Camera-Based Photoplethysmography for Measuring Heartbeat Intervals During General Anesthesia

Abstract

BACKGROUND:

Photoplethysmography has been used to assess vital signs since the late 19th century. Recently, camera-based photoplethysmography systems have gained attention due to their noninvasive nature. However, challenges such as low perfusion, motion artifacts, and ambient light interference limit their use during surgical anesthesia. This study evaluated the efficacy of a camera-based system (FaCare) compared with that of a conventional contact monitor (GE HealthCare CARESCAPE B850 patient monitor) in measuring heartbeat intervals during various stages of surgical anesthesia.

METHODS:

Thirty patients undergoing video-assisted thoracic surgery were included. Data were collected using a webcam and FaCare software at 4 stages: preanesthesia, postanesthesia, postanesthesia with a shadowless lamp, and postsurgery. Six remote photoplethysmography techniques using artificial intelligence algorithms processed the data.

RESULTS:

The results demonstrated a high level of agreement between the FaCare and GE HealthCare monitor. Pearson correlation analysis, Bland–Altman plots, and Welch’s t test indicated that 88.1% of the heart rate correlation coefficients between the 2 devices were >0.8. Furthermore, their heartbeat interval measurements showed strong agreement in the Bland–Altman plots. FaCare showed comparable functionality, offering a noninvasive alternative suitable for operating rooms.

CONCLUSIONS:

This study evaluated the FaCare camera-based photoplethysmography system, integrating 6 remote photoplethysmography techniques with artificial intelligence algorithms, and compared it to a conventional contact monitor for measuring heartbeat intervals during surgical anesthesia. The results showed strong consistency between FaCare and the GE contact monitor across different anesthesia stages. These findings indicate that the noninvasive FaCare system reduces infection risks and improves patient comfort. Future research is recommended to optimize artificial intelligence algorithms, data synchronization, and sampling frequency to enhance its clinical application.

KEY POINTS.

• Research question: Can a camera-based system like FaCare effectively monitor vital signs, such as heart rate, while avoiding issues associated with traditional finger-clipped devices?

• Finding: In this study, we tested a camera-based system called FaCare, which uses advanced technology to measure vital signs during surgery. The system was compared to traditional monitors that touch the skin. FaCare was just as accurate as the traditional devices and improved patient comfort by avoiding direct contact.

• Meaning: This technology could lead to better patient monitoring methods in the future.

Since the late 19th century, photoplethysmography (PPG) has been widely used in clinical devices to monitor vital signs such as heart rate (HR), respiratory rate, blood pressure, and blood oxygen saturation by detecting blood volume changes in the microvascular bed of tissues.¹ Over the past 20 years, interest in the medical applications of PPG has shifted from conventional contact devices to camera-based PPG, also known as imaging PPG (iPPG, PPGi, and PPGI), remote PPG (rPPG), distance PPG, noncontact PPG, and video PPG.² Camera-based PPG has been applied in various clinical settings, including for older patients,³ neonates and infants,^4,5 patients undergoing dialysis,⁶ patients with diabetic peripheral neuropathy,⁷ and patients in the intensive care unit (ICU).⁸ These populations benefit from camera-based PPG as it reduces infection risk, enhances comfort, and minimizes issues like wire entanglement and skin damage.

However, little research has explored the application of camera-based PPG in patients under surgical anesthesia, likely due to the lack of simple, suitable devices for the operating room. Surgical anesthesia presents challenges such as low perfusion,⁹ motion artifacts, ambient light interference, and facial occlusion from surgical drapes or masks.

This study focused on applying camera-based PPG in patients under surgical anesthesia. Traditionally, PPG research has focused on the amplitude and interval of its waveforms^10,11; however, camera-based PPG faces greater challenges from environmental light interference and technical limitations. Therefore, we concentrated on measuring HR and heartbeat interval (HBI) during anesthesia. We collected data from 30 patients using an affordable and widely available 30 fps webcam. The data were processed with a medical personal computer (PC) using the FaCare program. FaCare uses optical technology to measure PPG remotely, avoiding the inconvenience and risks of invasive or contact methods. It employs an artificial intelligence (AI) algorithm to reduce interference from skin, hair, blood vessels, and other factors on the rPPG signal, improving accuracy and reliability. The aim of this study was to compare the performance of camera-based PPG with conventional contact PPG monitors in measuring HBIs under surgical anesthesia across different phases (preanesthesia, postanesthesia, postanesthesia with a shadowless lamp, and postsurgery), demonstrating the feasibility and robustness of our method under various lighting conditions and anesthesia levels.

