Deep Learning Algorithm for Reducing CT Slice Thickness: Effect on Reproducibility of Radiomic Features in Lung Cancer

Authors
Sohee Park, MD, Sang Min Lee, MD, Kyung-Hyun Do, MD, June-Goo Lee, PhD, Woong Bae, PhD, Hyunho Park, MD, Kyu-Hwan Jung, PhD, and Joon Beom Seo, MD
Journal
Korean Journal of Radiology
Related Product

Research

Date Published
2019.08
Summary

This study evaluates the impact of CT slice thickness on the reproducibility of radiomic features (RFs) in lung cancer and explores whether convolutional neural network (CNN)-based super-resolution (SR) algorithms can enhance reproducibility. CT images with 1-, 3-, and 5-mm slice thicknesses from 100 lung cancer patients were analyzed. A CNN-based SR algorithm was developed to convert thicker slices into 1-mm slices. Semi-automatic segmentation of lung cancers was performed using AVIEW software, allowing extraction of 702 RFs, including tumor intensity, texture, and wavelet features. The study found that reproducibility of RFs decreased with increasing slice thickness, with mean concordance correlation coefficients (CCCs) of 0.41 (1 mm vs. 3 mm), 0.27 (1 mm vs. 5 mm), and 0.65 (3 mm vs. 5 mm). Tumor intensity features were the most reproducible, while wavelets were the least. Applying the CNN-based SR algorithm significantly improved reproducibility, with mean CCCs rising to 0.58, 0.45, and 0.72, respectively. The percentage of reproducible RFs increased substantially after SR application.

Contact

Please leave your inquiry if you have any questions regarding our products, recruitment, investment, or any other matters.

Contact us