Idea
Multi-condition respiratory disease classification platform from smartphone cough audio with high accuracy and balanced detection.
Research Paper
Core Innovation
This paper introduces active-frame QKV attention pooling to focus on relevant cough audio frames, mitigating silence dilution in long input windows. It combines Whisper encoder fine-tuning with dual-encoder cross-attention fusion and balanced contrastive learning to handle class imbalance and domain shifts across multiple datasets, achieving superior multi-class respiratory disease classification.
Why It Matters
Accurate multi-class respiratory disease detection from cough recordings enables low-cost, accessible screening on consumer devices. This reduces reliance on expensive clinical tests and supports early diagnosis and monitoring at scale. It transforms respiratory healthcare by providing scalable, remote, and rapid screening tools.
Market Size (TAM)
$10–20B TAM for digital respiratory diagnostics; $2–5B SAM from healthcare providers and telemedicine platforms. Driven by rising respiratory disease prevalence and demand for remote diagnostics.
Potential Customers & Pain Points
- Healthcare providers – Need scalable low-cost respiratory screening
- Telemedicine platforms – Require remote diagnostic tools
- Public health agencies – Need rapid disease surveillance
- Smartphone manufacturers – Seek value-added health features
- Insurance companies – Want early disease detection to reduce costs.
Business Model
SaaS platform licensing to healthcare providers and telemedicine companies; API access for smartphone manufacturers and health app developers; subscription model for continuous updates and support.
Competitive Landscape
- ResApp Health
- Hyfe AI
- Sonavi Labs
- CoughDetect
Implementation Challenges
- Data privacy and regulatory approval for medical diagnostics
- Variability in cough audio quality across devices and environments
- Integration with existing healthcare workflows and systems
Validation Strategy
- Pilot studies with healthcare providers to validate diagnostic accuracy
- Partnerships with telemedicine platforms for real-world deployment
- Regulatory submissions for medical device certification
- User studies to assess usability and integration in clinical workflows
Research Paper Overview
CoughSense: Five-Class Respiratory Disease Classification via Whisper Encoder Fine-Tuning and Dual-Encoder Cross-Attention Fusion with Balanced Contrastive Learning
Summary
CoughSense classifies cough recordings into five respiratory conditions using a fine-tuned Whisper encoder and dual-encoder fusion, addressing class imbalance and domain shifts. It achieves high balanced accuracy and recall, outperforming existing models, enabling multi-condition respiratory screening from smartphone audio.