Enhanced lung cancer prediction using multi-layer perceptron with backpropagation and combined feature extraction techniques

Abstract

Lung cancer is characterized by the uncontrolled growth of abnormal cells in the lung tissue. Early detection of lung abnormalities is essential for efficient oxygen exchange and carbon dioxide removal. The application of AI to interpret CT scans for lung cancer detection has attracted significant attention due to its remarkable performance in medical imaging analysis. However, existing AI systems still struggle with low recognition accuracy and, consequently, limited applicability in lung cancer classification. The proposed study presents an AI system for lung cancer detection from CT images, operating automatically with very high accuracy. For this purpose, six CT image categories were included, comprising a normal case and various lung cancer stages. The images were first resized and normalized as part of the processing. After that, image feature extraction was performed using the Gray-Level Co-occurrence Matrix (GLCM) and Scale-Invariant Feature Transform (SIFT) methods. The features from GLCM, SIFT, and the combination of both were provided to the AI models, which included Support Vector Machine (SVM) and Multi-Layer Perceptron with Backpropagation (MLPBP). Combined GLCM and SIFT features with the MLPBP classifier achieved an accuracy of 89.9%, ensuring reliable lung cancer classification based on experimental evaluation.