Deep convolutional neural network for reduction of contrast-enhanced region on CT images

被引：11

作者：

Sumida, Iori ^{[1
]}

Magome, Taiki ^{[2
]}

Kitamori, Hideki ^{[3
,4
]}

Das, Indra J. ^{[5
]}

Yamaguchi, Hajime ^{[6
]}

Kizaki, Hisao ^{[6
]}

Aboshi, Keiko ^{[6
]}

Yamashita, Kyohei ^{[6
]}

Yamada, Yuji ^{[6
]}

Seo, Yuji ^{[1
]}

Isohashi, Fumiaki ^{[1
]}

Ogawa, Kazuhiko ^{[1
]}

机构：

[1] Osaka Univ, Dept Radiat Oncol, Grad Sch Med, 2-2 Yamada Oka, Suita, Osaka 5650871, Japan

[2] Komazawa Univ, Fac Hlth Sci, Dept Radiol Sci, Setagaya Ku, 1-23-1 Komazawa, Tokyo 1548525, Japan

[3] Kyushu Univ, Grad Sch Med Sci, Dept Hlth Sci, Higashi Ku, 3-1-1 Maidashi, Fukuoka, Fukuoka 8128582, Japan

[4] Osaka Univ, Dept Oral & Maxillofacial Radiol, Grad Sch Dent, 1-8 Yamada Oka, Suita, Osaka 5650871, Japan

[5] NYU, Dept Radiat Oncol, Langone Med Ctr, Laura & Isaac Perlmutter Canc Ctr, 160 E 34th St, New York, NY 10016 USA

[6] NTT West Osaka Hosp, Dept Radiat Oncol, Tennoji Ku, 2-6-40 Karasugatsuji, Osaka 5438922, Japan

来源：

JOURNAL OF RADIATION RESEARCH | 2019年 / 60卷 / 05期

基金：

日本学术振兴会;

关键词：

deep learning; convolution neural network; CT; contrast enhancement; ITERATIVE RECONSTRUCTION METHODS; COMPUTED-TOMOGRAPHY; RADIATION; EXPOSURE;

D O I：

10.1093/jrr/rrz030

中图分类号：

Q [生物科学];

学科分类号：

07 ; 0710 ; 09 ;

摘要：

This study aims to produce non-contrast computed tomography (CT) images using a deep convolutional neural network (CNN) for imaging. Twenty-nine patients were selected. CT images were acquired without and with a contrast enhancement medium. The transverse images were divided into 64 x 64 pixels. This resulted in 14 723 patches in total for both non-contrast and contrast-enhanced CT image pairs. The proposed CNN model comprises five two-dimensional (2D) convolution layers with one shortcut path. For comparison, the U-net model, which comprises five 2D convolution layers interleaved with pooling and unpooling layers, was used. Training was performed in 24 patients and, for testing of trained models, another 5 patients were used. For quantitative evaluation, 50 regions of interest (ROIs) were selected on the reference contrast-enhanced image of the test data, and the mean pixel value of the ROIs was calculated. The mean pixel values of the ROIs at the same location on the reference non-contrast image and the predicted non-contrast image were calculated and those values were compared. Regarding the quantitative analysis, the difference in mean pixel value between the reference contrast-enhanced image and the predicted non-contrast image was significant (P < 0.0001) for both models. Significant differences in pixels (P < 0.0001) were found using the U-net model; in contrast, there was no significant difference using the proposed CNN model when comparing the reference non-contrast images and the predicted non-contrast images. Using the proposed CNN model, the contrast-enhanced region was satisfactorily reduced.

引用

页码：586 / 594

页数：9