Abstract

About 20% of cancer deaths wordwide are caused by lung cancer, making it one of the major causes of cancer-related mortality. For lung cancer to be diagnosed and treated effectively, early and precise identification of lung nodules in medical images is essential. While vision transformers (ViTs) are skilled at modeling global contextual information, traditional convolutional neural networks (CNNs) excel at capturing local features. In this work, a hybrid deep learning model that combines ViT-Tiny and ResNet18 as a CNN backbone is proposed for better feature representation in lung computed tomography (CT) images. After combining features from both backbones using a weighted sum fusion technique, the model uses a linear classifier for binary lung nodule classification. The model is evaluated using a benchmark lung nodule dataset, and the results demonstrate the effectiveness of the proposed model. The proposed model performed exceptionally well on the hold-out test set, with precision of 0.9889, accuracy of 0.9905, recall of 0.9926, and an F1-score of 0.9907.

Authors

S. Saranya, R. Rajeswari, T. Ramesh
Bharathiar University, India

Keywords

Lung Cancer, Hybrid CNN-VIT, Vision Transformer, Preprocessing, Tiny VIT