Automated Respiratory Event Detection Using Deep Neural Networks

  • 2021-01-12 17:43:17
  • Thijs E Nassi, Wolfgang Ganglberger, Haoqi Sun, Abigail A Bucklin, Siddharth Biswal, Michel J A M van Putten, Robert J Thomas, M Brandon Westover
  • 4


The gold standard to assess respiration during sleep is polysomnography; atechnique that is burdensome, expensive (both in analysis time and measurementcosts), and difficult to repeat. Automation of respiratory analysis can improvetest efficiency and enable accessible implementation opportunities worldwide.Using 9,656 polysomnography recordings from the Massachusetts General Hospital(MGH), we trained a neural network (WaveNet) based on a single respiratoryeffort belt to detect obstructive apnea, central apnea, hypopnea andrespiratory-effort related arousals. Performance evaluation includedevent-based and recording-based metrics - using an apnea-hypopnea indexanalysis. The model was further evaluated on a public dataset, theSleep-Heart-Health-Study-1, containing 8,455 polysomnographic recordings. Forbinary apnea event detection in the MGH dataset, the neural network obtained anaccuracy of 95%, an apnea-hypopnea index $r^2$ of 0.89 and area under the curvefor the receiver operating characteristics curve and precision-recall curve of0.93 and 0.74, respectively. For the multiclass task, we obtained varyingperformances: 81% of all labeled central apneas were correctly classified,whereas this metric was 46% for obstructive apneas, 29% for respiratory effortrelated arousals and 16% for hypopneas. The majority of false predictions weremisclassifications as another type of respiratory event. Our fully automatedmethod can detect respiratory events and assess the apnea-hypopnea index withsufficient accuracy for clinical utilization. Differentiation of event types ismore difficult and may reflect in part the complexity of human respiratoryoutput and some degree of arbitrariness in the clinical thresholds and criteriaused during manual annotation.


Quick Read (beta)

loading the full paper ...