Open Access
Add to BibSonomy
Abstract
In this study, we explored non-contact pulse wave measurements from the extremities, particularly the soles, and examined their differences. Two experiments were conducted. First, we identified the optimal method for capturing pulse waves and discovered that the peak intensity of the green signal was the most effective. Then, we analyzed the temporal deviations between the electrocardiogram (ECG) and extremities based on these findings. Differences were observed in the face, palms, and soles of the feet in observing only three subjects. Previous attempts at extremity measurements have been made; however, our study is the first to focus on the foot sole. This study will pave the way for broader medical and research applications.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Human vital signals are usually required at the time of a physician consultation. However, these signals must be monitored repeatedly in certain diseases and situations. For instance, the heart rate is an essential indicator for assessing health status and analyzing cardiovascular function. Conventional heart rate measurement methods primarily use contact methods. However, non-contact methods with simultaneous regions of exploration are gaining attention, such as during infectious disease pandemics (e.g., COVID-19) [1,2].
Most instruments used to register human vital signs are measured by contact; that is, the instrument must be in contact with certain body parts. For instance, a stethoscope, used to listen to heart sounds, touches several parts of the chest. Even new stethoscopes [3] equipped with a microphone and communication module that enable remote heart-sound hearing and measurement must touch the chest. Pulse oximeters [4,5] were attached to the fingers to measure heart rate and oxygen saturation. Oximeters can also be attached to other appendages of the human body (e.g., toes and earlobes) to obtain information at the location where they are attached [6]. An electrocardiogram (ECG) was used to observe the heart movements in detail.
In general, these methods are accurate, reliable, and widely used. However, these technologies are inconvenient for newborns, the elderly, patients with dementia, and subjects with damaged skin, or because of the limitations of the sensing equipment, which may produce secondary effects on the human body [7].
Non-contact measurement methods, which are considered non-invasive, provide relatively accurate results. Non-contact methods, such as microwave- and camera-based methods, can minimize user discomfort. Microwave-based methods can pass through clothing and other materials, making it possible to make measurements without directly touching the body. However, microwave reflection patterns are affected not only by the human body but also by the surrounding environment and the movement of objects, making noise reduction a challenge. In addition, microwaves expose the human body to electromagnetic waves; therefore, control of electromagnetic radiation is required [8].
Camera-based non-contact measurement methods use optical methods to detect pulse waves. Specifically, the pulse wave pattern is determined by capturing color changes in the skin. As the heartbeat and blood flow change, the skin color also changes slightly. This technique, called “Remote Photoplethysmography (rPPG),” can non-invasively estimate heart rate [9–12].
The rPPG signals were obtained from digital camera data. It provides information about the blood volume; thus, the heart rate can be computed. The challenges of this technique include the lighting conditions and movement of the subject. However, research has been conducted in terms of subject positions and signal processing, which improves the accuracy and quality of the analysis. The face is the most common region for obtaining an rPPG signal. Owing to small changes in the rPPG signal, several algorithms based on different mechanisms have been proposed and developed, such as independent component analysis (ICA). Recently, deep-learning models have been developed as the size of the rPPG database has grown [13–16].
The advantage of using cameras is that information can be obtained simultaneously from multiple locations, and data from multiple locations may provide new information for medical care. An example of a potential contribution to medicine is abnormal blood vessels in the lower extremities, such as in patients with diabetes. There are indications for the diagnosis of peripheral arterial disease (PAD). Conventionally, a cuff is applied to the extremities to diagnose PAD, which requires manual labor and is prone to measurement errors. Although attempts have been made to automate this diagnostic method, it is still invasive [17,18]. In addition, blood flow measurements in the lower extremities are important because patients with Buerger's disease develop arteriosclerotic defect occlusion. However, owing to severe pain, it is difficult to measure blood flow by applying a cuff around the extremities and pressurizing them.
