BiTimeCrossNet: Time-Aware Self-Supervised Learning for Pediatric Sleep

2026-02-02

AI Summary

This work addresses the limitation of existing methods in modeling long-duration pediatric sleep physiological recordings, which often neglect the global temporal position of segments and struggle to capture temporal dependencies and multimodal interactions. To overcome this, we propose BiTimeCrossNet, a novel framework that explicitly incorporates a time-aware mechanism into self-supervised sleep representation learning for the first time. Our approach leverages a cross-attention architecture to model pairwise interactions among multiple physiological signals without requiring sequence-level labels. Combined with multimodal fusion and linear probing using frozen backbones, BiTimeCrossNet consistently outperforms non-time-aware baselines across six downstream pediatric sleep tasks, demonstrating particularly strong performance in respiratory event detection and exhibiting robust generalization on an independent dataset.

Technology Category

• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Multi-modal Vision

Application Category

• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

We present BiTimeCrossNet (BTCNet), a multimodal self-supervised learning framework for long physiological recordings such as overnight sleep studies. While many existing approaches train on short segments treated as independent samples, BTCNet incorporates information about when each segment occurs within its parent recording, for example within a sleep session. BTCNet further learns pairwise interactions between physiological signals via cross-attention, without requiring task labels or sequence-level supervision. We evaluate BTCNet on pediatric sleep data across six downstream tasks, including sleep staging, arousal detection, and respiratory event detection. Under frozen-backbone linear probing, BTCNet consistently outperforms an otherwise identical non-time-aware variant, with gains that generalize to an independent pediatric dataset. Compared to existing multimodal self-supervised sleep models, BTCNet achieves strong performance, particularly on respiration-related tasks.

Problem

• self-supervised learning
• pediatric sleep
• time-aware modeling
• multimodal physiological signals
• long physiological recordings

Innovation

• time-aware self-supervised learning
• multimodal physiological signals
• cross-attention
• pediatric sleep analysis
• long-sequence modeling