3.1 Experimental Design
We designed an experiment primarily to evaluate the application effectiveness and suitability of the system in remote and home environments. The benefits of HRVB in the CR field have been substantiated by numerous academic studies. Therefore, to further its broader application in the field of CR, especially for the rehabilitation of heart disease patients in long-term home settings, our experiment shifted its focus to verifying the system's suitability and effectiveness, rather than merely demonstrating the effects of HRVB. In light of this, we selected university students as our test subjects because this group is more readily accessible. Throughout the entire experimental process, we used the HRV indicators of these university students as the core evaluation criteria for assessing the application results of our system.
We successfully recruited 30 male university students, aged between 18 to 24 years, none of whom had recently taken psychotropic drugs, asthma, or cardiac medications and hadn't received any psychological therapy. Moreover, all participants abstained from smoking, alcohol consumption, or vigorous exercise an hour prior to the experiment. During the recruitment phase, metrics such as height, weight, and BMI of each participant were measured.
The study is divided into two segments. The first segment primarily aims to validate the performance of our proposed RF selection methodology. Additionally, it seeks to determine the number of participants who, in the absence of researcher intervention, could not sustain RF breathing due to issues like lack of concentration. The second segment focuses on whether patients can ensure that each breathing training session effectively enhances their HRV indices under the supervision of a human-machine collaborative cognitive decision-making model.
Participants initially familiarized themselves with the RF breathing concepts and techniques in a laboratory environment. The research team demonstrated and instructed participants on the correct pursed-lip abdominal breathing technique. Subsequently, participants practiced slow breathing at 6BPM guided by an animation, during which the research team evaluated participants' blink rate to gauge their attention and fatigue levels. For the "Stepped method" assessment, participants breathed at pre-determined BPM rates, ranging from 8BPM to 4.5BPM, with each frequency change of 0.5BPM lasting for 3 minutes. After completing the "Stepped method", 10 participants were randomly selected to undergo the "sliding method" assessment. In the next phase, one participant was chosen to undertake daily breathing exercises for a span of four weeks. The therapist, based on the cognitive decision-making model, determined the user's training progression. Should the participant fall asleep or if there isn't a notable improvement in the HRV, the therapist would discreetly awaken the participant and restart the training session.
3.2 Experimental Results
3.2.1 Signal Quality Testing of Wearable Devices
Prior to conducting the overall RF experiment, we first validated the performance and feasibility of our designed wearable shirt for ECG data collection. We compared the performance of our ECG shirt with conventional ECG electrodes during RF-breathing training, as shown in Fig. 5(a). Our ECG shirt demonstrated no significant difference in ECG signal acquisition compared to conventional ECG electrodes during both resting and low-frequency breathing states. The R-wave feature in the ECG signal was clear, indicating that we could obtain the required HRV-related indicators.
Additionally, we tested the effect of the airbag in guiding subjects' diaphragmatic breathing. Figure 5(e) shows the comparison of abdominal circumference between a subject's breathing without the airbag and their breathing guided by the airbag. It is evident that the subject's abdominal circumference significantly increased under airbag guidance, indicating improved breathing depth. This suggests that our designed breathing-guidance airbag has a significant effect on helping subjects perform effective diaphragmatic breathing.
3.2.2 “Stepped and Ranked Method”
We successfully recruited 30 healthy adult male participants for respiratory training. Upon the completion of HRVB, one participant withdrew from the study due to difficulties in mastering diaphragmatic breathing. The remaining participants successfully completed the slow-breathing experience with a target of 6 BPM. However, during the process of RF assessment, five participants were unable to continue due to fatigue and attentional deficits, one among them requested to temporarily halt the experiment for rest. In remote CR training, especially during the long-term training process, the requirement for sustained attention in daily HRVB training becomes considerably challenging in the absence of on-site supervision by a therapist, particularly over extended training periods [38, 39]. The data from these six participants were excluded, and further analysis and processing were conducted on the remaining 24 participants. This involved selecting the high-ranking (top three) frequencies of three indicators: "HRmax-HRmin (Peak-trough amplitude)," "percent total LF power as LF/ (LF + HF)," and "Phase synchrony." The Table 1 shows the ranking data of one participant's experiment [34].
Table.1 Ranking of three HRV indicators for one participant.
Phase synchrony | HRmax-HRmin | LF power | RF |
4.5 / 5 / 6.5 / 6 / 5.5 | 4.5 / 5 / 5.5 / 6 / 6.5 | 4.5 / 5 / 6 / 6.5 / 5.5 | 4.5 |
Therefore, this participant's top three rankings for HRmax-HRmin were 4.5/5/5.5, for LF power were 4.5/5/6, and for Phase synchrony were 4.5/5/6.5. The participant's RF was 4.5 BPM. It's important to note that the RF doesn't necessarily rank first across all indicators; it might only rank first in two or even just one of the indicators.
