Sleep apnea is caused by several endophenotypic traits, namely pharyngeal collapsibility, poor muscle compensation, ventilatory instability (high loop gain), and arousability from sleep (low arousal threshold). Measures of these traits have shown promise for predicting outcomes of therapies (e.g. oral appliances, surgery, hypoglossal nerve stimulation, CPAP, and pharmaceuticals), which may become an integral part of precision sleep medicine. Currently the methods Sands et al. [1] developed for endotyping sleep apnea from polysomnography (PSG) are embedded in the original authors’ code, which is computationally expensive and requires technological expertise to run. We present a re-implementation and validation of the integrity of the original authors’ code by reproducing the endo-Phenotype Using Polysomnography (PUP) method of Sands et al. [1, 2] The original MATLAB methods were reprogrammed in Python; efficient methods were developed to detect breaths, calculate normalized ventilation (moving time-average), and model ventilatory drive (intended ventilation). The new implementation (PUPpy) was validated by comparing the endotypes from PUPpy with the original PUP results. Both endotyping methods were applied to 38 manually scored polysomnographic studies. Results of the new implementation were strongly correlated with the original (p<10 -6 for all): ventilation at eupnea V̇passive (ICC=0.97), ventilation at arousal onset V̇active (ICC=0.97), loop-gain (ICC=0.96), and arousal threshold (ICC=0.90). We successfully implemented the original method by Sands et.al. [1, 2] providing further evidence of its integrity. Additionally, we created a cloud-based version for scaling up sleep apnea endotyping that can be used more easily by a wider audience of researchers and clinicians.

Introduction Sleep stage classifications are of central importance when diagnosing various sleep-related diseases. Performing a full PSG recording can be time-consuming and expensive, and often requires an overnight stay at a sleep clinic. Furthermore, the manual sleep staging process is tedious and subject to scorer variability. Here we present an end-to-end deep learning approach to robustly classify sleep stages from Self Applied Somnography (SAS) studies with frontal EEG and EOG signals. This setup allows patients to self-administer EEG and EOG leads in a home sleep study, which reduces cost and is more convenient for the patients. However, self-administration of the leads increases the risk of loose electrodes, which the algorithm must be robust to. The model structure was inspired by ResNet (He, Zhang, Ren, Sun, 2015), which has been highly successful in image recognition tasks. The ResTNet is comprised of the characteristic Residual blocks with an added Temporal component. Methods The ResTNet classifies sleep stages from the raw signals using convolutional neural network (CNN) layers, which avoids manual feature extraction, residual blocks, and a gated recurrent unit (GRU). This significantly reduces sleep stage prediction time and allows the model to learn more complex relations as the size of the training data increases. The model was developed and validated on over 400 manually scored sleep studies using the novel SAS setup. In developing the model, we used data augmentation techniques to simulate loose electrodes and distorted signals to increase model robustness with regards to missing signals and low quality data. Results The study shows that applying the robust ResTNet model to SAS studies gives accuracy > 0.80 and F1-score > 0.80. It outperforms our previous model which used hand-crafted features and achieves similar performance to a human scorer. Conclusion The ResTNet is fast, gives accurate predictions, and is robust to loose electrodes. The end-to-end model furthermore promises better performance with more data. Combined with the simplicity of the SAS setup, it is an attractive option for large-scale sleep studies. Support This work was supported by the Icelandic Centre for Research RANNÍS (175256-0611).

No abstract

No abstract

No abstract

Introduction Sleep apnea is caused by several key endophenotypic traits namely pharyngeal collapsibility, poor muscle compensation, ventilatory instability (high loop gain), and arousability from sleep (low arousal threshold). Already, measures of these traits have shown promise for predicting outcomes of therapies (oral appliances, surgery, hypoglossal nerve stimulation, CPAP, or pharmaceuticals) and thus may be an integral part of future precision sleep medicine with treatments administered based on underlying pathophysiology. However, currently, the novel methods developed for endotyping from polysomnography are computationally expensive and can only be performed by the original authors or their collaborators due to the need for technological expertise. Here we present a cloud-based method for endotyping sleep apnea from polysomnography for use in the clinical arena. Methods For cloud-based use, we optimized the Phenotype Using Polysomnography (‘PUP’) method of Sands et.al. (2015-2018) by performing the following: Code was translated from MATLAB to Python; efficient methods were developed to detect breaths, calculate normalized ventilation (moving time-average), and model ventilatory drive (intended ventilation). The new implementation (‘PUP.py’) was validated by comparing the measured traits against the original values. Results 38 manually scored clinical polysomnographic studies were endophenotyped using the two implementations. Results of the new implementation (‘PUP.py’) were strongly correlated with the original (p<10-6 for all): collapsibility and compensation (ventilation at eupneic drive ‘Vpassive’: r=0.98; ventilation at arousal threshold, r=0.97), loop gain (r=0.96), and arousal threshold (r=0.92). Conclusion We successfully implemented the original method by Sands et.al. to scale up sleep apnea endotyping and make it available to a broader audience. Support This work was supported by the Icelandic Centre for Research RANNÍS, the European Union’s Horizon 2020 SME Instrument (733461), and the American Heart Association (15SDG25890059).