AUTHOR=Wang Hongfeng , Zhu Hai , Ding Lihua 

TITLE=Accurate classification of lung nodules on CT images using the TransUnet

JOURNAL=Frontiers in Public Health

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2022.1060798

DOI=10.3389/fpubh.2022.1060798

ISSN=2296-2565

ABSTRACT=Background: Computed tomography (CT) is an effective way of scanning lung cancer. Classification of lung nodules in CT screening is totally doctor-dependent, which has drawbacks including difficulty classifying tiny nodules, subjectivity, and high false positive rates. Deep convolutional neural networks, a deep learning technology, have shown effective in medical imaging diagnosis in recent years. Herein, we propose a deep convolutional neural network technique (TransUnet) to automatically classify lung nodules accurately.
Methods: The TransUnet is consists of three parts, transformer, Unet, and global average pooling (GAP). The transformer encodes discriminative features via global self-attention modeling on CT image patches. The Unet, which collects context by constricting route, enables exact lunge nodule localization. The GAP is used to categorize CT images, allowing each sample to be assigned a score. Python was employed to pre-process all CT image in the LIDI-IDRI, and the obtained 8474 images (3259 benign and 5215 lung nodules) was used to evaluate the performance of the method.
Results: The accuracies of TransUnet in the training set and testing set were 87.90% and 84.62%. The sensitivity, specificity, and AUC of proposed TransUnet on the testing dataset achieved 70.92%, 93.17%, and 0.862 (0.844-0.879). We also compared the TransUnet to three well-known methods, and the TransUnet did the best of all of these methods.
Conclusions: The experimental results on LIDI-IDRI demonstrated that the proposed TransUnet has great performance in classifying lung nodules and it has great potential application in the diagnosis of lung cancers.