Non-contact lung disease classification via orthogonal frequency division multiplexing-based passive 6G integrated...

Area of Science:

• Medical Diagnostics
• Wireless Communication
• Machine Learning

Background:

• Current respiratory disease screening relies on methods like spirometry, CT scans, X-rays, and sputum analysis.
• These existing methods can be resource-intensive and may not be universally accessible.

Purpose of the Study:

• To investigate a new diagnostic approach for respiratory diseases using non-ionizing 6G/WiFi radio signals.
• To develop a method for screening five common respiratory diseases: asthma, COPD, ILD, pneumonia, and tuberculosis.

Main Methods:

• Subjects were exposed to 5.23 GHz 6G/WiFi multi-carrier radio signals.
• Respiratory diseases were identified by analyzing how each disease modulates the radio signals' amplitude, frequency, and phase.
• A new dataset, OFDM-Breathe, was collected from 220 individuals, containing over 26,000 seconds of radio signal recordings across 64 frequencies.
• Machine learning and deep learning models were evaluated for disease classification.

Main Results:

• A vanilla convolutional neural network achieved 98% accuracy in differentiating between the five respiratory diseases.
• The model demonstrated strong performance in precision, recall, and F1-score.
• An ablation study showed that 96% accuracy could be achieved using only 12.5% of the total bandwidth, leaving the rest for data communication.

Conclusions:

• The proposed method offers potential for real-time respiratory disease screening.
• This technology could improve health equity, particularly in developing countries.
• It lays the foundation for future integrated sensing and communication platforms in healthcare systems.