Akademik Veri Yönetim Sistemi
2025 10th International Conference on Computer Science and Engineering (UBMK), İstanbul, Türkiye, 15 - 21 Eylül 2025, cilt.1, sa.1, ss.1-6, (Tam Metin Bildiri)
Lung cancer is one of the most common malignancies worldwide and has the highest mortality rate. In this study, three open access datasets, namely IQ-OTH/NCCD, SPIE-AAPM-NCI and CT-Scan, were used to conduct the diagnosis of lung cancer by automatic analysis of computed tomography (CT) images. Images were subjected to various pre-processing steps and evaluated by 5-fold cross-validation method. In the study, a total of seven deep learning models based on convolutional neural network (CNN), vision transformer (ViT) architecture and hybrid structure were statistically compared in terms of performance before and after data augmentation. Classification performance was measured by accuracy, precision, sensitivity and F1 score; the results were analyzed by independent samples t-test at %90 confidence level. SwinV2-CR-Small-224 model showed the highest success without data augmentation, and ConvNeXtV2-Base model showed the highest success after data augmentation. With the findings obtained, it has been revealed that ViT-based architectures offer clinically significant high classification performance and deep learning-based decision support systems have a strong potential in lung cancer diagnosis.