Journal highlights

The following are highlights from the current issues of RSNA’s peer-reviewed journals.

RSNA Publications Brochure Cover
 

RSNA Journals Move to All-Digital Format

We know our members increasingly enjoy the convenience of accessing our peer-reviewed journals online. To support the growth of dynamic, multimedia-rich content and acknowledge reader preferences and their changing needs, Radiology and RadioGraphics will soon be available exclusively online.

The final RadioGraphics print issue will be the November/December issue. The final Radiology print issue will be its 100th anniversary issue in January 2023.

RSNA digital journal subscriptions offer all the peer-reviewed research, authoritative reviews, editorials and commentaries, and quality education readers have come to expect from the print journals. Digital access allows for the bonus of exceptional multimedia content including podcasts, slides and videos.

For more information about journal subscriptions, contact the RSNA customer service team by phone at 1-877-776-2636 or 1-630-571-7873 (outside U.S. or Canada), or via email at customerservice@rsna.org.

Conveniently access all RSNA journals online from anywhere in the world at RSNA.org/Journals.

Radiology Logo

Comparison of Quantitative Liver US and MRI in Patients with Liver Disease

Nonalcoholic fatty liver disease is the most common cause of liver disease in children. While biopsy is the reference standard for diagnosing and defining liver disease severity, imaging methods like US shearwave elastography and MR elastography have shown promise for detecting and characterizing pediatric chronic liver disease.

In a prospective study published in Radiology, Vinicius P. V. Alves, MD, Cincinnati Children’s Hospital Medical Center, and colleagues defined the associations between quantitative US measures and MRI measures of chronic liver disease in children, adolescents and young adults with known or suspected chronic liver disease. The authors further sought to define the predictive ability of quantitative US measures to detect abnormal liver stiffening and steatosis according to the MRI reference standard.

Participants aged 8-21 years, with known or suspected liver disease and a body mass index less than 35 kg/m2, underwent 1.5-T MRI and quantitative liver US at a pediatric academic medical center. The study authors compared US parameters with liver MR elastography and liver MRI proton density fat fraction. They found that for individuals in this age range with known or suspected liver disease, there was moderate to high correlation between US shear-wave speed (SWS) and MR elastography-derived stiffness. US SWS predicted an abnormal liver shear stiffness with high performance.

“Quantitative MRI and US play an important and growing role in the assessment of liver disease and our study provides important data on the diagnostic performance of quantitative US in children with chronic liver disease, using MRI as a reference standard,” the authors conclude.

To read the full article, go to RSNA.org/Radiology. Follow the Radiology editor on Twitter @RadiologyEditor.

Fig 1 Alves

Representative MR elastography and quantitative US images in a 16-year-old boy with Fontan-associated liver disease and elevated liver shear stiffness (5.4 kPa with gradient-recalled echo MR elastography). (A) Axial MR elastogram with 95% confidence map overlay shows a stiff (5.4 kPa) heterogeneous liver. Image colors are indicative of stiffness (in kilopascals) according to the scale in the left of the elastogram image. (B) Transverse two-dimensional shear-wave elastography US image with shear-wave speed of 1.73 m/sec (9.0 kPa). (C) Transverse shear-wave dispersion map with dispersion of 16.86 m/sec/kHz. (D) Split-screen transverse image shows liver attenuation measurement of 0.46 dB/cm/MHz. (E) Split-screen transverse image shows normalized local variance measurement of 1.24. (F) Split-screen longitudinal image shows hepatorenal index measurement of 1.07.

https://doi.org/10.1148/radiol.212995 ©RSNA 2022

Radiograpics

Imaging Cancer in Pregnancy

Pregnancy-associated cancer (PAC) is defined as cancer that is detected during pregnancy and up to one year postpartum. Although rare, about 1 in 1,000 pregnancies, PAC is increasing because of postponed childbearing and advanced maternal age at conception.

