COMPARATIVE ANALYSIS OF DEEP LEARNING ARCHITECTURES FOR LUNG CANCER DETECTION

Abstract

Lung cancer remains one of the leading causes of cancer-related deaths worldwide, emphasizing the critical need for early and accurate detection. This thesis presents a comparative analysis of three deep learning architectures—Convolutional Neural Network (CNN) from scratch, pre-trained CNN, and Vision Transformer (ViT)—to identify the most effective model for lung cancer detection using medical imaging. By evaluating these models on key performance metrics such as accuracy, precision, recall, F1-score, training time, and model size, the study provides a comprehensive assessment of their strengths and limitations. The pre-trained CNN emerged as the top performer, achieving an impressive accuracy of 99%, demonstrating the significant advantages of transfer learning in medical diagnostics. The Vision Transformer also showed strong performance with an accuracy of 96%, albeit with higher computational demands. The CNN from scratch offered the benefits of shorter training times and smaller model size, making it suitable for resource-constrained environments. These findings underscore the importance of leveraging advanced deep learning models for early lung cancer detection, which can significantly enhance patient outcomes. The study also highlights potential areas for future research, including ensemble learning techniques, advanced architectural modifications, and the incorporation of diverse imaging modalities. This work contributes to the ongoing efforts to harness artificial intelligence in healthcare, aiming to improve diagnostic accuracy and early intervention strategies for lung cancer.