METHODS

This study was approved by the Institutional Review Board of Taipei Veterans General Hospital, Taiwan (approval number 2021-05-024CC), and written informed consent was obtained from all subjects participating in the trial. The trial was registered before patient enrollment at clinicaltrials.gov (NCT05023356, Principal investigator: Hui-Hsuan Ke, Date of registration: 2021-08-23). This manuscript adheres to the applicable Consolidated Standards of Reporting Trials (CONSORT) guidelines. Differences between the methods in our submission and those in the clinical trial registration are detailed in Supplemental Digital Content 2, Supplemental Material, https://links.lww.com/AA/F381.

Image Processing and Analysis

A flowchart of the image processing is shown in Figure 1. The Mediapipe Face Mesh was used to detect faces in the video in 2 steps: face detection (determining if a face was present in the frame) and facial landmark detection (obtaining the coordinates of the facial landmarks if a face was found). To capture rPPG, the region of interest was defined as the polygonal boundaries around the forehead, left cheek, and right cheek. Several facial landmark points were randomly selected within these boundaries. These points were separated into R, G, and B channels. Subsequently, the RGB signals were converted into rPPG signals to detect blood volume changes from facial videos. The 6 standard rPPG methods¹² used in this study are briefly described below:

Figure 1.

A processing pipeline using face detection, ROIs selection, color channel extraction, RGB to rPPG conversion, and peak and valley detection for accurate heart rate and HBI estimates from camera-based PPG signals. The lower part of the figure illustrates the FaCare program’s process for integrating 6 rPPG methods through a TCN and training it on data from awake, healthy individuals. The program integrates the 6 methods into a single rPPG signal through 3 TCN layers: the first layer extracts local time-series features and compresses the 6 methods into 4 features, the second layer refines these into 2, and the final layer produces a single rPPG signal optimized for pure heartbeat extraction. Below are waveform examples (X-axis: time, Y-axis: amplitude) from the 6 rPPG methods, which are ultimately fused into 1 rPPG signal, as depicted in the processing pipeline (left to right, fourth stage). CHROM indicates chrominance-based method; EKG, electrocardiogram; F, forehead; Green, green channel method; HBI, heartbeat interval; ICA, independent component analysis; L, left cheek; LGI, local group invariance; PCA, principal component analysis; POS, plane-orthogonal-to-skin PPG, photoplethysmography; R, right cheek; ROIs, region of interests; rPPG, remote PPG; TCN, temporal convolutional network.

Principal Component Analysis

Principal component analysis (PCA) was used to extract the main features from data for HR estimation, especially in rPPG. It reduced data dimensionality by transforming it into principal components, capturing most variability for a more straightforward analysis. PCA improved HR signal quality and accuracy by removing noise, making it suitable for noncontact HR monitoring using webcams and ambient light.¹³

Independent Component Analysis

Independent component analysis (ICA), a statistical method for separating independent signals from data, was applied in rPPG for HR estimation. The critical points of ICA included signal separation based on non-Gaussianity and blind source separation for HR signal extraction.¹⁴

Chrominance-Based Method

Chrominance-based method (CHROM) extracted HR signals using color based on red, green, and blue channels. It resisted light interference and relied on relative skin color changes, providing stability under different lighting conditions. A mathematical model converted the RGB color data into chrominance elements.¹⁵

Plane-Orthogonal-to-Skin Method

The plane-orthogonal-to-skin method (POS) method derived HR information from signals orthogonal to the skin by projecting RGB signals onto the skin plane to obtain pulse wave signals. It estimated blood volume changes based on light patterns and mitigated environmental light changes and motion artifacts using orthogonal projections, ensuring stability under various conditions.¹⁶

Green Channel Method

It operated on the principle that green light is most sensitive to blood volume changes due to its penetration depth into the skin and blood vessels. It worked under ambient light, making it versatile in various environments. This simple and efficient method extracted and filtered green channel signals from RGB images to provide stable HR measurements.¹⁷

Local Group Invariance Method

The local group invariance (LGI) method estimated HR from facial videos in natural environments by using local group invariance, ensuring performance under various conditions. It extracted HR signals from local facial features, making it suitable for HR estimation in noncontrolled environments and rPPG.¹⁸

