Contactless Algorithm for Identifying the Facial Area Containing the Most Vital Sign-Related Information Using Machine Learning.

Abstract

Introduction: In recent years, there have been significant advancements in the field of biological parameter measurement. The emergence of contactless technology now enables the extraction of vital sign data using a standard personal computer (PC) equipped with a webcam. This system captures facial footage to obtain a remote photoplethysmography (rPPG) signal.

Objectives: Identifying the correct extraction area, known as the Region of Interest (ROI), is a crucial part of this process. Traditionally, the forehead is used as the primary site for extracting physiological parameters. However, this method poses challenges when patients have injuries or fractures on the forehead that need to be covered. To overcome this issue, we developed an automated computer vision model utilizing a machine learning technique that can analyse and classify blood flow quality in real-time across three facial ROIs: the forehead, right cheek, and left cheek.

Methods: The procedure starts by identifying the subject's face in the first frame and ensuring it remains tracked throughout the video. Subsequently, three mask regions are created to facilitate feature extraction using the bitwise operation method. In the final step, a Support Vector Machine (SVM) model is used to ascertain which of the three regions holds the most crucial physiological information.

Results: The proposed system shows promising results when evaluated using the PURE testing dataset. It delivered outstanding performance, achieving an accuracy rate of 99% and a False Positive Rate (FPR) of merely 0.7%.

Conclusions: The suggested approach is highly efficient in monitoring the forehead, right cheek, and left cheek of a patient's face, identifying the region with the richest physiological data among them. This method was created and evaluated with 10 participants who were seated in a chair, facing the camera. The camera was positioned 1 meter away from the subjects.