Niu et al. used an RGB camera to create a dataset of exposed arm, leg, and face images and attempted pulse wave detection using rPPG [19]. This dataset includes pulse waves acquired using PPG. However, no data were obtained for the sole of the foot, which was the focus of the present study. If we could measure the foot, for example, the sole, we would have a greater range. Some patients with DM undergo dialysis. During dialysis, it would be better to monitor blood status by photographing the feet.
In our previous study, we simultaneously measured pulse rates in the palms and soles of rats [20]. Therefore, we attempted to apply these techniques to humans and decided to perform non-contact pulse-wave measurements on human extremities, especially on the sole of the foot. In addition, we investigated differences in pulse wave propagation in the extremities. A green channel (G) signal was used for the measurement. The advantage of using the G-signal of an RGB camera is that it can be quickly realized by spatially averaging the G-signals and processed in real time [21,22].
The primary objectives of this study are to investigate the viability of non-contact pulse wave measurements on human extremities, particularly the sole of the foot, and to explore differences in pulse wave propagation across different extremities. The study also aims to assess the potential of utilizing the green channel G-signal from RGB cameras for real-time processing and analysis of pulse wave data. This experiment was conducted on three healthy subjects with same skin tones as the preliminary study.
We found foot measurements to be useful for many reasons [23,24]. One primary factor is that impaired blood flow significantly contributes to the risk of ulcer development. Monitoring plantar blood flow becomes crucial since ulcers on the feet of diabetic patients can severely compromise their quality of life. Moreover, oxidative stress and metabolic disorders may lead to vascular alterations and damage to the blood vessel's inner lining. Such damage can, in turn, impair the structure and function of peripheral nerves, leading to diabetic peripheral neuropathy (DPN) [25]. Given that neuropathy tends to manifest earliest at the extremities, particularly the feet, measuring blood flow there can be instrumental in detecting DPN at its onset. Therefore, there is a clear need for non-invasive techniques that can simultaneously capture data from various body parts.
2. Experimental methods
2.1 Subjects
The experiment involved three adult male participants with heights ranging from 168 to 182 cm, weights ranging from 63 to 140 kg, and ages ranging from 30 to 63 years. All participants were comprehensively informed of the objectives and content of the experiment, and informed consent was obtained from all participants.
This study was approved by the Ethics Review Committee of the International University of Health and Welfare (approval number: 23-Io-18).
2.2 Equipment
Figure 1 shows a photograph of the experimental setup. The subject was seated in a chair with armrests and palms facing the camera. The camera captures the entire from the face to the sole of the foot as well as the flickering lamp of the ECG. This arrangement enables the time synchronization of multiple regions of interest (ROI).
Fig. 1. (A) Rear view of subject: Captured with the camera positioned slightly below knee height. Two LED lights are strategically set at a 45° angle relative to the body. (B) Front view of subject: Show the area photographed by camera. An ECG indicator light is visible within the frame, serving as the reference value. A mirror aids in viewing the soles. To mitigate excessive light interference, a light-shielding cloth covers the foot rest. (C) Sole observation tools: The setup is designed to facilitate easier capturing of the foot's sole using a camera. Ribbon LEDs are affixed around an acrylic plate, with mirrors positioned to ensure the sole is optimally photographed. (D) ECG lead setting.
We used a DFK-33UP1300 RGB camera (The Imaging Source, LLC) with a RICOH FL-CC0814A-2 M lens focal length: 8 mm (aperture: f/1.4). The lighting was provided by two LED lights (VILTROX L116T) positioned 1 m away at a 45-degree angle to the subject, with the output set at 100% and the color temperature fixed at 5600 K. Raw images were captured at a resolution of 1280 × 720 to 1280 × 1024 pixels, with a frame rate of 60 FPS, and were archived as a sequence of images in the BMP format. IC-EXPRESS software supplied with the camera was used for image acquisition.
