Lung Cancer Detection From CT Images With Clinical Feature Fusion

Abstract

Lung cancer diagnosis from computed tomography (CT) images is challenging because early abnormalities may be subtle and manual review of large image volumes is time-consuming. The supplied thesis develops a web-based computer-aided diagnosis system that integrates CT image preprocessing, lung segmentation, image-feature extraction, and Convolutional Neural Network (CNN) classification. Python, Flask, TensorFlow, Keras, OpenCV, HTML, CSS, JavaScript, and SQLite are used to support image upload, preprocessing, inference, result visualization, and prediction-history storage. The system classifies CT images into Normal and Lung Cancer categories and reports confidence scores. All 12 functional test cases in the thesis passed, including image upload, preprocessing, segmentation, feature extraction, CNN prediction, report generation, and end-to-end workflow execution. The epoch-wise accuracy graph reaches 95.9% after 20 epochs. A separate model-performance dashboard reports 94.8% overall accuracy, 96.5% sensitivity, 93.2% specificity, and an AUC-ROC score of 0.975; a five-fold cross-validation summary reports 94.8% average validation accuracy. The developed framework is presented as a decision-support tool intended to assist, rather than replace, qualified healthcare professionals.