A Recall-First CNN for Sleep Apnea Screening from Snoring Audio

TL;DR

This study targets scalable sleep apnea screening using snoring audio by training a ResNet-based CNN on mel-spectrograms to prioritize recall. Through oversampling and class weighting, the model achieves a recall of $90.55\%$ while retaining a modest overall accuracy of $36.42\%$, highlighting the value of sensitivity in screening contexts. The approach is device-independent, enabling potential home or basic-clinic deployment, yet it relies on a very small dataset ($n=18$) without external validation, which limits generalizability. Future work should expand data, incorporate multimodal signals, and validate across diverse devices to realize a practical screening aid for early detection of sleep apnea.

Abstract

Sleep apnea is a serious sleep-related breathing disorder that is common and can impact health if left untreated. Currently the traditional method for screening and diagnosis is overnight polysomnography. Polysomnography is expensive and takes a lot of time, and is not practical for screening large groups of people. In this paper, we explored a more accessible option, using respiratory audio recordings to spot signs of apnea.We utilized 18 audio files.The approach involved converting breathing sounds into spectrograms, balancing the dataset by oversampling apnea segments, and applying class weights to reduce bias toward the majority class. The model reached a recall of 90.55 for apnea detection. Intentionally, prioritizing catching apnea events over general accuracy. Despite low precision,the high recall suggests potential as a low-cost screening tool that could be used at home or in basic clinical setups, potentially helping identify at-risk individuals much earlier.

Figures (8)

• Figure 1: System Overview
• Figure 2: mel-spectrogram of apnea
• Figure 3: mel-spectrogram of non-apnea
• Figure 4: CNN architecture for apnea detection
• Figure 5: Apnea Recall Comparison