Proceedings of The 3rd International Academic Conference on Research in Engineering and Technology
Year: 2024
DOI:
[PDF]
An Explainable Transformer-Based Method for Lung Cancer Classification
Oguzhan Katar, and Ozal Yildirim
ABSTRACT:
Lung cancer is one of the leading causes of cancer-related mortality worldwide. Therefore, developing early diagnosis and effective treatment strategies is critically important. In recent years, deep learning methods have made significant advancements in this field, but most methods are often limited by their ‘black box’ nature. In this study, we propose a Vision Transformer (ViT)-based method for lung cancer classification. Our proposed method classifies lung computed tomography (CT) images into benign, malignant, and normal categories. Additionally, the focus areas of these predictions are visualized using the Eigen-CAM algorithm. The pre-trained ViT model was fine-tuned and validated on a public dataset (IQ-OTH/NCCD) comprising of 1,097 samples. The model achieved an accuracy rate of 98.18% on test images. Experimental results demonstrate that the proposed method provides high accuracy in lung cancer classification and could be a valuable tool in clinical applications due to its explainability attribute. The proposed method allows for visualization of the pixels that the model focuses on during the decision-making process, enabling clinical experts to better understand the model’s outputs. This method could reduce the workload of radiologists in the classification and localization of lung cancer.
keywords: deep learning, diagnostics, explainable ai, medical image analysis, radiology