Classification of Cheyne-Stokes breathing and obstructive sleep apnea using ECG

Detection of Cheyne-Stokes Breathing using a transformer-based neural network

Annotation of sleep disordered breathing, including Cheyne-Stokes Breathing (CSB), is an expensive and time-consuming process for the clinician. To solve the problem, this paper presents a deep learning-based algorithm for automatic sample-wise detection of CSB in nocturnal polysomnographic (PSG) recordings. 523 PSG recordings were retrieved from four different sleep cohorts and subsequently scored for CSB by three certified sleep technicians. The data was pre-processed and 16 time domain features were extracted and passed into a neural network inspired by the transformer unit. Finally, the network output was post-processed to achieve physiologically meaningful predictions. The algorithm reached a F1-score of 0.76, close to the certified sleep technicians showing that it is possible to automatically detect CSB with the proposed model. The algorithm had difficulties distinguishing between severe obstructive sleep apnea and CSB but this was not dissimilar to technician performance. In conclusion, the proposed algorithm showed promising results and a confirmation of the performance could make it relevant as a screening tool in a clinical setting.

Bioprofiles and mechanistic pathways associated with Cheyne-Stokes respiration: insights from the SERVE-HF trial

INTRODUCTION

The SERVE-HF trial included patients with heart failure and reduced ejection fraction (HFrEF) with sleep-disordered breathing, randomly assigned to treatment with Adaptive-Servo Ventilation (ASV) or control. The primary outcome was the first event of death from any cause, lifesaving cardiovascular intervention, or unplanned hospitalization for worsening heart failure. A subgroup analysis of the SERVE-HF trial suggested that patients with Cheyne-Stokes respiration (CSR) < 20% (low CSR) experienced a beneficial effect from ASV, whereas in patients with CSR ≥ 20% ASV might have been harmful. Identifying the proteomic signatures and the underlying mechanistic pathways expressed in patients with CSR could help generating hypothesis for future research.

METHODS

Using a large set of circulating protein-biomarkers (n = 276, available in 749 patients; 57% of the SERVE-HF population) we sought to investigate the proteins associated with CSR and to study the underlying mechanisms that these circulating proteins might represent.

RESULTS

The mean age was 69 ± 10 years and > 90% were male. Patients with CSR < 20% (n = 139) had less apnoea-hypopnea index (AHI) events per hour and less oxygen desaturation. Patients with CSR < 20% might have experienced a beneficial effect of ASV treatment (primary outcome HR [95% CI] = 0.55 [0.34-0.88]; p = 0.012), whereas those with CSR ≥ 20% might have experienced a detrimental effect of ASV treatment (primary outcome HR [95% CI] = 1.39 [1.09-1.76]; p = 0.008); p for interaction = 0.001. Of the 276 studied biomarkers, 8 were associated with CSR (after adjustment and with a FDR1%-corrected p value). For example, higher PAR-1 and ITGB2 levels were associated with higher odds of having CSR < 20%, whereas higher LOX-1 levels were associated with higher odds of CSR ≥ 20%. Signalling, metabolic, haemostatic and immunologic pathways underlie the expression of these biomarkers.

CONCLUSION

We identified proteomic signatures that may represent underlying mechanistic pathways associated with patterns of CSR in HFrEF. These hypothesis-generating findings require further investigation towards better understanding of CSR in HFrEF.

SUMMARY OF THE FINDINGS

PAR-1 proteinase-activated receptor 1, ADM adrenomedullin, HSP-27 heat shock protein-27, ITGB2 integrin beta 2, GLO1 glyoxalase 1, ENRAGE/S100A12 S100 calcium-binding protein A12, LOX-1 lectin-like LDL receptor 1, ADAM-TS13 disintegrin and metalloproteinase with a thrombospondin type 1 motif, member13 also known as von Willebrand factor-cleaving protease.

ResCSRF: Algorithm to Automatically Extract Cheyne-Stokes Respiration Features From Respiratory Signals

OBJECTIVE

Cheyne-Stokes respiration (CSR) related features are significantly associated with cardiac dysfunction. Scoring of these features is labor intensive and time-consuming. To automate the scoring process, an algorithm (ResCSRF) has been developed to extract these features from nocturnal measurement of respiratory signals.

METHODS

ResCSRF takes four signals (nasal flow, thorax, abdomen, and finger oxygen saturation) as input. It first detects CSR cycles and then calculates the respiratory features (cycle length, lung-to-periphery circulation time, and time to peak flow). It outputs nightly statistics (mean, median, standard deviation, and percentiles) of these features. It was developed and blindly tested on a group of 49 chronic heart failure patients undergoing overnight in-home unattended respiratory polygraphy recordings.

RESULTS

The performance of ResCSRF was evaluated against manual expert scoring (ES) (consensus between two independent sleep scorers). In terms of percentage of CSR per recording, the mean difference [reproducibility coefficient (RPC)] between ResCSRF and ES was 0.5(6.4) and 0.5(8.1) for development and test set, respectively. The nightly statistics of CSR-related features output by ResCSRF showed high correlation with ES on the blind test set with the mean difference of less than 3 s and RPC of less than 7 s.

CONCLUSIONS

These results indicate that ResCSRF is capable of automating the scoring of CSR-related features and could potentially be implemented into a remote monitoring system to regularly monitor patients' cardiac function.

ECG biomarkers for simultaneous detection of obstructive sleep apnea and Cheyne-Stokes breathing

Use of extended electrocardiography (ECG) for detection of sleep disordered breathing SDB when obstructive sleep apneas and Cheyne-Stokes breathing are simultaneously present is explored. A multi-tier algorithm is designed that uses quantitative changes in the morphology of the QRS complex of Lead 1 and V4 due of SDB events and combines those changes with variations in heart rate to detect each type of SDB. For this purpose, ECG signals are divided into 15 minute epochs. These epochs are then subjected to baseline wander removal and R peak detection. An envelope of R peaks is computed to derive R Wave Attenuation (RWA). Concurrently, the heart rate variability (HRV) is also computed. Various biomarkers derived from these trends are combined to develop an algorithm to classify Normal, OSA and CSR epochs. One hundred and five (105) data clips from 15 subjects were used to test the proposed algorithm. It produced detection rates of 93.75%, 100% and 83.3% for Normal, OSA and CSR epochs respectively in case of training set (66 clips). Detection rates of 75%, 85.71% and 70.5% for Normal, OSA and CSR epochs respectively were obtained in case of test set (39 clips).