Automated Chest Cancer Detection and Classification Using Deep Learning

Early detection of chest cancer is critical for effective treatment and improved patient outcomes. This study proposes a structured workflow for automated chest cancer detection using a novel hybrid deep learning model, CTAF-Net. The workflow begins with robust preprocessing of CT scan data, including resizing, normalization, and data augmentation, to ensure high-quality inputs for the model, spatial features such as edges, textures, and shapes are extracted from the pre-processed images. To improve model efficiency and performance, Ant Colony Optimization (ACO) is employed to select an optimal subset of features from the extracted set, reducing redundancy and computational complexity. Finally, the CTAF-Net model, integrating Convolutional Neural Networks (CNN) and Vision Transformer modules, classifies the selected features to accurately identify cancerous and non-cancerous cases. Finally, the performance evaluation stage assesses the model using metrics such as accuracy, precision, recall, and F1-score, providing quantitative evidence of its effectiveness. The proposed deep learning framework outperforms baseline models, achieving 99% accuracy, 98.9% precision, 99% recall, and 98.95% F1-Score, representing an improvement of 0.9–1% over the best baseline (InceptionV3). This demonstrates enhanced sensitivity to subtle tumor patterns, reduced false predictions, and superior overall classification performance.