Remote sensing of alcohol consumption using machine learning speckle pattern analysis.

Area of Science:

• Biomedical Optics
• Machine Learning
• Forensic Science

Background:

• Current alcohol monitoring methods (breath, blood) have limitations.
• There is a need for faster, non-invasive alcohol assessment techniques.
• Law enforcement in countries like China requires sensitive detection of minimal alcohol levels.

Purpose of the Study:

• To develop and evaluate a remote optical technique for alcohol consumption assessment.
• To utilize machine learning for binary classification of alcohol presence.
• To offer a non-invasive, rapid alternative to traditional methods.

Main Methods:

• A laser illuminates the radial artery, and a camera captures defocused speckle patterns.
• Machine learning models were developed for automatic feature selection from temporal speckle pattern analysis.
• Models were evaluated using multi-class and binary classification schemes.

Main Results:

• Binary classification models demonstrated superior performance over multi-class models.
• Model C achieved 88% accuracy with 99% sensitivity for binary alcohol detection.
• High specificity (97%) was also achieved by one of the binary models.

Conclusions:

• The developed binary classification model effectively distinguishes pre- and post-alcohol consumption.
• The technique offers high sensitivity and accuracy, crucial for clinical and forensic use.
• This non-invasive optical method presents a promising advancement in alcohol monitoring.