Depth Map Analysis for Predicting Obstructive Sleep Apnea Using Transfer Learning with VGG-19

Abstract

The rising stress levels in today's competitive educational and professional environments contribute significantly to various health issues, including obstructive sleep apnea (OSA). OSA is characterized by repeated airway blockages during sleep due to the relaxation of the tongue and surrounding muscles, leading to symptoms such as loud snoring, choking or gasping for breath, and persistent daytime fatigue. Diagnosing OSA can be a lengthy and costly process, resulting in many individuals remaining untreated and unaware of their condition. This study proposes a novel approach to diagnosing OSA using depth maps derived from human facial scans. By leveraging deep learning algorithms, we aim to enhance the identification process through the rich morphological information provided by depth maps compared to traditional 2D color images. Conventional machine learning models have struggled to achieve optimal prediction and classification accuracy for OSA detection. Therefore, we employ the VGG-19 deep learning architecture, extracting features from facial depth maps and applying transfer learning techniques with a pre-trained model on the IMAGENET dataset. This model is subsequently fine-tuned using a dataset of OSA-specific facial images. Our results demonstrate that the trained VGG-19 model effectively predicts OSA from new test images, offering a more efficient and accessible diagnostic tool for this prevalent sleep disorder.