Enhancing Lung Cancer Detection: A Framework for Lung Parenchyma Segmentation and Classification Using Transfer Learning

Lung cancer continues to be one of the most prevalent causes of cancer-related fatalities worldwide, with a high mortality rate attributed to late-stage diagnosis. Early detection is essential for enhancing patient outcomes, as it enables the implementation of appropriate therapies that can substantially increase survival rates. In the past few years, artificial intelligence (AI)-based methods have emerged as powerful instruments to assist radiologists, enabling them to conduct diagnoses that are more consistent, efficient, and accurate. This study introduces a comprehensive framework for the detection of lung cancer that combines deep learning models with advanced segmentation techniques, utilizing CT images. This framework implements techniques such as data augmentation, segmentation techniques, and transfer learning to improve diagnostic accuracy by addressing critical issues such as class imbalance and juxtapleural nodule segmentation. Among the models tested, ResNet50 outperformed EfficientNetB0, DenseNet121, and VGG16, achieving an F1 score of 93. 10% and an accuracy of 93. 04%.