Table 2 Studies on photoplethysmography based AF detection using ML approaches.
From: Photoplethysmography based atrial fibrillation detection: a review
Author (year) [ref.] | Number of patients | Dataset features | Age of population | Length PPG segments | Measurement device | Acquisition conditions | Input data | Methodology | Performance results for rhythms detection |
|---|---|---|---|---|---|---|---|---|---|
Shan et al. (2016)56 | 468 stroke patients | – | – | 2 min | Bedside monitor | Inpatient | RR times series features. Mean, median, standard deviation, RMSSD, power in very low-frequency range, low frequency (LF), high frequency (HF), ratio of power in LF and HF (LF/HF), multi-scale entropy, Shannon entropy, Turning point ratio | Support vector machines | AUC = 0.971, Sen = 0.942, Acc = 0.957 |
M. Lemay et al. (2016)57 | 20 |  | – | 10 s | PPG wrist- based device | Inpatient | RR times series features mean, minimum, median, and interquartile range of RR | Support vector machine | Acc = 0.9385 |
V. Corino et al. (2017)58 | 70 | 30 with AF, 9 ARR, 31 NSR | 40 ± 17, 76 ± 9, 65 ± 15 | 2 min | Empatica E4 wristband | Inpatient | RR times series features 24 features: spectral analysis, variability and irregularity analysis, PPG-waveform features | KNN classifier | NSR: Sen = 0.773, Sp = 0.928. AF: Sen = 0.754, Sp = 0.963. ARR: Sen = 0.758, Sp = 0.768 |
T. Schack et al. (2017)95 | 326 | 20 with AF, 294 of NSR, 12 of Noise | – | 20 s | Video camera of smartphone | Inpatient | RR times series features Mean, median, standard deviation and the mean absolute deviation (MAD); RMSSD; normalized RMSSD; Shannon entropy. PPG-waveform features: mean, median, SD and MAD, crest time, peak rise height, fall height, waveform width, cross-correlation of consecutive pulse segments, very low frequency, low frequency, high frequency and quotients of these spectral powers | Support vector machines | Perfect detection of AF |
S. Fallet et al. (2019)60 | 17 | AF NSR ventricular arrhythmias (VA) | 57 ± 13 | 10 s | Wrist-worn device | Inpatient | RR times series features: mean, standard deviation, median, interquartile, minimum, maximum, RMSSD. PPG-waveform features: adaptive organization index, variance of the slope of the phase difference, permutation entropy, spectral entropy, fractional spectral radius, and spectral purity index). | Decision trees | AF vs NSR: Acc = 0.981, Sen = 0.997, Sp = 0.924. AF vs VA: Acc = 0.959, Sen = 0.981, Sp = 0.887. AF vs (NSR&VA): Acc = 0.950, Sen = 0.962, Sp = 0.928 |