The 6 rPPG methods described above were based on different physiological principles, each offering advantages depending on the scenario and condition. Most camera-based PPG devices use 1 or more of these methods. However, the FaCare program utilized all 6 rPPG methods, integrating them through a temporal convolutional network (TCN) based model¹⁹ trained from awake, healthy individuals. After receiving the RGB signals, the first TCN layer performed causal and dilated convolutions to integrate signals based on local time-series features, compressing the 6 methods into 4 combined features to capture higher-level temporal characteristics. The second TCN layer refined these 4 features into 2, enhancing pattern recognition and noise filtering. Finally, the last TCN layer compressed the 2 features into a single rPPG signal, achieving global integration for pure heartbeat signal extraction. Subsequently, we implemented automated peak and valley detection using the find_peaks function from scipy, setting various parameters to describe waveform characteristics. For example, the peak height must exceed a specified threshold to ensure that only sufficiently high signals are identified as peaks. The peak amplitude must meet a threshold for the height difference relative to neighboring points. The distance between 2 peaks must exceed a specified interval to avoid misclassifying closely spaced signals as independent peaks. Peak prominence requires the distance from the peak to the lowest surrounding point to exceed a certain value. Furthermore, peak width is used to filter shape characteristics further. By applying these combined conditions, reliable HR and HBI estimates can be obtained from camera-based PPG signals.

Anesthesia Procedures and Intraoperative Monitor Settings

To evaluate the performance of FaCare in measuring the HR and HBI of patients under surgical anesthesia, we conducted a clinical experiment at the Department of Anesthesiology of Taipei Veterans General Hospital, Taiwan. Patients who underwent video-assisted thoracic surgery (VATS) lung wedge resection with propofol (n = 30) between December 2021 and June 2022 were enrolled, with all surgeries performed by a single thoracic surgical team.

Blood pressure was measured using a noninvasive blood pressure cuff, and pulse oximetry and electrocardiography (GE HealthCare CARESCAPE B850 patient monitor) were performed before anesthesia and throughout the surgery. The video recording system included a sensing component (30 fps webcam) mounted on a movable construction stand with an adjustable arm, and a medical PC with the FaCare program (SuperGenius, version 1). The sensing camera system was positioned approximately 1 m above the patient’s head in a supine position before each recording.

Figure 2 shows the anesthesiologist’s working area. The left side displays the GE vital signs monitor, showing electrocardiogram waveforms, pulse oximetry, blood oxygen saturation level, and blood pressure. Intravenous anesthesia was delivered via a target-controlled infusion pump adjacent to the GE monitor. The anesthesia machine on the right provided ventilatory settings for patients under sedation and muscle relaxation. Two shadowless lamps fixed to the ceiling turned off in stages 1, 2, and 4, and turned on in stage 3, focusing on the patient’s chest and abdomen during the surgery. The construction stand used in this study was adjustable in height and bending angle, fixing the sensing system 1 m above the patient’s face.

Figure 2.

Photograph during surgery: (i) construction stand with an adjustable arm for the sensing system; (ii) 30 fps webcam, 1 m from the patient’s face; (iii) recording PC; (iv) patient (facing camera); and (v) contact physiological monitors (GE HealthCare).

VATS with lung wedge resection and anesthesia involved several steps. After the patient was brought into the operating room and placed on the operating table, the anesthesia nurse confirmed the patient’s identity and attached basic contact physiological monitors, including a noninvasive blood pressure cuff, pulse oximetry, and electrocardiogram. After the anesthesiologist’s orders, analgesics, anesthetics, and muscle relaxants were administered to the patient. Once the anesthesia depth was confirmed, the anesthesiologist intubated the patient and used an anesthesia machine to manage oxygen delivery and carbon dioxide removal. After anesthesia induction, the surgeon activated the shadowless lamp to illuminate the surgical site, focusing on the patient’s chest and abdomen with sufficient brightness.

The surgery commenced after positioning the patient, disinfecting the skin, and draping with sterile sheets. The operation lasted several hours, during which the anesthesiologist and nurse continuously monitored vital signs and administered appropriate medications or interventions. After the surgery, the shadowless lamp was turned off, the sterile drapes were removed, and the anesthesiologist assisted the patient in emerging from anesthesia. The brief steps of VATS with lung wedge resection, anesthesia, and image-capturing timing are shown in Figure 3.

Figure 3.

An illustration of anesthesia and surgical steps for VATS with lung wedge resection. The 4 stages are preanesthesia, postanesthesia, postanesthesia with shadowless lamp, and postsurgery. VATS indicates video-assisted thoracic surgery.

