[論文レビュー] Continuous Telemonitoring of Heart Failure using Personalised Speech Dynamics

Yue Pan, Xingyao Wang+7 more|arXiv (Cornell University)|Feb 23, 2026

Phonocardiography and Auscultation Techniques被引用数 0

ひとこと要約

The paper introduces Longitudinal Intra-Patient Tracking (LIPT) with a Personalised Sequential Encoder (PSE) to monitor heart failure via speech, outperforming cross-sectional methods in HF trajectory detection and deterioration prediction.

ABSTRACT

Remote monitoring of heart failure (HF) via speech signals provides a non-invasive and cost-effective solution for long-term patient management. However, substantial inter-individual heterogeneity in vocal characteristics often limits the accuracy of traditional cross-sectional classification models. To address this, we propose a Longitudinal Intra-Patient Tracking (LIPT) scheme designed to capture the trajectory of relative symptomatic changes within individuals. Central to this framework is a Personalised Sequential Encoder (PSE), which transforms longitudinal speech recordings into context-aware latent representations. By incorporating historical data at each timestamp, the PSE facilitates a holistic assessment of the clinical trajectory rather than modelling discrete visits independently. Experimental results from a cohort of 225 patients demonstrate that the LIPT paradigm significantly outperforms the classic cross-sectional approaches, achieving a recognition accuracy of 99.7% for clinical status transitions. The model's high sensitivity was further corroborated by additional follow-up data, confirming its efficacy in predicting HF deterioration and its potential to secure patient safety in remote, home-based settings. Furthermore, this work addresses the gap in existing literature by providing a comprehensive analysis of different speech task designs and acoustic features. Taken together, the superior performance of the LIPT framework and PSE architecture validates their readiness for integration into long-term telemonitoring systems, offering a scalable solution for remote heart failure management.

研究の動機と目的

• Address inter-individual variability in speech-based HF assessment.
• Develop a longitudinal tracking framework to monitor intra-patient HF trajectories.
• Design a Personalised Sequential Encoder to encode continuous speech histories.
• Validate the approach on a cohort of hospitalized HF patients and follow-up data.

提案手法

• Extract global and frame-level acoustic features from speech tasks.
• Apply statistical screening to identify HF-relevant features (HF-voice A/B).
• Propose Longitudinal Intra-Patient Tracking (LIPT) with a Personalised Sequential Encoder (PSE) to model intra-patient trajectories.
• Train and compare cross-sectional and longitudinal models (XGBoost and FNN) for HF state transition detection.
• Evaluate across multiple speech tasks (vowels, short sentences, long sentence) and analyze task effectiveness.
• Validate the approach on decompensated vs post-treatment states and on follow-up rehospitalisation data.

実験結果

リサーチクエスチョン

• RQ1 Can longitudinal modelling outperform traditional cross-sectional approaches for HF status estimation from speech?
• RQ2 Which speech tasks and feature sets yield the strongest signals for HF trajectory tracking?
• RQ3 How effective is the Personalised Sequential Encoder at capturing intra-patient temporal dynamics?
• RQ4 How well does the LIPT/PSE approach generalise to follow-up data including rehospitalisation prediction?

主な発見

• LIPT significantly outperforms cross-sectional approaches across architectures, e.g., accuracy improved from around 69% (cross-sectional) to up to 81.8% (longitudinal FNN) for selected feature sets.
• RASTA frame-level features achieve very high performance; combining RASTA with selected global features yields sensitivity around 99.8% and specificity around 99.7%.
• PSE with frame-level RASTA features reaches macro-F1 of 99.5% (decompensated to post-treatment) and 99.7% precision, indicating strong detection of HF trajectory changes.
• In follow-up evaluation, RASTA-based models effectively identify rehospitalisation with AUROC up to 0.94, though stable cases show higher false-positive rates requiring calibration.
• Longer, more comprehensive speech tasks (counting 1–60) provide the best intra-patient longitudinal information, while vowels offer clinical practicality.
• The study supports the feasibility of personalised speech modelling for scalable remote HF monitoring and highlights directions for calibration and broader data.