RESPiRE: Respiratory Evaluation via Sensory Platform in Real Time Using Edge Learning

Overview

Cough remains one of the most prominent medical indicators of illness, amassing 120M+ clinical visits per year in the US alone. Although coughing is a common symptom, dismissing it as trivial risks lack of necessary clinical care and can result in outbreaks as people refrain from isolating until a possibly infectious virus subsides. Each cough has unique features linking it to certain conditions. Although clinicians can appreciate these features, non-medical individuals might struggle to do so. This study intends on building a low-cost, easily-accessible, and accurate early diagnostic tool for patients: RESPiRE. Using deep learning for performing cough audio data sample analysis, 34 features are extracted from labeled audio samples and visually represented via spectrograms on a simple Raspberry Pi 5 development platform. 13-MFCC coefficients capture the most important voice tract features including the timbre and spectral shape. The 12-Chroma coefficients represent pitch. Spectral contrast coefficients identify differences between loudest and quietest parts. ZeroCrossingRate and SpectralCentroid highlight the sound oscillation frequency and sudden bursts, along with sharp, deep and harsh sounds of cough spikes. Utilizing open source cough dataset (20,657 labeled samples), trained via supervised learning on common classification algorithms, RandomForest and CATBoost resulted in similar accuracies (74.9%), XGBoost (73%), and DecisionTreeClassifier (61.5%), demonstrating a promising low-cost easy to use “Doc in a Box” tool is possible to aid patients in early diagnosis and safeguard public health. Additionally, enabling more sensory input data such as throat videos and eye images would allow in categorizing cough types like bronchitis, pertussis, or asthma instead of simply being healthy or symptomatic.

Competition history

• CSEF 2026 Medicine & Physiology (Junior Division) · Entry J-15-09

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project