Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Global Health Technologies

Undergraduate Lightning Talks

A video selfie for vital sign monitoring: Comparison of techniques and region of interest on the performance of imaging-based heart rate assessment

Saturday, October 26, 2024
1:00 PM - 1:05 PM EST
Location: Room 342

Introduction:  Assessment of vital signs such as heart rate are a cornerstone of medical evaluations. Typical methods for heart rate assessment include the use of stethoscope, manual counting of pulses by hand or the use of vital sign monitors that require electrodes to be placed on the body. Unfortunately, these techniques have drawbacks in terms of infection control due to the need for close contact with the person and are ill-suited for remote healthcare applications or where mass assessments for public health purposes are required. A better technique for heart rate monitoring that does not rely upon close contact with an individual is needed. Imaging Photoplethysmography (iPPG) represents an innovative, non-contact approach to heart rate measurement, capturing blood volume changes through video analysis. However, knowledge of which of the commonly utilized iPPG algorithms and the regions of interest produces the best results are currently lacking. This study evaluated different iPPG techniques such as green pixel extraction, CHROM, Plane-Orthogonal-to-Skin (POS), Adaptive Green-Red Difference (AGRD), G-R Method; and the impact of different regions of interest (ROIs), such as the cheek, forehead, and full face, on heart rate measurement accuracy. The findings of this research will contribute to the evolution of iPPG-based heart rate monitoring, providing a foundation for more advanced, non-invasive, and more importantly contactless health assessment tools.

Materials and

Methods:  This study leverages the Univ. Bourgogne Franche-Comté Remote PhotoPlethysmoGraphy (UBFC-rPPG) public dataset, containing 34 minutes of facial video for 41 participants (9 females and 22 males) and the heart rate ground truth collected using a CMS50E transmissive pulse oximeter to evaluate the various extraction methods and ROIs. The pulse oximeter data was able to extract the Photoplethysmography (PPG) waveform, and served as the ground truth heart rate data. 10 participants were removed from further analysis due to data quality issues. The iPPG extraction methods assessed included green pixel extraction, Independent Component Analysis (ICA), CHROM, Plane-Orthogonal-to-Skin (POS), G-R, and Adaptive Green-Red Difference (AGRD). These methods were evaluated in combination with different regions of interest (ROIs), focusing on the cheek, forehead, and full face as seen in Figure 1. The above-mentioned methods and ROI choices were implemented and executed using Matlab R2024a to extract the heart rate from all participants’ video data. The extracted iPPG-derived heart rate information was compared to the ground truth heart rate information. Differences between the ground truth heart rate and the iPPG-derived heart rate for each participant, iPPG method and ROI were quantified using absolute error. Additionally, the proportion of heart rate measurements with differences of less than five beats was also calculated as an overall accuracy measure.

Results, Conclusions, and Discussions: Our results indicate that the best performance was derived from the combination of POS method and full face ROI which has a mean absolute error (MAE) of 2.3917. Additionally, the combination of the full face with the POS extraction method achieved the highest accuracy rate of 83.87% outperforming individual facial regions (Figure 2a). Specifically, the forehead achieved the best accuracy of 80.64% with the ICA method, while the cheek reached the best performance of 77.42%. Furthermore, the full face approach also has the lowest standard error of mean at 0.1, compared to those achieved by the cheek and forehead ROIs (Figure 2b).

Towards our primary goal of evaluating the optimal combination of iPPG method and ROI for imaging-based heart rate assessment, our results highlight the superior performance of full face ROI combined with the POS extraction method. Combined with the low standard error of mean for the full face, this indicates that utilizing the entire face provides more comprehensive information for heart rate estimation and may reflect the better tolerance against motion and variation among people and iPPG methods. Similarly, the superior performance of the POS method, which is among the newer methods, confirms the superiority of emerging techniques. By adopting this methodology, future iPPG systems can achieve greater precision and reliability, enhancing their practical applications in a real world setting.

The findings underscore the potential of full face scanning techniques and the POS extraction method in accurate iPPG heart rate assessment. This advancement promises to enable contactless and remote heart rate monitoring systems which is crucial for future healthcare. It can allow for real-time health assessments and early detection of potential issues without requiring physical visits to healthcare facilities, making it particularly important in increasing access to healthcare services, especially for individuals in remote or underserved areas. Additionally, with facilitation of continuous and accurate self-assessment, it empowers patients to actively manage their health, potentially leading to better health outcomes and reduced healthcare costs. Altogether, it supports the shift towards preventative care by enabling timely interventions and reducing the burden on traditional healthcare systems.

Acknowledgements (Optional):