Detection of obstructive apnea events in sleeping infants from thoracoabdominal movements

Abstract

The aim of the study was to determine whether in infants, the evaluation of thoracoabdominal movements alone, with no measurement of airflow, could be used to identify obstructive sleep apnea events (OA). Two different methods were used: first, we initially quantified thoracoabdominal asynchrony. Although 79.3% of OAs showed a significant increase of thoracoabdominal asynchrony, only 10.9% of the events scored by the identification of phase opposition were true OAs. Next, we developed two artificial neural networks (ANNs) as classifiers for the study of the thoracoabdominal signals. The first network was trained to locate obstructive and central apnea events. It correctly detected 75% of the OAs; however, only 6.2% of the detected events were true OAs. When a second network was used, OAs could not be discriminated from other portions of the signals showing similar phase characteristics. It was concluded that the information available in uncalibrated signals of thoracic and abdominal respiratory movements was insufficient to unambiguously detect OA events in sleeping infants.