Physiological and morphometric biomarkers for synthetic media detection

Area of Science:

• Digital Forensics
• Biomedical Signal Analysis
• Artificial Intelligence Security

Background:

• The increasing realism of synthetic media (deepfakes) challenges digital forensics and medical data integrity.
• Synthetic videos can compromise biometric systems, patient identification, and telemedical imaging.
• Existing detection methods often lack interpretability and robustness to media degradation.

Purpose of the Study:

• To introduce and evaluate a dual-framework combining physiological and morphometric anomalies for detecting synthetic audiovisual content.
• To assess the interpretability and degradation-awareness of biologically grounded forensic indicators.
• To differentiate authentic from synthetic media under various conditions using a large-scale dataset.

Main Methods:

• Utilized the DeepFake RealWorld (DFRW) dataset (46,371 clips).
• Extracted physiological features: remote photoplethysmography (rPPG) variability, oculomotor dynamics, speech-motion synchrony.
• Extracted morphometric/topological features: curvature variance, bilateral symmetry, persistent homology.

Main Results:

• Both physiological and morphometric markers demonstrated measurable discrimination (mean Δp ≈ 0.21).
• Classification metrics showed high specificity but moderate sensitivity, suitable for triage signals.
• Morphometric features exhibited greater stability under compression and rescaling compared to physiological ones.

Conclusions:

• Physiological and morphometric anomalies serve as complementary, degradation-aware indicators of synthetic media.
• These biologically grounded features offer interpretable signals for digital security and biomedical signal analysis.
• The findings provide a methodological basis for synthetic content detection and telemedical data integrity assurance.