AMARA: Quality-gated adaptive fusion for robust respiratory disease screening from cough acoustics

Cough and breathing patterns carry acoustic biomarkers of respiratory illness, enabling non-invasive smartphone-based screening. This paper presents AMARA (Acoustic Multimodal Analysis for Respiratory Assessment), a multimodal framework fusing statistical acoustic features, spectro-temporal representations, and clinical metadata for respiratory disease screening, addressing limitations of existing approaches through three contributions. Quality-Gated Adaptive Fusion (QGAF) incorporates per-modality signal quality estimates into the attention scoring function, down-weighting unreliable inputs rather than treating all modalities as equally trustworthy. Modality Dropout Consistency (MDC) regularization ablates entire modalities during training and penalizes prediction divergence, enforcing stability under missing or degraded inputs. A Curriculum-Based Modality Introduction (CBMI) strategy introduces modalities sequentially by reliability, with consistency constraints ensuring each addition improves the ensemble. An optional language model reasoning module generates candidate explanatory narratives requiring clinical verification. Validated on 909 curated COUGHVID samples via five-fold cross-validation, the framework achieves 95.2% sensitivity and 93.1% specificity. On low-quality recordings, QGAF + MDC achieves 89.8% accuracy, a 12.3 percentage-point improvement over unimodal baselines. Evaluation across five external datasets spanning four continents, combined with domain-adaptive regularization, reduces mean performance degradation from 7.7 to 3.9 percentage points. Pilot extensions to four-class differential diagnosis and three-class severity grading indicate the architecture generalizes beyond binary detection. The framework shows clinical potential but requires further large-scale prospective validation.

Authors

Institutions

Publication Details

Journal
Intelligent Data Analysis
Published
2026-11-18
DOI
https://doi.org/10.1177/1088467x261504172
Primary Topic
Phonocardiography and Auscultation Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

AMARA: Quality-gated adaptive fusion for robust respiratory disease screening from cough acoustics

F. Marir, Naila Marir, Takieddine Lombarkia
6 links
Intelligent Data Analysis
Phonocardiography and Auscultation Techniques
article

AMARA: Quality-gated adaptive fusion for robust respiratory disease screening from cough acoustics

F. Marir, Naila Marir, Takieddine Lombarkia
article en

Abstract

Cough and breathing patterns carry acoustic biomarkers of respiratory illness, enabling non-invasive smartphone-based screening. This paper presents AMARA (Acoustic Multimodal Analysis for Respiratory Assessment), a multimodal framework fusing statistical acoustic features, spectro-temporal representations, and clinical metadata for respiratory disease screening, addressing limitations of existing approaches through three contributions. Quality-Gated Adaptive Fusion (QGAF) incorporates per-modality signal quality estimates into the attention scoring function, down-weighting unreliable inputs rather than treating all modalities as equally trustworthy. Modality Dropout Consistency (MDC) regularization ablates entire modalities during training and penalizes prediction divergence, enforcing stability under missing or degraded inputs. A Curriculum-Based Modality Introduction (CBMI) strategy introduces modalities sequentially by reliability, with consistency constraints ensuring each addition improves the ensemble. An optional language model reasoning module generates candidate explanatory narratives requiring clinical verification. Validated on 909 curated COUGHVID samples via five-fold cross-validation, the framework achieves 95.2% sensitivity and 93.1% specificity. On low-quality recordings, QGAF + MDC achieves 89.8% accuracy, a 12.3 percentage-point improvement over unimodal baselines. Evaluation across five external datasets spanning four continents, combined with domain-adaptive regularization, reduces mean performance degradation from 7.7 to 3.9 percentage points. Pilot extensions to four-class differential diagnosis and three-class severity grading indicate the architecture generalizes beyond binary detection. The framework shows clinical potential but requires further large-scale prospective validation.

Intelligent Data Analysis
University Frères Mentouri Constantine 1 (DZ), Effat University (SA)
Good health and well-being
Phonocardiography and Auscultation Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon