3D CNN Lung Cancer Classification with Dual feature fusion

Lung cancer remains a significant cause of cancer-related fatalities worldwide, emphasizing the critical need for early diagnosis to improve patient outcomes and reduce mortality rates. Recent advancements in deep learning techniques have shown promising results in medical image analysis, particularly in lung cancer identification. In this paper, we propose a novel approach that integrates a multi-kernel 3D Convolutional Neural Network (CNN) with dual-feature fusion using the Histogram of Oriented Gradients (HOG) and Local Binary Patterns (LBP) for lung cancer classification. Our comprehensive evaluation across three distinct datasets (IQ-OTH/NCCD, LIDC-IDRI, and NSCLC-Radiomics) demonstrates the method's robustness and generalizability, achieving 99.16% accuracy on the primary dataset. The fusion strategy combines the strengths of CNNs in automated feature extraction with the descriptive power of HOG in capturing shape information and LBP in characterizing local texture patterns, resulting in statistically significant improvements (p<0.01) over existing single-dataset approaches. This multi-scale, multi-feature integration addresses key limitations in current pulmonary nodule analysis while maintaining clinical applicability.