Breath-based lung cancer detection using an ML-driven low-cost sensor array

Abstract

Abstract Lung cancer is the leading cause of cancer-related mortality worldwide. Lately, electronic nose (e-nose) systems have emerged as a promising method for non-invasive lung cancer detection. These systems, however, have several limitations, including low accuracy rates and long detection times. To address these challenges, we conducted a pilot study involving the development of an affordable e-nose device that can detect more than 30 volatile organic compounds, using twelve metal oxide semiconductor sensors and one chemi-resistive alkane sensor. The device recorded data for 28 healthy controls and 18 lung cancer breath samples that were then analyzed using a multilayer perceptron neural network. The dataset was expanded through a novel use of data augmentation, where Gaussian noise was applied to generate synthetic samples while preserving the original data’s statistical properties. The model was evaluated by 5-fold cross-validation and achieved an accuracy of 96.26%, sensitivity of 92.88%, specificity of 97.75%, and an area under the curve of 0.9286. Our system outperforms existing e-nose detection methods by more than 5% and is capable of classifying in approximately 5 minutes. These findings highlight the potential of this breath analyzer system as a rapid and cost-effective tool for preliminary lung cancer screening.