A Few-Shot SE-Relation Net-Based Electronic Nose for Discriminating COPD.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Chronic Obstructive Pulmonary Disease (COPD) diagnosis often requires multiple tests.
• Early detection of COPD is crucial for timely intervention and risk mitigation.
• Limited breath samples pose a challenge for accurate diagnostic models.

Purpose of the Study:

• To develop an advanced electronic nose system for COPD diagnosis.
• To evaluate the performance of SE-RelationNet with limited breath samples.
• To demonstrate the feasibility of early COPD detection using minimal data.

Main Methods:

• Utilized SE-RelationNet, integrating residual blocks, BiGRU layers, and squeeze-excitation attention.
• Employed few-shot learning tasks (1-shot and few-shot) for model evaluation.
• Conducted ablation studies to validate model components and stability.

Main Results:

• Achieved 85.8% mean accuracy in 4-way 1-shot tasks (F1-score = 0.852).
• Reached 93.3% accuracy (F1-score = 0.931) with four samples per class.
• SE-RelationNet outperformed SiameseNet and ProtoNet by over 15% in 1-shot tasks.

Conclusions:

• The proposed electronic nose effectively diagnoses COPD with minimal breath samples.
• SE-RelationNet offers a robust and accurate solution for few-shot COPD detection.
• This technology facilitates early COPD diagnosis, potentially reducing lung cancer risks.