Enhanced Lung Cancer Detection Using Advanced Deep Learning Techniques

Abstract

Lung cancer is one of the leading causes of mortality worldwide, making early detection and accurate classification crucial for effective treatment. This research investigates the use of deep learning (DL) models for lung cancer detection through histopathological images. The study looks at three main models Convolutional Neural Networks (CNNs), U-Net, and Region-based CNN (R-CNN), along with different methods to extract features like Gabor Filtering, Local Binary Pattern (LBP), Wavelet Transform, Scale-Invariant Feature Transform (SIFT), and Harris Corner Detection. A balanced dataset comprising 5,000 histopathological images-2,500 lung adenocarcinoma (lung-aca) and 2,500 normal lung (lung-n) images were used for classification. Among the models tested, U-Net with LBP achieved the highest performance with 99.00% accuracy, 98.00% sensitivity, 99.00% specificity, and an AUC of 0.99. U-Net also performed exceptionally with Wavelet Transform and SIFT, yielding accuracies of up to 98.00% and an AUC of 1.00. CNNs showed good performance, particularly with Wavelet Transform and SIFT 96.00% accuracy, while R-CNN demonstrated reasonable accuracy with Wavelet Transform 93.00% and Gabor Filtering 91.00%. However, R-CNN struggled with Harris Corner Detection, achieving only 53.00% accuracy. This research emphasizes the importance of selecting appropriate feature extraction techniques to improve the accuracy of DL models in medical image analysis. While DL models show great potential for lung cancer detection, challenges such as high computational demand, feature extraction inconsistencies, and model interpretability need to be addressed. Future work may involve ensemble learning, 3D medical imaging, and advancements in explainable AI to further enhance model performance and clinical applicability. Keywords- Lung cancer Detection, Deep learning models, Histopathological image analysis.