A customized footrest ensured that the sole was within the camera's field of view. Custom-made footrests with transparent plates and mirrors were used to mirror the foot soles. The footrest was crafted by substituting a segment of a readily available footrest, designed for foot placement, with a 15-mm transparent acrylic plate. For simultaneous imaging, it is vital to adjust lighting to suit the skin. The LED light strips were fixed to the edge of the transparent plate to maintain an adjustable light intensity. The LED light source of the footrest can be adjusted to correspond to the light intensity of the upper body. A black light-shielding cloth was placed over each leg to prevent external interference.
Nippon Kohden Corporation products were used to acquire ECG data. A telemetry-type electrocardiograph (model ZS-920P) with three leads was used as a sensor. The sensors were wirelessly linked to a monitoring device (BSM-2401). ECG management and the correct installation of the sensors were performed by a trained technician. The monitoring device was equipped with an indicator light that flashed synchronously with the R-wave signal.
In this study, the R signal of the ECG was derived by recording the flicker of the indicator light. The illumination of this indicator light, serving as our reference for the R peak, holds particular significance. For the specifications of the ECG device, the light is designed to illuminate in synchrony with the R peak, within a maximum delay of 32 ms following the detection of the R peak point in the waves of QRS complex. The signal captured by the camera from the indicator light, is shown in Fig. 2(D).
Fig. 2. (A) Pipeline schematic: Illustrates the process extracting the G-signal from the image of the right sole. (B) G-signal schematic: Shows the steps from the obtained G-signal to its peak detection. (C) Signal processing example: Displays the processing of rPPG (from subject 1's right sole) juxtaposed with ECG signal data in the final panel. The original and resampled signals are superimposed, making them appear as a single color in the top panel. (D) R signal of ECG: Obtained using ECG light synchronized with R peak.
2.3 Signal data preprocessing
The signal data processing pipeline is shown in Fig. 2. Data were acquired from multiple locations on the body by selecting data from a set portion of the ROI from a series of pictures captured using an RGB camera. The ROI was set at six locations: the face, right and left palms, right and left soles of the feet, and the indicator lamp of the ECG. The ECG’s indicator lamp was used to obtain the correct values. The ROI was configured as approximately 50 × 50 pixels. The pixel values of the G-channel in each ROI were averaged for each image to obtain the time-series signal, G(t), for each ROI. This process was executed in a program we created using MATLAB 2019b. The filtfilt function in MATLAB was used because the typical bandpass processing results in a time shift. The band-pass bandwidth was 0.7–2.0 Hz. In the third graph in Fig. 2(C), the detrended signal and the signal data after bandpass processing are superimposed, and it can be seen that they are time-matched. In the bottom graph of Fig. 2(C), the signal of the indicator light used for acquiring the ECG data is clear. After de-trending, the data were differentiated to obtain a signal for peak detection. The rPPG signal tends to be noisier than the ECG signal, making the accuracy of the peak detection crucial. To evaluate the efficacy of our peak detection method, we compared the maximum value of the G-signal (intensity peak) with the minimum value (absorption peak).
MATLAB was used for the peak detection. For the obtained peaks, the first and last peaks were removed to prevent false positives. Therefore, the peak interval (PI) was obtained from the extracted peak detection time, T = [ T2,…, Tk,…, TN-1 ]. To avoid false peak detection, T was set from 2 to N-1.
HR was calculated from PI
Figure 3(A) shows the actual calculated signal and the detected peaks from the right sole of Subject 1. Figure 3(B) shows a Poincaré plot of the peak signal. To discuss the time of the peaks, the correspondence of the peaks used for statistical processing is shown in Fig. 3(A). The rPPG signal, which indicated a change in blood flow, corresponded to the peak that appeared after the R peak signal of the ECG was generated. The bottom of the absorption peak, denoting the peak of the G-signal intensity, occurred thereafter. Therefore, for a, b, c, etc., respectively, we compare indicate the intensity, absorption, and ECG peaks, as well as the corresponding relationships.
Fig. 3. The peaks used in the calculations and Poincaré plot. (A) The correspondence of the peaks used in the data analysis was a, b, c, etc. Example intensity, absorption, and electrocardiogram (ECG) results are shown. The corresponding peak interval time is similar to those of a-b, b-c, c-d, etc. The first and last peaks were not used because they could cause errors. (B) Poincaré plot created based on peak (A).