Four 2-minute videos were recorded for each patient at different surgery stages. The 4 stages were as follows: (i) baseline condition (preanesthesia), (ii) postanesthesia induction and endotracheal intubation, (iii) immediately after stage 2 with the shadowless lamp (postanesthesia with shadowless lamp), and (iv) after surgical wound closure (postsurgery). During the last 3 stages, patients were under general anesthesia and unconscious. The illumination for the webcam video in stages 1, 2, and 4 was provided by the room’s fluorescent lamps, while in stage 3, both fluorescent and surgical shadowless lights provided illumination. Physiological signals from the GE HealthCare monitor (GE HealthCare CARESCAPE B850 patient monitor) were used as a reference.

Statistical Analysis

Demographic data were obtained from medical records. The GE vital signs monitor provided PPG data from pulse oximetry and electrocardiogram, whereas FaCare obtained PPG from video recordings. Statistical analyses were performed using a custom-written Python program. Data were analyzed using Pearson correlation analysis, Bland–Altman plots, and Welch’s t test.

RESULTS

Supplemental Digital Content 1 (Table S1, https://links.lww.com/AA/F380) presents the characteristics of the study participants (13 male and 17 female). The average patient age was 58.5 ± 10.4 years, and the American Society of Anesthesiologists grades ranged from I to III. The details of patients who underwent VATS are also summarized in Table S1, https://links.lww.com/AA/F380. Figure 4 shows the data for the first patient (ID: 1) using an X-Y scatter plot to display HR values from the GE monitor and FaCare across 4 surgical stages. The correlation coefficient (R) values for the 4 stages are 0.87, 0.87, 0.83, and 0.75, respectively. Of the 120 HR values collected, 88.1% had R > 0.8 (Supplemental Digital Content 1, Table S2, https://links.lww.com/AA/F380), indicating a strong positive correlation between the HR measurements from the GE monitor and FaCare. Subsequently, we used Bland–Altman plots to assess the consistency and interchangeability of the 2 methods.

Figure 4.

X–Y scatter plot of HR value of GE monitor and FaCare (ID: 1). A, Baseline condition before anesthesia (preanesthesia). B, Postanesthesia induction and endotracheal intubation. C, Immediately after the previous stage with shadowless lamp (postanesthesia with shadowless lamp). D, After surgical wound closure (postsurgery). The x-axis represents the HR value of GE monitor, while the y-axis represents the HR value of FaCare. The red solid lines indicate the linear fit of 2 measurements, and the dashed lines indicate the ideal calibration. HR indicates heart rate.

Figure 5A–D shows Bland–Altman plots for the first patient (ID: 1), illustrating HBI values from both the GE monitor and FaCare across the 4 stages. The plot shows the mean difference between the 2 methods. The Bland–Altman plots (Figure 4A–D) show the mean difference (mean bias) ± standard deviation (SD) for HBI values as 0.02 ± 0.22, 0.01 ± 0.19, −0.01 ± 0.21, and 0.03 ± 0.26, respectively.

Figure 5.

Bland–Altman plot of HBI value of GE monitor and FaCare (ID: 1) and HBI mean difference histogram. A, Baseline condition before anesthesia (preanesthesia). B, Postanesthesia induction and endotracheal intubation. C, Immediately after the previous stage with shadowless lamp (postanesthesia with shadowless lamp). D, After surgical wound closure (postsurgery). The x-axis represents the mean HBI value of GE monitor and FaCare, while the y-axis represents the difference in HBI values between GE monitor and FaCare (HBI value of FaCare minus the HBI value of GE monitor). The solid lines represent the mean of 2 measurements, and the dashed lines represent the limits of agreement (−1.96 to +1.96 s). E, HBI mean difference histogram. Each stage aggregates the bias values from the Bland–Altman plots of 30 patients. HBI indicates heartbeat interval.

In all stages, most data points fall within the 95% confidence interval (limits of agreement), defined as the mean difference ±1.96 SD of the differences. As the mean HBI value increased, the HBI value of FaCare was larger than that of the GE monitor; therefore, the data points cluster in the positive difference area. Conversely, when the mean HBI value decreased, data points clustered in the negative-difference area. The limits of agreement in Figure 4A–D are 0.24 and −0.21, 0.2 and −0.18, 0.2 and −0.22, and 0.29 and −0.24, respectively. The HBI results for the remaining 29 patients followed a similar trend (Supplemental Digital Content 1, Table S3, https://links.lww.com/AA/F380). The Bland–Altman plot shows strong agreement between the GE monitor and FaCare for detecting HBI.

