Computer-Assisted Diagnosis of Obstructive Sleep Apnea Hypopnea Syndrome (OSAHS) using Signal Processing and Machine Learning

This study considered the detection of sleep apnea-hypopnea events as a sample classification and time domain segmentation task. The proposed sleep apnea-hypopnea detection framework comprises SCM and ESM. SCM and ESM are responsible for the classification and segmentation tasks, respectively.

We collected PSG examination records from 39 cases. SCM classified samples with and without sleep apnea-hypopnea events using signal processing, feature extraction, and classification algorithms, achieving a classification accuracy of 90.72%, a sensitivity of 77.22%, and a specificity of 93.33%.

The proposed ESM takes the positive samples identified from SCM as input and performs threshold segmentation using the sequential features of the nasal pressure signal to accurately label AH events. Experimental results demonstrate that ESM accurately annotates AH events and is comparable to expert reference.

The precise segmentation outcomes of sleep apnea-hypopnea events can be utilized to calculate AHI and the duration of sleep apnea-hypopnea events. We computed AH duration in 39 cases and found the regression slope fitted to the reference true value was close to 1. AHI was determined by classifying and segmenting the samples using AHDF. The severity of OSAHS for the 39 cases was then categorized based on AHI, resulting in an accuracy rate of 82.1%. Moreover, the identification of OSAHS cases yielded an accuracy of 97.4% and a recall of 95.8%.

Thus, our research has the potential for practical applications in clinical computer-assisted professional screening with implications for labor reduction. We conclude by summarizing our findings and identifying valuable future research directions for OSAHS computer-aided diagnostic algorithms and diagnostic systems.