3. Results
3.1 Comparison of accuracy of intensity peaks and absorption peaks
The accuracy of the HRs obtained from the rPPG was evaluated. For validation, we compared the rPPG HR with the correct value, which was the HR derived from an indicator lamp that flashed in sync with the R-wave of the ECG signal. Each measurement session spanned 15 s and was conducted in triplicate. Tables 1 and 2 present the average results of the three sessions. Table 1 shows the results of the calculation for the peaks obtained from the G-signal intensity. In comparison, Table 2 shows the results obtained using the peak with the highest absorption in the calculation, which inverted the intensity of the G-signal. Notably, the method leveraging the intensity peak exhibited minimal error, underscoring its high accuracy. Moreover, this precision was on par with that of ECG PI values, which demonstrated comparable deviation metrics. The standard deviations of the normal and normal intervals (SDNN) were calculated from the PI.
Table 1. Accuracy verification of HR calculated from intensity peak
Table 2. Accuracy verification of HR, calculated from absorption peak
The observed error rate was approximately 1%, demonstrating the functionality of the proposed device. Given the variability in PI, it is preferable for the SDNN results from the rPPG to align closely with those derived from the ECG. The SDNN values obtained from the ECG signals varied between 0.03 and 0.04 seconds. A comparison of the peaks obtained from the signal intensity with the absorption peak indicated that the peak obtained from the signal intensity was superior.
Table 3 summarizes the results for the ROIs set up on different parts of the body using the data presented in Table 1. Mean absolute error (MAE) and Pearson correlation were used as the indicators. The results in Table 3 show that the use of the intensity peaks yielded more significant results.
Table 3. Intensity and absorption measurements across different ROIs
3.2 Time difference between ECG and rPPG using INTENSITY peaks
Table 4 presents the time differences between the ECG and rPPG signals across the face, left palm, right palm, left sole, and right sole. These results are based on the peak intensity values. The measurements for each subject included SD and SDNN. The average heart rate (HR) was obtained using an ECG. All statistical analyses were performed using EZR [26],which is for R. More precisely, EZR is a modified version of the R commander designed to add statistical functions frequently used in biostatistics.
Table 4. Comparison of time differences between ECG and rPPG signals across various body locations using INTENSITY (sec)
Table 5 and Fig. 4 present the differences in peak times across the extremities based on the differences between the peak of R-wave in ECG signal detailed in Table 3 and the peak rPPG intensity data sourced from the extremities.
Table 5. Differences in Peak Times across Extremitiesa
Tukey's test, a method of multiple comparisons, was used for the face, palms (both left and right), and soles (both left and right) to determine whether statistically significant differences existed. Because the pulse wave propagation speed varied among the individuals, it was computed individually for each participant. The results showed no statistically significant differences between the left and right palms or between the left and right soles, as evidenced by P-values greater than 0.05. However, the results varied between subjects for the face and palms. In the comparisons between the face and palms, as well as between the face and soles, the p-value was less than 0.01, indicating a statistically significant difference.
The time difference in the peaks between the palm and sole, ΔT, was between 0.06-0.12 seconds. Considering the difference in distance from the heart to the palm and sole as L ≈ 0.6 m, the pulse wave propagation speed can be calculated as
From this equation, V = 5–10 m/s. This range aligns with previously reported values in the reference, which are between 4 and 15 m/s [27].
4. Discussion
Efforts to enhance the accuracy of rPPG include improvements in mobile device precision and the application of machine learning. Although adopting machine-learning models can contribute to accuracy enhancement, it is imperative to increase the accuracy of the pulse wave signal itself. It is more effective to process a signal with minimal noise than one overwhelmed by noise because it leads to fewer errors.
In this study, we revisited these foundational elements to explore ways to improve accuracy. Our initial focus was on a peak-detection method using G-signals. To detect peals from pulse waves, there does not seem to be a detailed discussion on whether peaks of intensity or absorption should be applied.
