EchoSpiro™

Zero-hardware, zero-cloud edge ultrasonic spirometry turning commodity microphones and speakers into an 18.5 kHz pulmonary triage radar.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of## Inspiration

Respiratory diseases such as COPD, asthma, and acute lower respiratory infections affect hundreds of millions worldwide, with the vast majority of deaths occurring in low- and middle-income regions. Gold-standard diagnostic tools like clinical spirometry and plethysmography face severe deployment bottlenecks in low-resource frontline settings:

  • High Equipment Costs: Traditional spirometers and micro-manometers are cost-prohibitive for rural clinics.
  • Cross-Contamination Risks: Forced-air mouthpieces demand single-use disposable bacterial filters or rigorous sterilization.
  • Patient Compliance Barriers: Spirometry requires forced, uncomfortable exhalation maneuvers that are difficult for pediatric, geriatric, and critically ill patients.

EchoSpiro™ was conceived to overcome these hurdles by turning ordinary consumer hardware into an acoustic sonar workstation—delivering continuous, contact-free, zero-consumable respiratory monitoring and clinical triage.

What it does

EchoSpiro™ is an edge-based, non-contact active sonar diagnostic workstation that turns a standard laptop or mobile device into an acoustic pulmonary screening station:

  • Active CW Sonar Telemetry: Emits an inaudible, continuous 18.5 kHz ultrasonic carrier wave and listens for subtle Doppler shifts and phase modulations reflected off the patient's chest wall.
  • Sub-Millimeter Kinematic Tracking: Detects micrometric chest displacement in real time, reconstructing instantaneous tidal volume ($V_T$), respiratory rate (BPM), and inspiratory-to-expiratory ratios ($I:E$).
  • Dual-Band Visualizer & Waterfall Spectrogram: Displays a live 18.5 kHz Doppler waterfall spectrogram alongside a real-time displacement oscilloscope with ~12.2 ms loop latency and a continuous Signal Quality Index (SQI).
  • Automated Respiratory Pathology Screening: Uses high-frequency acoustic profiling to categorize respiratory patterns (Eupneic Normal, Tachypnea, Bradypnea, and COPD/Obstructive sawtooth signatures with real-time Wheeze Index calculation).
  • Volatile Longitudinal Tracking: Captures baseline and post-intervention respiratory curves to evaluate bronchodilator response or acute distress progression directly in client RAM.
  • Bilingual Attested Referral Generation: Generates tamper-evident referral slips (supporting English and Bengali / বাংলা) embedded with offline-verifiable QR codes and WebCrypto SHA-256 digital attestation seals.
  • Print-Ready Prescription Layout: Automatically formats into a crisp, high-contrast black-and-white A4 clinical triage slip under @media print for rural paper-based clinical handoffs.

How we built it

  • Acoustic DSP Engine: Built entirely in the browser using the Web Audio API (AudioContext, BiquadFilterNode, custom AudioWorkletNode). It synthesizes an 18.5 kHz Continuous Wave (CW) pilot tone and performs real-time quadrature demodulation (I/Q decomposition) and bandpass filtering directly on raw microphone PCM streams.
  • Kinematic Feature Extraction: Applied Short-Time Fourier Transforms (STFT), phase unwrapping, and dynamic background noise cancellation to isolate chest displacement from room reverberation and ambient acoustic noise.
  • Visualization Layer: Engineered with HTML5 Canvas and hardware-accelerated WebGL pipelines to render the real-time Doppler waterfall spectrogram and kinematic dual-band displacement curves at a consistent 60 FPS.
  • Frontend & Interface: Built with React, Next.js, and modern CSS Grid/Flexbox optimized for ultra-high-definition presentation displays (up to 4K UHD) as well as offline-first field deployment.
  • Security & Integrity: Implemented client-side WebCrypto API for in-browser SHA-256 digest creation and asymmetric signing, enabling tamper-evident physical handoffs via embedded SVG QR codes without requiring active cloud connectivity.
  • Validation Pipeline: Automated end-to-end clinical UI testing and high-fidelity screen asset capture using Playwright with custom device scale factors.

Challenges we ran into

  1. Multipath Reflection & Acoustic Clutter: Low-amplitude chest excursions (~1–5 mm) are easily drowned out by acoustic reflections off walls, desks, and clothing. We resolved this by combining adaptive comb notch filtering with dynamic background phase-subtraction algorithms.
  2. AudioWorklet Latency on Consumer Microphones: Consumer-grade laptop microphones introduce variable hardware phase jitter and automatic gain control (AGC) distortion. We bypassed default browser audio processing flags (echoCancellation: false, noiseSuppression: false, autoGainControl: false) and handled filtering strictly in our custom DSP pipeline to maintain sub-15 ms loop latency.
  3. Print Media CSS Normalization: Bridging dark-mode HUD aesthetics with regulatory paper-based medical records required precise @media print rule architecture to suppress telemetry backgrounds, maximize SVG contrast, and fit strict single-page A4 dimensions.

Accomplishments that we're proud of

  • Near-Zero Equipment Cost: Transformed any commodity laptop or tablet into a functional non-contact pulmonary monitor with zero proprietary hardware required.
  • Rigorous Ground-Truth Validation: Benchmarked across an $N=128$ simulated respiratory dataset, achieving an agreement coefficient of $r = 0.968$, $\pm 0.78\text{ BPM}$ Mean Absolute Error (MAE) on Bland-Altman analysis, and $96.88\%$ multi-class classification concordance.
  • Zero-Consumables & Infection Control: Completely eliminated disposable mouthpieces, spirometer turbines, and cross-contamination vectors for infectious triage scenarios.
  • True Offline Field Readiness: Engineered the entire DSP pipeline, signature verification, and report generator to run 100% locally in-browser without an internet connection.

What we learned

  • Modern client-side web technologies (Web Audio Worklets, WebCrypto, WebGL) are powerful enough to execute real-time, low-latency biomedical signal processing traditionally restricted to dedicated embedded hardware or MATLAB environments.
  • Designing for frontline global health requires prioritizing local context early—from multi-lingual referral generation (Bengali/English) to high-contrast paper print styling for low-connectivity rural health clinics.

What's next for EchoSpiro™

  • Expanded Multi-Center Clinical Trials: Partnering with regional healthcare centers to validate performance across diverse patient demographics, pediatric populations, and varied ambient noise profiles.
  • Acoustic Breath Sound Fusion: Combining active ultrasonic chest-motion tracking with passive acoustic auscultation to simultaneously detect lung crackles, rhonchi, and wheezes using lightweight on-device neural networks.
  • Mobile Hardware Profiling: Calibrating speaker-microphone phase responses across low-cost Android and iOS hardware configurations to support universal smartphone-based field deployment.
  • FHIR / OpenMRS Integration: Exporting encrypted referral slips directly into standard electronic health record (EHR) schemas for seamless integration into public health monitoring systems.

What we learned

What's next for EchoSpiro™

Built With

Share this project:

Updates

Submission history