Performance Comparison and Analysis of Multiple Methods for Lung Cancer Detection Based on LUNA16

Zhongxing Ge

doi:10.54097/v9xq5368

Authors

Zhongxing Ge School of Communications and Information Engineering, Xi’an University of Posts & Telecommunications, Xi’an, Shaanxi, China

DOI:

https://doi.org/10.54097/v9xq5368

Keywords:

3D Swin Transformer; 3D Deep Learning; Micronodule Detection; Lung Cancer Detection; LUNA16.

Abstract

Lung cancer is one of the most common and deadly types of cancer worldwide. While computed tomography (CT) of the chest is considered the gold standard for early detection, manual evaluation faces two major challenges: Micronodules are often overlooked, and radiologists suffer from reduced efficiency due to overwhelming daily case numbers and persistent fatigue. To solve these problems, we employ the LUNA16 dataset (including 888 CTs, 1186 annotated nodules and more than 550K candidate samples) to build an integrated experimental platform and systematically compare four kinds of detection methods: traditional manual features, 2D deep learning, 3D deep learning and transformer model. The evaluation metrics include detection accuracy (AUC-ROC curve and recall), ability of micronodule detection, robustness on limited training data (10% / 25% / 50% data amount) and computation efficiency (parameter scale and frame rate). The experimental results show that 3D Swin Transformer T has the best performance. Compared with the traditional method HOG+Random Forest, test AUC-ROC reaches 0.948 and the recall of micro-nodule is 89.7%, which is 34.4% improvement. 3D ResNet-50 has a good robustness in small sample situation (AUC could still be 0.902 when using 10% data for training), and 2D ResNet-50 could get a balance between fast speed and accuracy (85 fps). The experimental results provide support for the selection of model in clinical computer-aided system for lung cancer detection.

Downloads

Download data is not yet available.

References

[1] Sung H, Ferlay J, Siegel R L, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA: A Cancer Journal for Clinicians, 2021, 71(3): 209-249.

[2] Lyu J, Ling S H. Using multi-level convolutional neural network for classification of lung nodules on CT images. In: 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE, 2018: 686-689.

[3] Wan C, Ma L, Liu X, Fei B. Computer-aided classification of lung nodules on CT images with expert knowledge. In: Medical Imaging 2021: Image-Guided Procedures, Robotic Interventions, and Modeling, Vol. 11598, 2021 Feb: 673-678. SPIE.

[4] Arenberg D. Micronodules detected on lung cancer screening CT scans. Journal of Thoracic Oncology, 2019, 14(9): 1501-1503.

[5] Ma W B, Yang Y, Fang W C. An effective tuberculosis detection system based on improved faster R-CNN with ROI align method. In: 2023 IEEE Biomedical Circuits and Systems Conference (BioCAS), 2023 Oct: 1-5. IEEE.

[6] Setio A A A, Traverso A, De Bel T, Berens M S, Van Den Bogaard C, Cerello P, et al. Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: the LUNA16 challenge. Medical Image Analysis, 2017, 42: 1-13.

[7] Liu Z, Lin Y, Cao Y, Hu H, Wei Y, Zhang Z, et al. Swin transformer: hierarchical vision transformer using shifted windows. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 10012-10022.

[8] Jaeger S, Candemir S, Antani S, et al. Two public chest X-ray datasets for computer-aided screening of pulmonary diseases. Quantitative Imaging in Medicine and Surgery, 2014, 4(6): 475.

[9] Dragon J M, Guha S, Salvatore M M. Hounsfield units: future applications in clinical practice, radiomics, and artificial intelligence. Clinical Imaging, 2024, 110: 110141. DOI: 10.1016/J.CLINIMAG.2024.110141.

[10] Shorten C, Khoshgoftaar T M. A survey on image data augmentation for deep learning. Journal of Big Data, 2019, 6(1): 1-48.

[11] Ronneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention, 2015 Oct: 234-241. Cham: Springer International Publishing.

[12] He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.

[13] Milletari F, Navab N, Ahmadi S A. V-net: fully convolutional neural networks for volumetric medical image segmentation. In: 2016 Fourth International Conference on 3D Vision (3DV), 2016 Oct: 565-571. IEEE.

[14] Dosovitskiy A. An image is worth 16x16 words: transformers for image recognition at scale. arXiv preprint, 2020. arXiv:2010.11929.