The contact PPG method detects two waves: the primary absorption peak influenced by hemoglobin and the reflected wave that follows the primary absorption peak. The proximity between these absorption peaks increases with age, hypertension, and atherosclerosis, leading to a blurred distinction between incident and reflected waves [28]. Shin et al. utilized contact PPG to compare Intensity peaks and Absorption peaks, as well as signal processing, for subjects aged 17-30 years [29]. As a result, the analysis using absorption and intensity peaks showed almost the same results, partly due to the fact that younger age groups were targeted.
We hypothesized that utilizing the intensity of the G-signal, which is not affected by reflected waves, would lead to fewer peak detection errors. The results of our investigation showed that using the peak value obtained from the intensity was more effective. Moreover, to capture the phenomenon in which the concentration of hemoglobin changes due to heartbeats and to calculate the pulse wave propagation speed correlating with heartbeats, further interpretation is considered a future challenge.
Our subsequent investigation focused on the time lag for each location when using a peak with a high-accuracy G-signal intensity. Even if there is no synchronization with the heartbeat, a significant time difference exists as long as the intensity signal condition remains the same. Our findings showed that when calculating the pulse wave propagation speed, we obtained a result of 5–10 m/s, which aligns with real-world data.
This study was designed as a preliminary investigation with a small sample size (N = 3). Number of data obtained for each individual subject was sufficient to allow statistical processing of the data. However, the small sample size of these three subjects makes generalization difficult beyond individual difference. In the subsequent phase of the research, we will consider various clusters, such as skin tone, age, BMI, and health status, including specific diseases, to assess the validity of including sole measurements.
In this experiment, because there was no device that directly outputs signals from the ECG, a camera was used to capture the flickering lamp of the ECG. In the future, when verifying the accuracy in detail, it will be necessary to confirm the validity of the method using a camera to capture the flickering lamp light.
In our study, only participants in the resting state were subjected to the experiment. We believe that the scope of application of this technology will become more clearer after conducting measurements in the resting and stimulated states. For example, it is known that pulse wave propagation velocity changes with exercise [30]. In future studies, we intend to measure the change in pulse wave propagation velocity due to exercise.
If the pulse wave propagation speed changes owing to blood flow issues and different pulse wave peaks are observed, it indicates an anomaly. Patients with diabetes are treated while lying down during dialysis. Capturing the state of pulse wave propagation by filming the sole of the foot during dialysis may enable improved management.
Measuring rPPG from the sole presents significant advantages. One notable benefit is the reduction of motion artifacts during measurement. When compared to other body parts like the face or fingers, utilizing a footrest for sole measurements allows for conditions less prone to motion disturbances. Furthermore, considering that diabetic patients often receive dialysis while in bed, enabling an environment where the foot can be photographed non-invasively on the bed allows for the monitoring of foot blood flow during dialysis. This could facilitate the early identification of complications in patients undergoing dialysis.
When pulse waves can be easily measured from multiple locations on the sole, the state of the blood flow in the foot can be ascertained. In patients with diabetes, blood vessels in the feet can become occluded, leading to necrosis. However, this cannot be detected using an ECG. With a pulse oximeter attached to the fingertip, detection can be difficult unless blood flow is present in that area. Our technology can also contribute to these observations. As the accuracy of smartphone cameras improves, regular measurement of the soles of the feet with a smartphone could visualize changes in the condition of diabetic patients, allowing them to feel the effects of treatment.
The measurement of pulse waves from the plantar surface offers an advantage over facial images from the perspective of personal data protection. We implemented pulse wave measurements from the face for telemedicine [31]. The current technology for measuring the heart rate from facial images, which involves capturing the face, is often avoided by individuals concerned about personal data and privacy protection. If the plantar surface is used instead, this hurdle is significantly reduced, which could broaden the application of this technology.
Measurements of infant non-contact heart rate are also attracting attention, and measurements from the soles of the feet may be useful [32].
