Improvement of image quality for pancreatic cancer using deep learning-generated virtual monochromatic images: Comparison with single-energy computed tomography. (May 2021)

• Record Type: Journal Article
• Authors: Ohira, Shingo; Koike, Yuhei; Akino, Yuichi; Kanayama, Naoyuki; Wada, Kentaro; Ueda, Yoshihiro; Masaoka, Akira; Washio, Hayate; Miyazaki, Masayoshi; Koizumi, Masahiko; Ogawa, Kazuhiko; Teshima, Teruki
• Is Part Of: Physica medica. Volume 85(2021)
• Page Start: 8
• Page End: 14
• Publication Date: 2021-05
• Subjects: Dual-energy CT -- Deep learning -- Virtual monochromatic image -- Pancreas; Medical physics -- Periodicals; Biophysics -- Periodicals; Biophysics -- Periodicals; Imagerie médicale -- Périodiques; Radiothérapie -- Périodiques; Rayons X -- Sécurité -- Mesures -- Périodiques; Physique -- Périodiques; Médecine -- Périodiques; 610.153
• DOI: 10.1016/j.ejmp.2021.03.035
• Languages: English
• ISSNs: 1120-1797

Abstract

Highlights: A deep learning model was developed to generate 60 keV VMI from single-energy CT. Pancreatic tumor imaging quality in 60 keV VMI is superior to SECT. 60 keV VMI generated from SECT may better support pancreatic tumor treatment planning. Abstract: Purpose: To construct a deep convolutional neural network that generates virtual monochromatic images (VMIs) from single-energy computed tomography (SECT) images for improved pancreatic cancer imaging quality. Materials and methods: Fifty patients with pancreatic cancer underwent a dual-energy CT simulation and VMIs at 77 and 60 keV were reconstructed. A 2D deep densely connected convolutional neural network was modeled to learn the relationship between the VMIs at 77 (input) and 60 keV (ground-truth). Subsequently, VMIs were generated for 20 patients from SECT images using the trained deep learning model. Results: The contrast-to-noise ratio was significantly improved ( p < 0.001) in the generated VMIs (4.1 ± 1.8) compared to the SECT images (2.8 ± 1.1). The mean overall image quality (4.1 ± 0.6) and tumor enhancement (3.6 ± 0.6) in the generated VMIs assessed on a five-point scale were significantly higher ( p < 0.001) than that in the SECT images (3.2 ± 0.4 and 2.8 ± 0.4 for overall image quality and tumor enhancement, respectively). Conclusions: The quality of the SECT image was significantly improved both objectively and subjectively using the proposed deep learning model for pancreatic tumors in radiotherapy.