Early Diagnosis of Lung Cancer Through VOC Detection Using an Affine Deep Learning E-Nose System

Abstract

Lung cancer continues to be a leading cause of cancer-related mortality globally, primarily due to late-stage diagnosis. Non-invasive diagnostic techniques, especially those based on exhaled breath analysis, have recently garnered significant interest. This study explores a novel approach to lung cancer detection by analyzing volatile organic compounds (VOCs) such as formaldehyde, ethanol, acetone, and chloroform present in exhaled breath. Signal processing techniques, including Wavelet Transform (WT) and Wavelet Packet Transform (WPT), were employed to generate and analyze signals derived from VOC data. Features extracted from these transformed signals were subsequently classified using various machine learning and deep learning models, namely Support Vector Machine (SVM), k-Nearest Neighbours (KNN), Random Forest (RF), Multi-Layer Perceptron (MLP), and Convolutional Neural Network (CNN). A comprehensive comparative evaluation was conducted to assess the impact of WT and WPT on model performance. Results indicate that WPT significantly enhances feature representation, improving classification accuracy across models. Among all evaluated models, the CNN outperformed others, achieving a maximum accuracy of 98.21% when applied to WPT-processed signals. This demonstrates the superior capability of deep learning models in handling complex VOC patterns and highlights their potential in clinical applications. The findings of this research affirm that integrating advanced signal processing techniques with deep learning can substantially improve the accuracy and reliability of non-invasive lung cancer diagnostics. The proposed method offers a promising, efficient, and patient-friendly alternative to conventional diagnostic procedures.