Evaluation of different classification methods using electronic nose data to diagnose sarcoidosis

Area of Science:

• Pulmonary Medicine
• Biomedical Engineering
• Artificial Intelligence

Background:

• Sarcoidosis is a challenging disease to diagnose due to the lack of a definitive test.
• Electronic nose (eNose) technology offers a novel approach for disease diagnosis through breath pattern analysis.
• Artificial intelligence (AI) can be applied to eNose data for classifying complex biological patterns.

Purpose of the Study:

• To evaluate various dimensionality reduction techniques and AI classifiers for sarcoidosis diagnosis.
• To develop and validate an accurate diagnostic model for sarcoidosis using eNose data.
• To compare the performance of different methods for optimal model selection.

Main Methods:

• A dataset of 224 pulmonary sarcoidosis patients and 317 other interstitial lung disease patients was utilized.
• Multiple dimensionality reduction methods and hyperparameter-optimized classifiers were tested.
• Nested cross-validation was employed to assess diagnostic performance, with Random Forest (RF) selected as the best classifier.

Main Results:

• The Random Forest classifier combined with feature selection achieved the highest accuracy.
• The developed diagnostic model demonstrated an overall accuracy of 87.1%.
• The model achieved an area-under-the-curve (AUC) of 91.2%, indicating strong diagnostic capability.

Conclusions:

• An accurate diagnostic model for sarcoidosis using eNose technology has been successfully developed.
• The Random Forest classifier and feature selection proved to be the most effective components for this diagnostic task.
• This systematic methodology can be adapted for developing diagnostic models for other conditions using eNose data.