A Comparative Study on Lung Nodule Detection: 3D CNN vs Vision Transformer

Early detection of lung cancer is essential to reduce mortality. CT scans are an effective imaging technique for detecting lung cancer but often produce false positives that can lead to unnecessary invasive procedures. This study compared the reduction of false positives in the detection of lung nodules using two 3D methods: CNN and ViT. This study using LUNA16 dataset obtained the results that CNN, with the ability to capture complex spatial features, showed superior performance in reducing false positives compared to ViT, with an accuracy of 92%, precision of 84%, sensitivity of 95%, specificity of 90%, Fl-score of 89%, and false positives value of 40. In addition, CNN has a shorter training time than ViT, which requires higher computation and longer training time. Therefore, 3D CNN is more effective in reducing false positives in detecting lung nodules.