Figure 5E shows histograms of mean bias in the Bland–Altman plots for the HBI of 30 patients across 4 stages, separately for the GE monitor and FaCare. The graph shows that for most stages, the HBI bias between the 2 methods was concentrated between −0.02 and 0.07 seconds, and the differences approximated a normal distribution.

The Bland–Altman plot is a popular evaluation tool in method comparison studies; however, it does not clearly define an acceptable range for the difference between the new and old methods. Therefore, we also used Welch’s t test (unequal variances t test) to compare the differences in detecting HBI between patients using the GE monitor and those using FaCare. Figure 6 shows the P values using colors ranging from yellow (1.0) to purple (0.0), indicating increasing significance levels from lighter to darker shades. A P value <.05, marked with a slanted red underline, was considered statistically significant.

Figure 6.

Analysis of P values from Welch’s t test for detecting HBI with GE monitor and FaCare. From top to bottom are IDs 1 to 30, and from left to right are stages 1 to 4. The magnitude of P values is represented by color, with P < .05 marked with a slanted red underline. HBI indicates heartbeat interval.

In Figure 6, among the 120 data points representing the 4 stages for each of the 30 patients, 13 data points (11%) had P value <.05, indicating significant differences between the GE monitor and FaCare in HBI detection. Specifically, there were 4 significant P values (slanted red underlines) in stage 1, 3 in stage 2, 3 in stage 3, and 3 in stage 4, indicating overall consistency between the 2 methods.

DISCUSSION

A noncontact, camera-based PPG device in surgical settings must address specific challenges in the operating room.²⁰ They should focus on capturing PPG signals from facial skin to minimize interference from the surgical field and eliminate surgical light effects. Their algorithms must handle nearby movement and perform real-time vital sign analysis. The devices should be simple, easy to assemble, adjustable for height, and compact to avoid obstructing the operating room space.

The FaCare program integrates 6 rPPG methods using a TCN-based model to optimize strengths and minimize limitations. Each method is based on distinct physiological principles, excelling under specific conditions. PCA effectively reduces dimensionality, removes noise, and extracts features, enhancing computational efficiency and signal quality.¹³ ICA effectively handles mixed signals with overlapping sources and non-Gaussian distributions, making it ideal for noncontact HR monitoring.¹⁴ The CHROM excels in remote HR monitoring under natural lighting, surpassing PCA and ICA during static activities.¹⁵ The POS method shows high stability and accuracy across various environments, including different lighting and motion artifacts.¹⁶ The Green Channel Method offers consistent HR measurements, maintaining high stability against light changes and motion artifacts in static settings.¹⁷ The LGI method is resilient in uncontrolled environments, effectively recording HR during remote health monitoring and activities with slight motion or outdoor interference.¹⁸

In this study, the noncontact PPG system, FaCare, was compared to the traditional GE monitor across 4 stages: preanesthesia, postanesthesia, postanesthesia with shadowless lamp, and postsurgery. The X-Y scatter plot showed a strong positive correlation between HR measurements from the GE monitor and FaCare, while the Bland–Altman analysis revealed significant concordance in HBI measurements between the 2 monitoring systems. Furthermore, Figure 5E illustrates the histogram of HBI mean differences, with slight variation between −0.02 and 0.07 seconds. However, factors such as blood pressure, body position, blood flow velocity, and vascular stiffness may introduce errors in pulse intervals obtained through PPG signals.²¹

Previous studies have mainly focused on HRV analysis using pulse intervals, with errors minimally affecting HRV. Due to the strong correlation between HRV and HBI, HRV findings remain relevant to this research. Past studies²² confirmed that interval inaccuracies negligibly affect HRV analysis. However, the Bland–Altman plot showed deviations as mean values between FaCare and the GE monitor shifted (x-axis), indicating a positive bias in FaCare. Results were consistent across all stages, with bias and 95% limits detailed (Supplemental Digital Content 1, Table S3, https://links.lww.com/AA/F380). Quality assessment followed the Bland–Altman agreement analysis standards.²³ Supplemental Digital Content 1, Figure S1, https://links.lww.com/AA/F380 highlights varying agreement levels, with superior (ID: 9, stage 4) and inferior (ID: 8, stage 2) plots. Challenges like noise, drift, calibration errors, data loss, and suboptimal signal quality persist in camera-based PPG devices.^24,25 Future advancements in self-learning models and neural networks^26,27 promise improved error correction and automation.

