Abstract
Introduction: Lung carcinoma is a leading cause of death. Early detection of the disease will improve the survival rate. An easy, inexpensive method for early detection or even screening at the practitioners office is lacking. Electronic noses use arrays of chemosensors for patterns describing individual breath. This abstract describes a proof of principle study for discriminating NSCLC from COPD using an electronic nose.
Methods: 66 newly diagnosed (nd) NSCLC patients in the lung cancer group were compared to 73 COPD patients using a Cyranose 320 (Smith Detection) electronic nose. Gas samples were taken immediately after exhalation and raw files were transferred to a self developed pattern recognition software (DiagNose) to learn a common pattern for each ndNSCLC and COPD and then group each file.DiagNose software empoys neuronal network principles for pattern recognition.
Results: Pattern analysis of all files correctly identified 62 of the 73 samples as COPD and 59 of the 66 samples as Carcinoma patients (sensitivity 84%; specificity 89%).
Diskussion Patterns derived from breath analysis of NSCLC and COPD patients analyzed with a special software based on neuronal network techniques were able to correctly discriminate between cancer and COPD patients in the great majority of cases. Sensitivity and Specificity were both high. Further improvements with even more sophisticated algorithms may further improve these results. Our study demonstrates the potential of pattern recognition for early detection of lung carcinoma in a populationat risk.
- © 2011 ERS