background:

• Early cancer detection using in vitro diagnostic biosensors is limited by insufficient molecular biomarkers.
• Digital biomarkers offer comprehensive phenotyping but suffer from clinical data scarcity and generalization issues.

Purpose of the Study:

• To introduce Coupled Explainable Artificial Intelligence Recursive (CEAIR) learning, a framework for interpretable few-shot learning.
• To enable the extraction of digital biomarkers from limited serum samples using high-dimensional biosensor data.
• To improve early cancer detection, specifically for hepatocellular carcinoma.

Main Methods:

• Developed CEAIR, integrating computer vision and cooperative game theory for interpretable few-shot learning.
• Applied CEAIR to surface-enhanced Raman spectroscopy biosensor data from limited serum samples.
• Utilized classic machine learning algorithms for classification and external validation.

Main Results:

• CEAIR-derived digital biomarkers significantly outperformed traditional circulating molecular biomarkers in hepatocellular carcinoma detection.
• Achieved area under the curve values consistently exceeding 0.97 across multiple classifiers.
• Demonstrated strong generalization capabilities upon external validation.

Conclusions:

• CEAIR effectively overcomes limitations in generating diagnostic knowledge from high-dimensional, small-sample biosensor data.
• The framework learns reliable digital biomarkers for robust, non-invasive, and timely diagnosis of complex diseases.
• Highlights the potential of AI-driven digital biomarkers in clinical diagnostics.