• AI Health Checks

    • Lung AI - The Science

    Clinically validated screening for a range of respiratory diseases

    The Science behind

    Helfie's Cough AI 

    Helfie Cough AI turns a compatible, camera-enabled smartphone into a rapid respiratory screening tool that analyses cough sounds to support self-monitoring and inform personal choices about your lung health. 

    Developed in partnership with Swaasa®, and powered by advanced machine-learning models trained on real-world clinical datasets, Cough AI provides fast and accessible early screening for tuberculosis, COVID-19, COPD and asthma, anytime, anywhere.

    Using only a 10-second cough recording, the system extracts acoustic features linked to airflow obstruction, airway inflammation and respiratory infection. These signals are processed through clinically validated algorithms that deliver reliable, data-driven screening outcomes within seconds.

    Why this matters?


    Respiratory diseases are among the world’s leading causes of illness and death³. Chronic respiratory diseases such as COPD and asthma accounted for around 4 million deaths in 2019⁴, while infectious diseases like TB and COVID-19 caused a further 5 million deaths in 2021⁵. These conditions place a substantial burden on healthcare systems, particularly in low- and middle-income countries⁵.

    Cough is often the earliest and most universal symptom across these diseases, but timely screening remains limited by cost, availability of diagnostics and access to specialised care⁶. Many people delay help until symptoms worsen, leading to late-stage diagnosis and poorer outcomes.

    Helfie Cough AI addresses this gap by enabling immediate, low-cost respiratory screening directly from a smartphone. By identifying cough patterns linked to disease earlier, it supports timely follow-up, reduces delays to diagnosis and expands access for people who would otherwise struggle to reach a clinic.

    The science in your microphone

    Cough AI uses advanced acoustic analysis to transform raw cough audio into meaningful clinical insights, following a two-stage process:

    Cough Event Detection

    The smartphone records approximately 10 seconds of coughing. Machine-learning models identify cough events and segment them from background noise. The system analyses temporal and spectral features, such as frequency distribution, amplitude, airflow signatures and sound envelope, to isolate clinically relevant cough components.

    Disease Classification

    Once segmented, cough events are processed using state-of-the-art deep learning architectures:

    • Tuberculosis & COVID-19:

    Convolutional neural networks (CNNs) analyse spectrograms of the cough, while a feedforward neural network processes user information. Combined, these models classify coughs as TB-positive, TB-negative or inconclusive, and identify COVID-19 patterns.

    • COPD & Asthma:

    An Audio Spectrogram Transformer (AST) evaluates airway obstruction signatures, inflammation markers and sound modulations associated with chronic respiratory disease.

    Through this pipeline, the models distinguish between benign coughs and those associated with the four major respiratory conditions screened by the platform.

    Independent Validation 


    The underlying Swaasa® models have undergone clinical validation using data from nearly 4,700 participants across multiple hospital sites in India. Studies were conducted in partnership with Apollo Hospitals, the Government TB and Chest Hospital, and Christian Medical College (CMC) Vellore.

    Reported diagnostic performance includes¹,²,³:

    • Tuberculosis: 86.82% accuracy

    • COVID-19: 75.54% accuracy

    • COPD: 86.3% accuracy

    • Asthma: 83% accuracy

    Sensitivity and specificity values demonstrate strong agreement with gold-standard diagnostic tests, confirming the models’ suitability as an early screening tool. Performance has been tested across diverse populations and environments, supporting broad, equitable use.