We intend to conduct further investigations on this topic in the future. In this study, we did not examine methods that use all the RGB signals or machine learning. These issues are topics for future research. In addition, we did not conduct any studies using a pulse wave propagation speed measurement device simultaneously, which is a future research challenge.
5. Conclusion
Using the rPPG method, pulse waves can be measured simultaneously from the face, palms, and soles of the feet without contact with each other, with high accuracy. The G-signal from the RGB camera is used to calculate the pulse wave. When we examined whether the intensity or absorption values were more accurate, we found that the intensity value was superior. This high accuracy increases the possibility of determining the propagation velocity of the pulse waves. The ability to measure pulse waves in the extremities is expected to facilitate the diagnosis of diseases and blood pressure measurements based on differences in blood flow in the body.
Funding
Japan Science and Technology Agency (JPMJSP2109).
Disclosures
MT and NT: Imaging Tech Laboratory LLC, 727-2, Konakadai, Inage-ku, Chiba-shi, Chiba 263-0044, Japan (I, P, S).
Data availability
The data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
Supplemental document
See Supplement 1 for supporting content.
Reference
1. W. Taylor, Q. Abbasi, K. Dashtipour, et al., “A review of the state of the art in non-contact sensing for COVID-19,” Sensors 20(19), 5665 (2020). [CrossRef]
2. A. Smith, E. Thomas, C. Snoswell, et al., “Telehealth for global emergencies: Implications for coronavirus disease 2019 (COVID-19),” J. Telemed. Telecare 26(5), 309–313 (2020). [CrossRef]
3. M. Chowdhury, A. Khandakar, K. Alzoubi, et al., “Real-time smart-digital stethoscope system for heart diseases monitoring,” Sensors 19(12), 2781 (2019). [CrossRef]
4. J. Severinghaus and Y. Honda, “History of blood gas analysis. VII. pulse oximetry,” J Clin Monitor Comput 3(2), 135–138 (1987). [CrossRef]
5. T. Toshiyo, “Current progress of photoplethysmography and SPO2 for health monitoring,” Biomed. Eng. Lett. 9(1), 21–36 (2019). [CrossRef]
6. H. Lee, E. Kim, Y. Lee, et al., “Toward all-day wearable health monitoring: an ultralow-power, reflective organic pulse oximetry sensing patch,” Sci. Adv 4(11), 9530 (2018). [CrossRef]
7. F. Khanam, A. Al-Naji, A. Perera, et al., “Non-contact automatic vital signs monitoring of neonates in NICU using video camera imaging,” Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 11(2), 278–285 (2023). [CrossRef]
8. A. Singh, S. Rehman, S. Yongchareon, et al., “Multi-resident non-contact vital sign monitoring using radar: a review,” IEEE Sens. J. 21(4), 4061–4084 (2021). [CrossRef]
9. J. Park, H. Seok, S. Kim, et al., “Photoplethysmogram analysis and applications: an integrative review,” Front. Physiol. 12, 808451 (2022). [CrossRef]
10. W. Wang, S. Stuijk, and G. Hann, “Exploiting spatial redundancy of image sensor for motion robust rPPG,” IEEE Trans. Biomed. Eng 62(2), 415–425 (2015). [CrossRef]
11. A. Kamshilin and B. N. Margaryant, “Origin of photoplethysmographic waveform at green light,” Phys. Procedia 86, 72–80 (2017). [CrossRef]
12. A. Moço, S. Stuijk, and G. de Hann, “New insights into the origin of remote PPG signals in visible light and infrared,” Sci. Rep. 8(1), 8501 (2018). [CrossRef]
13. A. Al-Naji, K. Gibson, S. Lee, et al., “Monitoring of cardiorespiratory signal: principles of remote measurements and review of methods,” IEEE Access 5, 15776–15790 (2017). [CrossRef]
14. S. Premkumar and D. J. Hemanth, “Intelligent remote photoplethysmography-based methods for heart rate estimation from face videos: a survey,” Informatics 9(3), 57 (2022). [CrossRef]
15. E. Sabeti, N. Reamaroon, M. Mathis, et al., “Signal quality measure for pulsatile physiological signals using morphological features: applications in reliability measure for pulse oximetry,” Informatics in medicine unlocked 16, 100222 (2019). [CrossRef]
16. N. Podolyan, I. Mizeva, O. Mamontov, et al., “Imaging photoplethysmography quantifies endothelial dysfunction in patients with risk factors for cardiovascular complications,” Biomedical Signal Processing and Control 86, 105168 (2023). [CrossRef]
17. I. Arjaleena, R. Pekka, and M. Jussi, “Infrared thermography and vascular disorders in diabetic feet,” J. Diabetes Sci. Technology 14(1), 28–36 (2020). [CrossRef]
18. I. Arjaleena, R. Pekka, P. Matti, et al., “Infrared thermography follow-up after lower limb revascularization,” J. Diabetes Sci. Technology 15(4), 807–815 (2021). [CrossRef]
19. L. Niu, J. Speth, N. Vance, et al., “Full-body cardiovascular sensing with remote photoplethysmography,” In Proceedings of the IEEE CVF Conference on Computer Vision and Pattern Recognition, 5993–6003 (2023).
