Journal of the Korean Society of Radiology最新文献

The use of AI in radiology is rapidly transitioning towards the use of foundation models that learn universal representations from large-scale imaging and multimodal clinical data. In this review, we outline the key technical components of these models, including self-supervised encoders, fusion modules for feature alignment, and task-specific decoders, and further summarize the recent work in three categories: 1) image-only models, trained on millions of unlabeled radiology scans to enable robust transfer learning with minimal annotation, 2) Chest X-ray image-report models, which leverage large chest X-ray-reported corpora for joint visual and textual embedding, and 3) image-report models for other modalities, which fuse volumetric images with structured reports or clinical metadata. We further discuss the relevant evaluation strategies, including vision-centric, language-centric, and benchmark-based metrics, and outline approaches for clinical validation. Finally, we highlight the persistent challenges in the application of these models, and propose future directions for multimodal integration and human-AI collaboration to advance personalized radiology.

In the field of radiology, clinical practice guidelines (CPGs) have been established as a core tool for ensuring consistency in clinical diagnosis and patient safety. In 2012, aligning with the global emphasis on evidence-based medicine that emerged in the early 2000s, the CPG Committee of the Korean Society of Radiology was founded, with the aim of establishing or revising various guidelines across radiology and related medical fields. Since then, the Committee has developed and disseminated diverse CPGs to ensure the appropriateness of radiological examinations and to minimize radiation exposure. This report reviews the Committee's major achievements over the past decade, including the development of justification guidelines, support for subspecialty-led creation of guidelines, safety protocols for contrast media, rapid guidelines for COVID-19 imaging, and integration with clinical decision support systems. Through active collaboration with government agencies and academic institutions, the Committee has enhanced the scientific rigor and clinical relevance of its guidelines. Furthermore, the launch of an online archive has improved accessibility and utilization. Looking forward, the Committee aims to establish AI-integrated guideline frameworks and expand globally through international cooperation and alignment with national health policies.

Biliary adenofibroma, an extremely rare benign liver tumor with potential malignancy, lacks well-established imaging features because of its scarcity. Here, we report the imaging findings in a case of biliary adenofibroma, focusing on its characteristic solid microcystic features. During the screening of a healthy 44-year-old male, a mostly solid echogenic mass resembling a cavernous hemangioma was found in the right lobe of the liver on ultrasound examination. Subsequent CT revealed a low-density mass with a central nodular area, and MRI scans revealed a mostly cystic mass comprising numerous microcysts with a honeycomb appearance and a small scar-like nodular soft tissue area. Following hepatic segmentectomy, histopathological examination confirmed the presence of a biliary adenofibroma. This review describes a pathologically confirmed case of biliary adenofibroma. This rare hepatic tumor resembles cavernous hemangiomas of the liver on ultrasonography and pancreatic serous cystadenomas on MRI.

Since the introduction of Prostate Imaging Reporting and Data System (PI-RADS), prostate MRI has become an essential tool for detecting clinically significant prostate cancer, determining tumor stage, and guiding targeted biopsy. However, variability in image quality across institutions continues to affect diagnostic accuracy. To address this issue, the Prostate Imaging Quality (PI-QUAL) scoring system was developed as a standardized framework for evaluating the quality of prostate MRI. PI-QUAL v1 established the first quality control system for multiparametric MRI (mpMRI) but had limitations, including the inability to assess biparametric MRI (bpMRI) and a strong reliance on biopsy correlation. The updated PI-QUAL v2 simplified technical requirements, incorporated bpMRI assessment, and introduced a reproducible 3-point scale with adjustment rules that balance quality assurance with the diagnostic advantages of mpMRI. Factors influencing image quality include magnetic field strength and susceptibility- and motion-related artifacts. Complementary approaches, such as the Prostate Signal Intensity Homogeneity Score and artificial intelligence-based evaluation tools, have also been proposed. Overall, PI-QUAL v2 enhances reproducibility and broadens clinical applicability, providing a practical framework for ensuring high-quality prostate MRI. Future progress is expected through standardized data collection, structured training, and multidisciplinary collaboration.

Müllerian cysts in the posterior mediastinum are extremely rare benign cysts of Müllerian origin. On radiographs, Müllerian cysts appear as well-defined round masses with homogeneous water density. As these radiologic features resemble those of other posterior mediastinal cysts, the preoperative diagnosis of Müllerian cysts can be challenging. Herein, we report the imaging and pathological findings of a 54-year-old female patient with a posterior mediastinal Müllerian cyst.

Bronchovascular fistulas are rare yet lethal conditions capable of causing massive hemoptysis. To date, only one case of a bronchial artery-bronchial fistula has been reported. We report a patient with a bronchial artery-bronchial fistula treated using a multidisciplinary approach, including surgery. An 80-year-old male presented to the emergency department with hemoptysis that had started the previous day. His past history included bronchial artery embolization for massive hemoptysis 13 years prior, and he had been diagnosed with pulmonary thromboembolism and protein C deficiency 6 years ago. The patient underwent two bronchial artery embolizations for the massive hemoptysis, and during the second procedure, a bronchial artery-bronchial fistula was identified. As embolization was deemed ineffective, an emergency bilobectomy was performed. The patient developed Candida empyema as a postoperative complication but was discharged a month after surgery following antifungal treatment.

The Korean Society of Radiology (KSR) has overcome numerous crises and achieved significant growth over its 80-year history, thanks to the unity of its members. Starting as a small, academic-focused group, the society evolved to recognize the importance of its role in promoting public health and protecting members' rights, which was reflected in its bylaws and objectives. KSR appropriately restructured its governance, introducing an executive board system and committees to enhance its professional operations. It transitioned from a dual president-chairperson system to a single president system and established a council. Since 2012, successive presidents have consistently guided the society with slogans and core values that align with the demands of the times. Facing ongoing challenges, KSR is committed to strategic efforts such as strengthening policy engagement, standardizing resident training, expanding social responsibility, and enhancing member communication. Through these initiatives, the society aims to use the next 20 years for a new leap forward, positioning itself as a global leader in radiology by 2045 and shaping the future of the field.

This review provides an overview of the latest trends in lesion detection using AI in musculoskeletal imaging. It describes the types of deep learning networks used in detection AI and briefly explains their principles. Fracture-detection AI has shown improved sensitivity and reduced reporting time in multiple meta-analyses, and real-world validation in clinical settings has begun. Although many AIs have been developed to detect joint injuries and degenerative changes in MRI and CT/MRI detection models for bone metastasis and multiple myeloma, they have not yet reached a robust validation stage. Achieving clinical value requires attention to explainability, external validation and post-market monitoring, Picture Archiving Communicating System (PACS)-level integration, and legal and ethical issues and, therefore, proactive adoption by radiology professionals.

AI-based software as a medical device (SaMD) using deep learning applications for musculoskeletal diseases is being clinically implemented in South Korea, although it is still in its early stages in the musculoskeletal field compared to other fields of radiology, such as neuroradiology, chest, and breast imaging. AI models for detecting various fractures, estimating pediatric bone age, calculating geometric skeleton measurements, grading arthritis, and osteoporosis screening have shown high diagnostic performance, and many of these applications are now commercially available for use in clinical practice. Many studies have documented the feasibility of using an AI model for detecting joint pathology on MRI and interpreting spine MRIs. This review provides information on the domestic and international commercialization status of AI-based SaMD for musculoskeletal imaging and beneficial considerations for its application in clinical practice, helping readers who are interested in the field application of musculoskeletal imaging AI models in their decision-making.