Deep Radiomics-Based Pulmonary Tumor Stratification via Optimized VGG16 Feature Extraction and Classification

Early diagnosis of pulmonary malignancies is critical for reducing lung-cancer-associated mortality; however, conventional diagnostic workflows suffer from subjectivity, latency, and limited sensitivity. This research introduces a transfer-learning–driven VGG16 diagnostic framework optimized for high-resolution CT imaging. The proposed model leverages multi-level convolutional abstractions, domain-specific augmentation strategies, and fine-grained classifier refinement to enhance nodule discriminability. Quantitative results indicate a performance efficiency of up to 95% accuracy with superior generalization across heterogeneous CT datasets. The study establishes VGG16 as a powerful radiomics engine capable of supporting real-time, computer-aided diagnosis (CAD) systems, thereby providing a scalable and clinically relevant solution for automated lung malignancy detection. Lung cancer continues to be one of the most significant causes of cancer-related mortality worldwide, primarily due to delays in diagnosis and the subtle nature of early-stage symptoms. Improving the accuracy and efficiency of lung cancer detection is therefore critical for enhancing patient survival rates. In this study, a deep learning–based classification framework is proposed using the VGG16 convolutional neural network with transfer learning to detect and categorize lung cancer from CT scan images. The methodology includes systematic preprocessing, image normalization, and extensive data augmentation to enhance model robustness and reduce overfitting. The VGG16 architecture is fine-tuned to extract high-level radiographic features, enabling accurate discrimination between benign and malignant nodules. Experimental results demonstrate that the proposed model achieves an accuracy ranging between 80% and 95%, depending on the dataset composition and tuning parameters. The findings indicate that VGG16 provides reliable and computationally efficient feature extraction compared to traditional machine learning approaches. This research highlights the practical potential of incorporating deep learning models into clinical decision-support systems to assist radiologists in early detection, reduce false diagnoses, and improve overall diagnostic performance. The framework offers a non-invasive, scalable, and cost-effective solution for supporting modern medical imaging workflows.