20. M. Takahashi, T. Yamaguchi, R. Takahashi, et al., “Non-invasive measurement of pulse waves in the soles of rats using an RGB camera,” OSA Continuum 4(12), 3014–3025 (2021). [CrossRef]
21. Y. Maeda, M. Sekine, and T. Tamura, “Relationship between measurement site and motion artifacts in wearable reflected photoplethysmography,” J. Med. Syst. 35(5), 969–976 (2011). [CrossRef]
22. B. Fallow, T. Tarumi, and H. Tanaka, “Influence of skin type and wavelength on light wave reflectance,” J. Clin. Monit. Comput 27(3), 313–317 (2013). [CrossRef]
23. DF. Bandyk, “The diabetic foot: pathophysiology, evaluation, and treatment,” Seminars in vascular surgery 31(2-4), 43–48 (2018). [CrossRef]
24. L. Vileikyte, “Diabetic foot ulcers: a quality of life issue,” Diabetes/Metab. Res. Rev. 17(4), 246–249 (2001). [CrossRef]
25. Z. Qiamg, Q. Zhihui, Y. Maoguang, et al., “Alterations in plantar vessel blood flow in patients with mild diabetic peripheral neuropathy,” BMJ Open Diab Res Care 10(1), e002492 (2022). [CrossRef]
26. Y. Kanda, “Investigation of the freely available easy-to-use software ‘EZR’ for medical statistics,” Bone Marrow Transplant 48(3), 452–458 (2013). [CrossRef]
27. J. Cavalcante, J. Lima, A. Redheuil, et al., “Aortic stiffness: current understanding and future directions,” J. Am. Col.l Cardiol 57(14), 1511–1522 (2011). [CrossRef]
28. C. Vlachopoulos, WW. Nichols, and MF. O’Rourke, McDonald’s Blood Flow in Arteries: Theoretical, Experimental and Clinical Principles (CRC Press,2011), pp. 195–209.
29. HS. Shin, C. Lee, and M Lee, “Adaptive threshold method for the peak detection of photoplethysmographic waveform,” Comput. Biol. Med 39(12), 1145–1152 (2009). [CrossRef]
30. K. Jiang and Z. Yuntao, “Biological analysis and application research of the influence of exercise on pulse wave velocity (PWV),” Network Modeling Analysis in Health Informatics and Bioinformatics 10(1), 5–12 (2021). [CrossRef]
31. M. Takahashi, R. Koike, K. Nagasawa, et al., “Development of telemedicine tools with an emphasis on visual observation,” Artif. Life. Robotics 27(1), 38–47 (2022). [CrossRef]
32. L. Maurya, P. Kaur, D. Chawla, et al., “Non-contact breathing rate monitoring in newborns: a review,” Comput. Biol. Med. 132, 104321 (2021). [CrossRef]
