The following are highlights from the current issues of RSNA’s two peer-reviewed journals.
Automated Abdominal Segmentation of CT Scans for Body Composition Analysis Using Deep Learning
While body composition is linked to clinical outcomes in conditions including cancer, cardiovascular disease and after major surgery, the impact of body composition on these conditions is poorly understood. CT and MRI are already part of the clinical workup for many diseases, but body composition is not routinely calculated as it requires laborious segmentation or tracing of abdominal compartments.
In an article published online in Radiology (RSNA.org/Radiology), Alexander D. Weston, BS, Mayo Clinic, Rochester, MN, and colleagues developed and evaluated a fully automated algorithm for segmenting the abdomen from CT to quantify body composition.
Researchers trained a convolutional neural network based on the U-Net architecture to perform abdominal segmentation on a data set of 2,430 2D CT examinations and tested it on 270 CT examinations. It was further tested on a separate data set of 2,369 CT examinations from patients undergoing treatment for hepatocellular carcinoma (HCC).
Compared with reference segmentation, the model for this study achieved Dice scores (mean ± standard deviation) of 0.98 ± 0.03, 0.96 ± 0.02 and 0.97 ± 0.01 in the test set and 0.94 ± 0.05, 0.92 ± 0.04 and 0.98 ± 0.02 in the HCC data set, for the subcutaneous, muscle and visceral adipose tissue compartments, respectively.
The model performance met or exceeded the accuracy of expert manual segmentation of CT examinations for both the test data set and the HCC data set. The model generalized well to multiple levels of the abdomen and may be capable of fully automated quantification of body composition metrics in 3D CT examinations.
“Despite the fact that our model is trained solely on 2D sections at the L3 level, we observed our model at other levels of the abdomen. This suggests that our algorithm is a useful tool for performing fully automated 2D segmentation despite being trained on 2D data. Using our tool to perform body composition analysis on several patients, we observed high variability in anatomy at a single section, suggesting that single-section analysis is of limited utility and that 3D analysis is a more accurate method,” the authors conclude.
An example of 3D segmentation using the author’s algorithm. The 3D scan was subsampled into a series of 2D images, and fully automated segmentation based on deep learning was performed on the series of images. Several views of the 3D volume are shown to demonstrate the model’s ability to generalize across multiple sections.
Weston et al, Radiology 2019 ©RSNA 2019
Selective Chemoembolization of Caudate Lobe Hepatocellular Carcinoma: Anatomy and Procedural Techniques
Transarterial chemoembolization is the most common treatment for unresectable hepatocellular carcinomas (HCCs). However, when an HCC is located in the caudate lobe, interventional radiologists may be reluctant to perform chemoembolization and percutaneous ablation due to the tumor’s complex vascular supply and deep location. With the advent of C-arm CT, rendering the 3D display of the hepatic artery and detecting the tumor-feeding vessels are possible and can guide interventional radiologists to the tumor.
In the January-February issue of RadioGraphics (RSNA.org/RadioGraphics), Hyo-Cheol Kim, MD, of Seoul National University College of Medicine, Korea, and colleagues review the anatomy of the caudate artery with C-arm CT and describe basic technical considerations in selective chemoembolization for caudate lobe HCCs.
HCCs are frequently fed by multiple caudate arteries, and the arteries are almost always connected to each other and to the A4 branch, making a communicating vascular arcade between the right hepatic artery and left hepatic artery. The origins of the tumor-feeding arteries of a caudate lobe HCC can vary depending on the subsegmental location of the tumor.
In general, chemoembolization is not recommended for patients with HCCs with portal vein thrombus or bile duct invasion.
Although it can be difficult to spot the caudate artery at digital subtraction angiography (DSA), C-arm CT images obtained at the proper or common hepatic artery can demonstrate the origin of the caudate artery and help render its 3D vascular anatomy. As the caudate artery frequently originates from a large vessel at an acute angle, pre-shaping the microcatheter by manual bending or steam-heating prior to performing the procedure is useful for selective catheterization.
“Using the meticulous shepherd’s hook technique for caudate artery chemoembolization while having knowledge of the artery’s anatomy, enables the safe and effective treatment of caudate lobe HCCs,” the authors write.
Figure 12. HCC in the caudate process in a 51-year-old man. (a) Axial arterial phase CT image shows a small HCC (arrowhead) in the caudate process. (b) Common hepatic angiogram shows a tumor in the caudate process (arrowhead). (c) Three-dimensional volume-rendered C-arm CT image shows a caudate lobe HCC fed by the caudate artery (arrowhead), which originates from the segment VII hepatic artery.
Kim et al, RadioGraphics 2019:39;1 © RSNA 2019
Listen to Radiology Editor David A. Bluemke, MD, PhD, discuss this month’s research you need to know. Podcasts summarize the importance and context of selected recent articles. Subscribe today at RSNA.org/Radiology-Podcasts and never miss a single episode.
February articles include:
“Mammographic Breast Density Assessment Using Deep Learning: Clinical Implementation. Lehman CD,” Yala A, Schuster T., et al.
“Quantitative Assessment of Liver Function by Using Gadoxetic Acid-enhanced MRI: Hepatocyte Uptake Ratio,” Yoon JH, Lee JM, Kang H, et al.
Listen to RadioGraphics Editor Jeffrey S. Klein, MD, and authors discuss the following articles from recent issues of RadioGraphics at RSNA.org/RG-Podcasts.
“In-Phase and Opposed-Phase Imaging: Applications of Chemical Shift andMagnetic Susceptibility in the Chest and Abdomen,” Shetty AS, et al.
“An Analysis of Nipple Enhancement at Breast MRI with Radiologic-Pathologic Correlation,” Gao Y., et al.
Audio summary podcasts (also available on ITunes and Google Play) include these studies:
“Complications of Intravesical BCG Immunotherapy for Bladder Cancer,” Green DB, et al.
“Leukemic Involvement in the Thorax,” Shroff GS, et al.
“Connective Tissue Disorders in Childhood: Are They All the Same?” Navallas M, et al.
Prepare Submissions for Radiology: Cardiothoracic Imaging and Radiology: Imaging Cancer
Radiology: Cardiothoracic Imaging is seeking manuscripts that emphasize research advances and technical developments in medical imaging that drive cardiothoracic medicine.
Radiology: Imaging Cancer will accept submissions that cover the best clinical and translational cancer imaging studies across organ systems and modalities, including leading-edge technological developments.
Available exclusively online, Radiology: Cardiothoracic Imaging will launch in mid-2019 and Radiology: Imaging Cancer will launch in late 2019.
For more information on the submission process, go to RSNA.org/Journals.