Using radio waves to accurately detect respiratory diseases

The potential for everyday wireless signals to be used to spot common respiratory diseases without scans, needles or physical contact has been evaluated in a recent study, showing that changes in breathing picked up by radio waves could help to identify respiratory diseases quickly, safely and with high diagnostic accuracy.

Published in the journal Communications Medicine, the study assessed whether passive 6G/WiFi radio signals could be used to screen for respiratory disease without physical contact or ionising radiation.

It leverages the fact that each respiratory disease is associated with a distinct abnormal breathing pattern that uniquely modulates the amplitude, frequency and phase of reflected radio-frequency signals.

The international research team aimed to develop and evaluate a non-invasive diagnostic approach capable of distinguishing between asthma, chronic obstructive pulmonary disease (COPD), interstitial lung disease (ILD), pneumonia and tuberculosis (TB) by analysing disease-specific alterations in breathing patterns.

Identifying respiratory diseases using radio waves

The researchers collected a new dataset, termed OFDM-Breathe, in a clinical setting using software-defined radios that exposed the chest to non-ionising orthogonal frequency division multiplexing (OFDM) signals at a centre frequency of 5.23 GHz.

The study cohort comprised 220 adults, including 190 clinically diagnosed patients and 30 healthy controls. Among the patient group, 45 had asthma, 43 had COPD, 41 had TB, 30 had ILD and 31 had pneumonia.

Data were acquired during 30-second recording sessions, repeated four times per participant, yielding 26,760 seconds of radiofrequency data across 64 subcarriers. The study population included 125 men (mean age 54.2 years) and 95 women (mean age 48.6 years), and all patients were clinically stable at the time of data collection.

Multiple machine learning and deep learning models were evaluated. A convolutional neural network achieved the strongest overall performance, with an accuracy of 98% in differentiating between the five diseases and healthy controls, alongside high precision and recall across classes.

High diagnostic accuracy with limited bandwidth

Long short-term memory and transformer models also performed strongly, each reaching approximately 97% accuracy. An ablation study showed that reliable classification with 96% accuracy was possible using only eight of the 64 subcarriers, corresponding to 12.5% of the available bandwidth.

The researchers acknowledged several limitations, including the fact that it was a feasibility study conducted at a single centre, with a modestly sized and slightly imbalanced dataset, particularly for ILD and pneumonia. The approach was evaluated in clinically stable adults, and performance in other populations or real-world settings was not assessed, they added.

Clinically, the technique could be used as a complementary screening tool for supporting rapid, low-cost respiratory disease assessment, particularly in remote or resource-limited settings, the researchers concluded.

They also highlighted future directions including validation in larger, multicentre cohorts and exploration of integration into 6G-enabled integrated sensing and communication platforms.

Previous research has shown that vocal changes recorded on a smartphone can signal the start of a COPD exacerbation.

Reference
Buttar HM et al. Non-contact lung disease classification via orthogonal frequency division multiplexing-based passive 6G integrated sensing and communication. Commun Med (Lond) 2026;6(1):9.