Detection is based on signs and symptoms, results of laboratory analyses and incidental findings during routine obstetric imaging. The care of PAC necessitates balancing the potential benefits of diagnosis and treatment of the mother with risks to the fetus.

In an article published in RadioGraphics, Priyanka Jha, MBBS, Department of Radiology and Biomedical Imaging, University of California, San Francisco, and colleagues provide an update of the imaging triage, safety considerations, cancer-specific imaging and treatment options for cancer in pregnancy. According to the authors, the most common PACs are breast cancer (41%), lymphoma (12%), uterine cervical cancer (10%), leukemia (8%) and ovarian cancer (7%).

Because the benefits of timely imaging diagnosis of PACs must be weighed against related risks to the fetus and treatment-related fetal side effects, screening for PACs and congenital fetal anomalies is typically performed with US and noninvasive prenatal tests. However, the use of other modalities such as CT or nuclear medicine requires a careful assessment of risk versus benefit from radiation exposure.

“This updated review of diagnostic imaging of PAC emphasizes safe imaging triage, minimization of risk to mother and fetus while optimizing outcomes, integration of imaging into clinical decision making and malignancy-specific solutions for extent of disease and staging workups,” the authors note.

Read the full article and invited commentary at RSNA.org/RadioGraphics. This article is also available for CME at RSNA. org/Learning-Center. Follow the RadioGraphics editor on Twitter @RadG_Editor.

Fig 6 Jha

Lymphoma in pregnancy in a 32-year-old woman who presented with respiratory distress at 22 weeks gestational age. The imaging workup showed a mediastinal mass that at tissue sampling by US guidance was proven to be B-cell lymphoma. Staging noncontrast MR images (not shown) showed no additional sites of disease. Chemotherapy was started at 24 weeks GA. The patient presented with premature rupture of membranes and had an uncomplicated vaginal delivery at 35 weeks GA. (A) Axial contrast-enhanced CT image of the chest shows a large anterior mediastinal mass (arrows). (B) Gray-scale split-screen US image through the supraclavicular space shows the heterogeneous hypoechoic mass (*). The dotted lines on the right image show the expected path of the needle when using the US biopsy guide device. US can provide percutaneous access to sonographically accessible lesions to minimize exposure to ionizing radiation.

https://pubs.rsna.org/doi/10.1148/rg.220005 ©RSNA 2022
AI journal logo
 

Rethinking Annotation Granularity for Overcoming Shortcuts in Deep Learning-based Radiograph Diagnosis: A Multicenter Study

Despite achieving expert-level accuracies on many disease-screening tasks, deep learning (DL)-based AI models can make correct decisions for the wrong reasons and demonstrate considerably degraded performance when applied to external data. When deep neural networks unintendedly learn dataset biases to fit the training data quickly, it is referred to as “shortcut learning.” Specifically, dataset biases are the patterns that frequently co-occurred with the target disease and are more easily recognized than the true disease signs.

In a new study published in Radiology: Artificial Intelligence, Luyang Luo, PhD, The Chinese University of Hong Kong and colleagues developed a classification model using chest radiograph-level annotations (CheXNet) and a detection model using fine-grained lesion-level annotations (CheXDet) for an extensive comparison on disease classification and lesion detection tasks. The researchers evaluated the ability of fine-grained annotations to overcome shortcut learning in DL-based diagnosis using chest radiographs.

The CheXNet and CheXDet models achieved radiologist-level performance on the internal testing. However, the CheXNet showed dramatically degraded external performance for external testing. The CheXDet showed significant improvement for external performance. CheXDet also achieved higher lesion localization performance than CheXNet for most abnormalities on all datasets.

“Our findings highlight the importance of using fine-grained annotations for developing trustworthy DL-based medical image diagnoses,” the authors conclude.

To read the full article, go to RSNA.org/AI. Follow the Radiology: Artificial Intelligence editor on Twitter @Radiology_AI.