We first compared the RF results of all participants with the RF values determined by evaluators through guidelines. Before the evaluators determined the RF values of the participants, they were unaware of the values calculated by the new weighting method. The RF values of the 24 participants are shown in Fig. 6(a), with 22 participants having consistent RF values. For the participants with inconsistent results, 4.5 BPM and 5 BPM were both in the top two rankings for all three evaluation indicators. The manually determined method selected the highest-ranking indicator for HRmax-HRmin, while our method selected 4.5 BPM, which ranked first in the other two indicators. Although the results differed, the difference was only 0.5 BPM, and the effect on training still needs to be compared through long-term training.
In addition, to improve the efficiency of determining the RF, we attempted to abandon frequency domain indicators and used only time domain indicators (HRmax-HRmin ranking) as the basis for RF determination. The determined RF is shown in Fig. 6 (b).
3.2.3 Differences between the different methods
In this study, we conducted a comprehensive comparative analysis between the "Stepped and Ranked Method" and the "Sliding Method." Our primary objective was to rigorously evaluate the accuracy and reliability of both approaches in determining RF, while also exploring a strategy that could combine the advantages of both methods, facilitating the determination of RF for patients even in remote settings without guidance. To this end, we selected 10 willing participants and obtained their RF values using the Sliding Method implemented in FreeResp, referred to as the "sliding session." These values were then compared with those determined through the "Stepped and Ranked Method," referred to as the "stepped session." It is noteworthy that the Sliding Method predominantly relies on time-domain metrics for evaluation. The comparative results of both methodologies have been detailed in Table 2.
Experimental results were analyzed using IBM SPSS Statistics 27. Among the 10 participants, 8 fell within the 0.5 BPM resolution range of our method. The differences in the remaining two participants were also less than 0.6 BPM. Using the Wilcoxon signed-rank test, there was no significant difference between the "Sliding Method" RF values and our method values (p > 0.05).
3.2.4 Breathing Training in Home Settings
This study employed HRV metrics to investigate whether participants could effectively regulate the balance of their ANS during each training session under the supervision of a human-machine collaborative cognitive decision-making model. We randomly selected one volunteer and provided a
Table.2 Comparison of the results of the two methods.
Participant | Sliding Methods (BPM) | Our Methods (BPM) | Absolute difference (BPM) |
P 1 | 4.34 | 4.5 | 0.16 |
P 2 | 4.92 | 5 | 0.08 |
P 3 | 5.27 | 5.5 | 0.23 |
P 4 | 5.46 | 5.5 | 0.04 |
P 5 | 4.77 | 4.5 | -0.27 |
P 6 | 5.07 | 4.5 | -0.57 |
P 7 | 5.42 | 6 | 0.58 |
P 8 | 4.55 | 4.5 | -0.05 |
P 9 | 4.34 | 4.5 | 0.16 |
P 10 | 4.32 | 4.5 | 0.18 |
Mean (absolute value) | 4.846 | 4.900 | 0.232 |
SD (absolute value) | 0.427 | 0.538 | 0.185 |
detailed explanation of the daily training guidelines. Over the following month, the participant engaged in continuous daily training within the comfort of their home environment, without the need to return to the laboratory.
During the experiment, researchers monitored the participants' physiological signals in real-time based on the human-machine collaborative cognitive decision-making model to assess the effectiveness of the training. If it was discovered that a participant had forgotten to train for some reason, or if the training outcomes were not satisfactory, researchers would immediately intervene and communicate with the participant to adjust the training strategy. We required each participant to train at least four times per week, with each session lasting at least 15 minutes. Furthermore, participants were asked to re-evaluate their RF every week using the "Stepped and Ranked Method." Ultimately, we successfully collected complete training data for one participant over a 27-day period. Among these 27 training sessions, the participant forgot to train five times due to personal reasons, but made up for those sessions following reminders from the researchers.
We chose HRV as the indicator to assess the effectiveness of remote CR [40], which includes the Standard Deviation of RR Intervals (SDNN), the Root Mean Square of Successive Differences (RMSSD) between adjacent heartbeats, and the Proportion of RR Intervals Greater than 50 Milliseconds (PNN50). An improvement in these metrics signifies the effectiveness and reliability of the training data [41]. SDNN measures the standard deviation of R-peak intervals, with larger SDNN values indicating higher HRV. RMSSD measures the difference between adjacent HR and serves as a time domain measure of HRV. The data results, as shown in Fig. 7(a), indicate that unsupervised RF-breathing training using FreeResp effectively improves various HRV indices (p < 0.05). Figure 7(b) shows the ratio of HRV data during training to pre-training HRV data can reflect the activation state of the participant during the training process.