An Optimized Hybrid Fusion Model Combining Dense and Visual Geometry Networks with Machine Learning for Lung Cancer Detection and Classification

Abstract

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, underscoring the critical need for accurate and early diagnostic solutions. By integrating Densely Connected Convolutional Network (DenseNet), Visual Geometry Group Network 19 (VGG19), and Multi-Class Support Vector Machine (MCSVM) into a hybrid architecture termed OncoFusion-Net, this study introduces a robust framework for the early and precise detection and classification of lung cancer. To enhance image contrast and suppress noise while preserving essential structural information, high-resolution Computed Tomography (CT) images are preprocessed using Anisotropic Diffusion Filtering (ADF) and Contrast Limited Adaptive Histogram Equalization (CLAHE). Segmentation is accomplished through Kapur’s Entropy-Based Thresholding Method, effectively isolating suspicious regions of interest. For feature extraction and representation, OncoFusion-Net harnesses the combined advantages of VGG19 and DenseNet, ensuring efficient feature reuse and propagation. The resulting features are classified using MCSVM, facilitating accurate differentiation among benign, malignant, and other lung cancer types. The model’s performance is rigorously validated on independent datasets, demonstrating high accuracy, sensitivity, and reliability, further confirmed through Receiver Operating Characteristic (ROC) curve analysis. Overall, this comprehensive framework advances lung cancer diagnosis by enabling early detection and supporting personalized treatment planning to improve patient outcomes.