Figure 6 shows Welch’s t test outcomes, with 11% indicating significant differences. P < .05 indicates significance; however, P > .05 does not imply equivalence due to potential high variability in datasets. Welch’s t test was selected for its simplicity and prior use,^28,29 although it may not be ideal for assessing equivalence. These statistical findings serve as a reference for interpretation alongside Bland–Altman plots. The significant differences observed likely stem from varying sampling rates. The GE monitor operated at 300 Hz, compared to FaCare at 30 Hz. Hoog Antink et al.³⁰ evaluated HRV indicators via PPG with rates exceeding 125 Hz, noting high-frequency HRV influences. Malik et al.³¹ recommend a minimum of 128 Hz sampling for electrocardiogram in HRV studies. However, the effects of low sampling rates on HRV analysis remain quantitatively unexamined.²⁰ Camera-based PPG devices, like FaCare, which uses a 30 fps webcam, may lack the sampling rate needed to capture rapid or subtle vital sign changes during anesthesia. Higher rates can increase noise and reduce signal-to-noise ratios; however, balancing sampling rate and signal quality is essential for accurate data.^32,33 Further research is required to determine the ideal rate for clinical use.

The study encountered data consistency challenges. Two experimenters handled the devices, and minor timing differences might have occurred despite timestamps. Accurate pairing of data points from FaCare and the GE monitor was necessary for Pearson correlation, Bland–Altman and Welch’s t test analyses, though minor misalignments could not be ruled out. A higher sampling rate for camera-based PPG collection devices, with an automated and synchronized data collection system, could significantly improve HR and HBI detection consistency, further expanding the application of camera-based PPG in clinical anesthesia and surgery.

Expanding the application of camera-based PPG in clinical anesthesia and surgery requires addressing respiratory-induced variations (RIV) in PPG signals. Contact-based monitors detect more pronounced RIV due to the high sensitivity of finger blood vessels to respiratory pressure changes, using infrared or red light to capture these signals accurately.³⁴ In contrast, RIV in noncontact, camera-based PPG devices is weaker, as facial blood flow is primarily influenced by heartbeats. Environmental lighting, motion artifacts, and other factors further affect signal stability and reliability.^35,36 Accurate respiratory frequency estimation requires advanced signal processing algorithms, such as wavelet transform or spectral analysis.^37,38 This study relied on time-domain methods rather than frequency-domain methods, as time-domain waveforms may be more easily used to derive other PPG-related parameters in the future.¹⁷ However, this approach may have caused information loss, highlighting the need for further research.

This study has some limitations. First, the sample size was limited to 30 participants from a single center, all of whom were Asian with yellow skin tones. Individual differences in skin tone existed; however, the absence of racial diversity limited the study’s applicability across populations. Overbye-Thompson et al.³⁹ examined the potential impact of algorithmic bias on individuals with darker skin tones, showing that skin tone can influence image recognition algorithms. Second, FaCare used a single sampling rate, and comparisons with devices featuring higher or multiple sampling rates are necessary. Furthermore, this study used only time-domain methods. Finally, PPG research usually examines waveform amplitude and intervals, but environmental interference poses significant challenges for camera-based PPG systems. This study focused on HR and HBI measurements during anesthesia, leaving amplitude analysis for future research. Algorithm improvements are needed to manage interference and enhance the clinical use of FaCare.

In conclusion, this study evaluated FaCare, a camera-based PPG system using 6 rPPG techniques with AI algorithms for operating room use. The system showed a high correlation and consistency with the GE monitor across 4 surgical stages, confirmed by Pearson correlation analysis, Bland–Altman analysis and Welch’s t test. FaCare offers a noninvasive alternative to traditional monitors, reducing infection risks⁴⁰ and improving patient comfort. Future research should focus on optimizing sampling rates, data synchronization, and AI algorithms for both anesthetized and nonanesthetized conditions.

ACKNOWLEDGMENTS

We would like to express our sincere gratitude to the Taipei Veterans General Hospital and National Science and Technology Council (NSTC), Taiwan, for their generous funding and support. We would also like to thank Editage for reviewing and editing the English in this manuscript.

DISCLOSURES

Conflicts of Interest: None. Funding: This research was financially supported by Taipei Veterans General Hospital Research Grant V112B-026, and the National Science and Technology Council (NSTC), Taiwan, under grant number 112-2314-B-075-019. This manuscript was handled by: Thomas M. Hemmerling, MSc, MD, DEAA.