IPCT-Net: Parallel information bottleneck modality fusion network for obstructive sleep apnea diagnosis

Area of Science:

• Artificial Intelligence in Medicine
• Sleep Medicine
• Biomedical Data Fusion

Background:

• Obstructive sleep apnea (OSA) is a prevalent sleep disorder with significant associated healthcare costs due to complications.
• Current deep learning (DL) models for OSA diagnosis often rely on single data modalities, limiting their representational capacity.
• Effective modality selection is crucial for optimizing clinical diagnostic performance in OSA screening.

Purpose of the Study:

• To develop a novel modality fusion representation enhancement (MFRE) framework to improve obstructive sleep apnea (OSA) diagnostic performance.
• To provide quantitative insights into the selection of clinical diagnostic modalities for OSA.
• To enhance the capabilities of artificial intelligence (AI)-assisted diagnosis for OSA.

Main Methods:

• Development of a parallel information bottleneck modality fusion network (IPCT-Net) for extracting local-global multi-view representations.
• Implementation of branch sharing mechanisms to eliminate redundant information in fused representations.
• Evaluation using large-scale, real-world home sleep apnea test (HSAT) multimodal data to assess various fusion strategies.

Main Results:

• The proposed MFRE framework demonstrated significantly superior OSA diagnostic performance compared to existing methods.
• The IPCT-Net effectively extracted rich, multi-view representations, enhancing diagnostic accuracy.
• Experiments validated the framework's effectiveness across a large cohort of participants.

Conclusions:

• The MFRE framework offers a robust approach to modality fusion for improved OSA diagnosis.
• The study provides valuable evidence for selecting optimal data modalities in AI-assisted OSA screening.
• This research advances AI-driven diagnostic tools for obstructive sleep apnea, potentially leading to earlier interventions and better patient outcomes.