Hemodynamic Pattern Recognition During Deception Process Using Functional Near-infrared Spectroscopy.

Area of Science:

• Neuroscience
• Psychology
• Biomedical Engineering

Background:

• Deception is a complex psychological process involving intentional falsehood.
• Deception detection is crucial in various fields, necessitating reliable methods.
• Existing methods for deception detection have limitations.

Purpose of the Study:

• To investigate the efficacy of functional near-infrared spectroscopy (fNIRS) for deception detection.
• To compare the performance of human interpretation versus machine learning (SVM) in analyzing fNIRS data for deception detection.
• To assess the accuracy, specificity, and sensitivity of fNIRS-based deception detection.

Main Methods:

• Utilized a mock theft paradigm with ten subjects.
• Recorded hemodynamic variations in the prefrontal cortex using fNIRS during responses to questions (Induced Lies, Induced Truths, Non-Induced).
• Compared classification performance of a human evaluator using hemodynamic topograms against a support vector machine (SVM) classifier.

Main Results:

• The human evaluator achieved 84.33% accuracy in a tri-class problem and 92% in a bi-class problem.
• The SVM classifier achieved 95.63% accuracy in a tri-class problem using cross-validation.
• A trade-off exists between classification accuracy and computational complexity.

Conclusions:

• fNIRS is a viable technique for deception detection by analyzing hemodynamic changes.
• Machine learning classifiers like SVM offer higher accuracy than human interpretation of fNIRS data.
• While human interpretation of fNIRS topograms is possible, it results in lower prediction accuracy compared to automated methods.