Journal highlights
The following are highlights from the current issues of RSNA’s peer-reviewed journals.
Unmet Needs in Synthetic Medical Imaging
AI models used in medical imaging rely on large, diverse datasets of medical images.
Data sharing in health care is complicated by ethical, technical and legal issues. Synthetic data, or AI-generated data that mimics real-world data, offers an alternative that could protect patient privacy, bridge gaps in data availability and help train radiologists. But current laws do not guarantee the safe and ethical use of synthetic data, and more oversight is needed.
A recent Radiology review provides a roadmap to guide future research and development. Lennart R. Koetzier, BSc, Delft University of Technology in the Netherlands, and colleagues highlight key challenges in the field, including unrealistic images, identity exposure and unethical models.
The authors emphasize that reliable image evaluation methods and close collaboration between regulatory bodies, physicians and AI developers are essential for synthetic images to be useful.
“This process requires constant updates and may lead to a paradigm of synthetic medical imaging that is as safe and reliable as it is innovative,” the authors conclude.
Read the full article, “Generating Synthetic Data for Medical Imaging,” at RSNA.org/Radiology.
Follow the Radiology editor on X @RadiologyEditor.
Top row shows synthetic chest radiographs that were misidentified as real by a radiologist. Bottom row shows synthetic chest radiographs that were correctly identified as synthetic. Images were generated by a diffusion model trained on samples of normal chest radiographs from the National Institutes of Health Chest X-ray Dataset.
Complete legend details can be found at https://doi.org/10.1148/radiol.232471 ©RSNA 2024.
The Emerging MASLD Crisis
Fatty liver disease, now called metabolic dysfunction-associated steatotic liver disease (MASLD), is on the rise alongside global obesity rates. Imaging plays a critical role in the diagnosis, monitoring and management of this disease. Extensive research has improved understanding of its pathology and prognosis, but fundamental knowledge gaps remain.
In a recent article in RadioGraphics, co-lead authors Sedighe Hosseini Shabanan, MD, MPH, and Vitor F. Martins, MD, PhD, University of California, San Diego, and colleagues summarize current insights into MASLD. The authors also highlight six key opportunities for the radiology community to address together, including problems with proliferation, reproducibility and reporting. The article concludes with a call to action.
“We call on the entire house of radiology—radiologists, technologists, administrators, researchers, industry and radiology societies—to take action, demand a seat at the table with other societies, and advance the development, validation, dissemination and accessibility of the technologies needed to confront the looming crisis,” the authors write.
Read the full article, “MASLD: What We Have Learned and Where We Need to Go—A Call to Action,” and the invited commentary at RSNA.org/RadioGraphics. This article is also available for CME on EdCentral. Follow the RadioGraphics editor on X @RadG_Editor.
MRI for assessment of hepatic steatosis and fibrosis in a 60-year-old woman with biopsy-proven at-risk metabolic dysfunction-associated steatohepatitis. Axial MRI proton-density fat fraction map (left) shows a fat fraction of 12% and axial 2D MRE stiffness map (right) shows a stiffness of 3.4 kPa. Cardiometabolic factors for MASLD diagnosis included elevated body mass index (26.9 kg/m2 ) and elevated fasting blood glucose levels. The patient’s alcohol use was less than one drink per week. Serum aspartate aminotransferase and alanine aminotransferase levels were elevated, and the fibrosis-4 score was 2.0. Note that as fibrosis increases, steatosis can decrease.
https://doi.org/10.1148/rg.240048 ©RSNA 2024
A Prostate Cancer Breakthrough With PSMA Therapy
Except for skin cancer, prostate cancer is the most common cancer in men in the U.S. For those with metastatic castration-resistant prostate cancer (mCRPC), an advanced form of the disease, survival rates are low. A promising treatment, known as PSMA radioligand therapy, uses nuclear medicine to target prostate-specific antigen (PSMA) on cancer cells.
The U.S. Food and Drug Administration (FDA) approved the radioactive drug lutetium 177 (177Lu) PSMA-617 in 2022 based on the landmark VISION trial. 177Lu PSMA-617 offers a personalized treatment approach that may improve survival and quality of life for patients with mCRPC.
A RadioGraphics article highlights the growing importance of PSMA therapy amid the evolving landscape of prostate cancer treatment. Laszlo K. Szidonya, MD, PhD, Oregon Health & Science University in Portland, and colleagues examine therapy administration, dosing, adverse effects and benefits of imaging after treatment.
“The integration of a multidisciplinary approach to patient management, alongside the utilization of imaging biomarkers and posttherapy SPECT/CT, can further refine the personalization of treatment of individual patients,” the authors conclude.
Read the full article, “PSMA Radiotheranostics in Prostate Cancer: Principles, Practice, and Future Prospects,” at RSNA.org/RadioGraphics. This article is also available for CME on EdCentral. Follow the RadioGraphics editor on X @RadG_Editor.
Nephrostomy tubes and extensive bone marrow involvement in a 74-year-old man with mCRPC referred for 177Lu-PSMA-RLT. 18F-DCFPyL-PET/CT maximum intensity projection (A) and axial fused (B) images show extensive PSMA-positive osseous matastases and retroperitoneal lymphadenopathy (green arrow). Bilateral nephrostomy tubes are present (blue arrows) to relieve bilateral hydronephrosis caused by the retroperitoneal lymphadenopathy.
Complete legend details can be found upon publication at https://doi.org/10.1148/rg.240080 ©️RSNA 2024.
Deep-Learning Denoising Technique for Brain MRI
MRI image acquisition speed has improved over time through techniques that include faster imaging sequences, parallel imaging, compressed sensing and deep learning-based reconstruction of incomplete images.
Deep learning methods work best when they incorporate principles of image physics. However, these methods usually need access to proprietary raw data and detailed information about an MRI system’s hard[1]ware and software.
In an article published in Radiology Advances, Laura Onac, MSc, Ezra AI Inc., and colleagues from New York University Grossman School of Medicine assessed a vendor-agnostic AI-based approach to removing image degradation artifacts in highly accelerated MRI scans.
The new model’s outputs rated better than the original images, with improvements in quality and feature visibility. Resolution was either improved or maintained, and scan time was reduced on average by 29%.
“This vendor-agnostic AI-based method achieved robust scan time savings without loss of image quality, potentially allowing for reduced cost and improved patient experience,” the authors conclude.
Read the full article, “An Image-Domain Deep-Learning Denoising Technique for Accelerated Parallel Brain MRI: Prospective Clinical Evaluation,” at Academic: Oup.com/RADADV.