Development of Machine Learning Model for VO2max Estimation Using a Patch-Type Single-Lead ECG Monitoring Device in Lung Resection Candidates

Figure 1. The segmentation of the electrocardiogram (ECG). The ECG data are standardized into baseline, P-wave, normal QRS complex, premature ventricular contraction (PVC) QRS complex, T-wave, noise, and fibrillation probabilities per sample. ECG, electrocardiogram; QRS complex, a combination of the Q, R, and S waves; PVC, premature ventricular contraction.

Figure 2. Flow diagram of the study.

Figure 3. The Bland–Altman plots of the heart rate measurements at each stage of the CPET. These plots showed the degree of agreement between the two methods during graded exercise testing. The y-axis shows the differences in heart rate values of the CPET and MC-100, while the x-axis represents the mean heart rate values. The plots also indicate the mean differences between the estimated heart rate and heart rate from MC-100 and the upper and lower limits of agreement. (a) Warming up, (b) Stage 1, (c) Stage 2, and (d) Stage 3. CPET, cardiopulmonary exercise test.

Figure 4. The Bland–Altman plots of the VO2max estimation by estimation model and clinical equations. (a) machine learning model, (b) ACSM equation, (c) FRIEND equation, and (d) HF-FRIEND equation. The deviation between the estimated VO2max value and the true measured value was minimal when using the machine learning model for estimation. VO2max, maximal oxygen consumption.

Figure 5. Comparisons of heart rate measurement between an electrocardiogram (ECG) and a single-lead ECG monitoring device during a CPET test in two patients. The results showed that the single-lead ECG monitoring device (MC-100) demonstrated significantly less fluctuation than the conventional ECG. In addition, when examining the 12-lead ECG (yellow section), it was found that the HR values had substantial discrepancies. However, the single ECG patch showed more approximate HR measurements and performed better in situations with high noise levels. W, warming up stage; S1, stage1; S2, stage2; S3, stage3; HR, heart rate; CPET, cardiopulmonary exercise test.

Figure 6. Comparisons between measured VO2max and estimated VO2max using machine learning model. (a) Patient No.15 has poor pulmonary function but good exercise function, while (b) Patient No.26 has normal pulmonary function and good exercise function. These graphs show that the machine learning model was able to estimate the VO2max value in a manner closely resembling the VO2 trend seen in the CPET, regardless of the patient’s initial lung function status. ML model, machine learning model; VO2max, maximal oxygen consumption; VO2, oxygen consumption; CPET, cardiopulmonary exercise test.