Voice EHR: introducing multimodal audio data for health

Frontiers in Digital Health, Journal Year: 2025, Volume and Issue: 6

Published: Jan. 28, 2025

Introduction Artificial intelligence (AI) models trained on audio data may have the potential to rapidly perform clinical tasks, enhancing medical decision-making and potentially improving outcomes through early detection. Existing technologies depend limited datasets collected with expensive recording equipment in high-income countries, which challenges deployment resource-constrained, high-volume settings where a profound impact health equity. Methods This report introduces novel protocol for collection corresponding application that captures information guided questions. Results To demonstrate of Voice EHR as biomarker health, initial experiments quality multiple case studies are presented this report. Large language (LLMs) were used compare transcribed (from same patients) conventional techniques like choice Information contained samples was consistently rated equally or more relevant evaluation. Discussion The HEAR facilitates an electronic record (“Voice EHR”) contain complex biomarkers from voice/respiratory features, speech patterns, spoken semantic meaning longitudinal context–potentially compensating typical limitations unimodal datasets.

Language: Английский

Vowel segmentation impact on machine learning classification for chronic obstructive pulmonary disease

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 22, 2025

Abstract Vowel-based voice analysis is gaining attention as a potential non-invasive tool for COPD classification, offering insights into phonatory function. The growing need data has necessitated the adoption of various techniques, including segmentation, to augment existing datasets training comprehensive Machine Learning (ML) modelsThis study aims investigate possible effects segmentation utterance vowel "a" on performance ML classifiers CatBoost (CB), Random Forest (RF), and Support Vector (SVM). This research involves individual models using three distinct dataset constructions: full-sequence, segment-wise, group-wise, derived from which consists 1058 recordings belonging 48 participants. approach comprehensively analyzes how each categorization impacts model's results. A nested cross-validation (nCV) was implemented with grid search hyperparameter optimization. rigorous methodology employed minimize overfitting risks maximize model performance. Compared full-sequence dataset, findings indicate that second segment yielded higher results within four-segment category. Specifically, CB achieved superior accuracy, attaining 97.8% 84.6% validation test sets, respectively. same category also demonstrated best balance regarding true positive rate (TPR) negative (TNR), making it most clinically effective choice. These suggest time-sensitive properties in production are important classification can aid capturing these properties. Despite promising results, size demographic homogeneity limit generalizability, highlighting areas future research. Trial registration registered clinicaltrials.gov ID: NCT06160674.

Language: Английский

The Bridge2AI-voice application: initial feasibility study of voice data acquisition through mobile health

Frontiers in Digital Health, Journal Year: 2025, Volume and Issue: 7

Published: April 15, 2025

Introduction Bridge2AI-Voice, a collaborative multi-institutional consortium, aims to generate large-scale, ethically sourced voice, speech, and cough database linked health metadata in order support AI-driven research. A novel smartphone application, the Bridge2AI-Voice app, was created collect standardized recordings of acoustic tasks, validated patient questionnaires, reported outcomes. Before broad data collection, feasibility study undertaken assess viability app clinical setting through task performance metrics participant feedback. Materials & methods Participants were recruited from tertiary academic voice center. instructed complete series tasks application on an iPad. The Plan-Do-Study-Act model for quality improvement implemented. Data collected included demographics including time completion, successful task/recording need assistance. Participant feedback measured by qualitative interview adapted Mobile App Rating Scale. Results Forty-seven participants enrolled (61% female, 92% primary language English, mean age 58.3 years). All owned smart devices, with 49% using mobile apps. Overall completion rate 68%, successfully recorded 41% cases. requested assistance completed challenges mainly related design instruction understandability. Interview responses reflected favorable perception voice-screening apps their features. Conclusion Findings suggest that is promising tool acquisition setting. However, development improved User Interface/User Experience broader, diverse studies are needed usable tool. Level evidence : 3.

Language: Английский