System and computer-implemented method for automatically diagnosing paediatric sleep apnoea

Abstract

The present invention relates to a system and a computer-implemented method for automatically diagnosing paediatric sleep apnoea. The system comprises a pulse oximeter (2) and a data processing unit (3) configured to: segment the oximetry signal (5) from the pulse oximeter (2) into SpO2 segments (30); estimate the number of apnoeic events (31) in each SpO2 segment (30) using a convolutional neural network (22); obtain the number of apnoeic events for each recording period (23); obtain a plurality of samples (40) of the oximetry signal (5); classify the sleep stages of the samples (40) into a multi-class classification using a U-Net neural network (24); obtain the patient's total sleep time (25) from the classified samples (46); and, using linear regression (26), estimate the number of apnoeas and hypopnoeas per hour of sleep (11) of the patient based on the number of apnoeic events per recording time (23), the total sleep time (25), and the age (28) and sex (29) of the patient.

Images