Recognizing lung cancer using a homemade e-nose: A comprehensive study.

Area of Science:

• Biomedical Engineering
• Analytical Chemistry
• Respiratory Medicine

Background:

• Breath analysis is a non-invasive method for lung cancer detection.
• Current gas sensors for medical applications have limitations in performance.
• Diverse sensor arrays may enhance the accuracy of breath analysis for disease detection.

Purpose of the Study:

• To investigate the potential of a diverse sensor array in an electronic nose (e-nose) for lung cancer detection.
• To compare the performance of different feature extraction and classification algorithms for breath analysis.
• To evaluate the effectiveness of combining multiple sensor types for improved diagnostic accuracy.

Main Methods:

• Fabrication of an e-nose with 10 gas sensors of 4 types.
• Direct testing of the e-nose with breath samples from 153 healthy individuals and 115 lung cancer patients.
• Evaluation of five feature extraction algorithms (LDA, Fast ICA) and three classification strategies (Random Forest) with cross-validation.

Main Results:

• Successful acquisition of unique "breath-prints" for all participants.
• LDA and Fast ICA combined features yielded the optimal feature space.
• Clustering algorithms achieved high separation (Adjusted Rand Index > 0.95).
• Random Forest classification achieved 86.42% mean accuracy and 0.87 AUC.

Conclusions:

• A diverse sensor array in an e-nose can improve lung cancer detection performance.
• Feature extraction and classification methods significantly impact diagnostic accuracy.
• Further research is needed to fully evaluate the extent of performance improvement from sensor diversity.