Creation of synthetic contrast-enhanced computed tomography images using deep neural networks to screen for renal cell carcinoma
概要
Recently, the number of accidentally discovered small-diameter renal tumors has increased1
More than 50% of kidney tumors are asymptomatic or discovered while screening for other
illnesses.2,3 It is necessary to perform both plain computed tomography (CT), which does not
use a contrast medium, and contrast-enhanced CT (CECT), which uses a contrast medium, to
diagnose renal cell carcinoma (RCC).4 The methods of CECT were determined by different
clinical indications according to the renal protocol, balancing diagnostic accuracy and radiation
exposure.5 CECT shows the blood flow, blood flow velocity, degree of capillary development,
and stromal status by comparing the Hounsfield units (HU) before and after the injection of the
contrast medium; an enhancement of the contrast effect by ≥15 HU when compared with plain
CT indicates the presence of a kidney tumor.6 Additionally, CECT angiography is useful for
visualizing the location of blood vessels before surgery.7 However, the use of a contrast medium
is contraindicated in patients with contrast medium-related allergies and moderate or greater
renal dysfunction.8 Moreover, RCC also occurs in younger patients, and thus, these patients are
subjected to frequent medical exposure to CT during screening and follow-up following radical
surgery. Imaging methods aimed at reducing medical exposure have been attempted previously.9
Magnetic resonance imaging (MRI) is recommended to reduce the risk of secondary carcinogenesis owing to medical exposure.10-12 MRI is useful for determining the presence or absence of
tumor thrombus in inferior vena cava in patients with RCC. MRI, including diffusion-weighted
imaging, is very useful in diagnosing kidney cancer and also useful in that there is no radiation
exposure. However, plain CT is taken frequently for screening in many clinical situations, including the emergency room and medical practitioner due to the shorter examination time than MRI.
These CT scans may be useful for diagnosing renal tumors. Imaging modalities with reduced
exposure doses and better image detection capabilities for screening small renal tumors have
not yet been developed.13 Additionally, the European Association of Urology (EAU) guidelines
recommend the development of a postoperative CT schedule according to the risk and frequency
of RCC recurrence-based tumor staging to reduce medical exposure.14
The progress in image composition technology has been remarkable. There have been many
reports in the medical field on improving diagnostic imaging assistance using artificial intelligence
(AI). AI is used to distinguish between benign and malignant renal tumors.15-18 Some studies have
sought to determine the grade and type of malignant and nuclear atypia of RCC.19,20 However,
all studies utilizing AI have used previously obtained CECT images and not image composition
technology. Furthermore, while previous studies have also reported CT image generation by
image-to-image translation using deep neural networks (DNNs),21 there have been no reports on
synthetic CECT images created for the purpose of reducing medical exposure and avoiding the
use of a contrast medium. In this study, we first created a DNN based on plain CT images. We
subsequently aimed to evaluate whether a synthetic CECT image created using the DNN could
Nagoya J. Med. Sci. 85. ...