<![CDATA[Lung cancer ranks among the primary contributors to cancer-related deaths globally, highlighting the need for accurate and efficient detection methods to enable early diagnosis. However, deep learning models such as VGG16 and VGG19, commonly used for CT scan image classification, often face challenges related to class imbalance, resulting in classification bias and reduced sensitivity to minority classes. This study contributes by proposing an integration of the VGG architecture and Generative Adversarial Networks (GANs) to improve lung cancer classification performance through balanced and realistic synthetic data augmentation. The proposed approach was evaluated using two datasets: the IQ-OTH/NCCD Dataset, which classifies patients into Benign, Malignant, and Normal categories based on clinical condition, and the Lung Cancer CT Scan Dataset, annotated with histopathological labels: Adenocarcinoma, Squamous Cell Carcinoma, Large Cell Carcinoma, and Normal. The method involves initial training of the VGG model without augmentation, followed by GAN-based data generation to balance class distribution. The experimental results show that, prior to augmentation, the models achieved relatively high overall accuracy, but with poor performance on minority classes (marked by low precision and F1-scores and FPR exceeding 8% in certain cases). After augmentation with GAN, all performance metrics improved dramatically and consistently across all classes, achieving near-perfect precision, TPR, F1-score, and overall accuracy of 99.99%, and FPR sharply reduced to around 0.001%. In conclusion, the integration of GAN and VGG proved effective in overcoming data imbalance and enhancing model generalization, making it a promising solution for AI-based lung cancer diagnostic systems.]]>
Copyrights © 2025
