Arterial embolization is a rapidly growing therapy for joint pain, particularly in cases where conservative measures like nonsteroidal anti-inflammatory drugs and joint injections have failed. This article presents the technical aspects of embolization for knee osteoarthritis, frozen shoulder, and tennis elbow. The guide outlines patient selection criteria, procedural techniques, and postprocedural care. Several trials have produced results with high rates of clinical success and functional improvement. As embolization techniques evolve, procedural standardization will enhance treatment efficacy and minimize adverse events, potentially integrating embolization into broader treatment guidelines for joint pain.

Background Patients diagnosed with ductal carcinoma in situ (DCIS) may also have undetected invasive breast cancer. Radiomic features of calcifications at mammography can predict occult invasive disease among women diagnosed with DCIS at core-needle biopsy, which could affect treatment recommendations. However, the generalizability of these radiomic models must be tested before they are adopted in clinical practice. Purpose To evaluate the generalizability of radiomic models based on mammography features to predict occult invasive cancer among women diagnosed with DCIS at core-needle biopsy from three national datasets. Materials and Methods In this retrospective, cross-national study, digital mammograms from women diagnosed with DCIS at breast core-needle biopsy were collected in the United States, United Kingdom, and the Netherlands between January 1, 2000, and December 31, 2021. Only asymptomatic women who had calcifications but did not have associated masses, architectural distortions, or asymmetries were included. Radiomic models were developed using cross-validated logistic regression on each national dataset, then round-robin tested on the other datasets. Differences across the three datasets in terms of the upstaging rate, age, lesion size, and estrogen and progesterone receptor levels were assessed using Kruskal-Wallis or χ test. Results The study included 1498 women (age range, 31-89 years; mean age, 59 years ± 9 [SD]), as follows: 696 women from the United States, 618 women from the United Kingdom, and 184 women from the Netherlands, with upstaging rates of 16.1%, 16.7%, and 14.1%, respectively. Internal cross-validation areas under the receiver operating characteristic curve (AUCs) were 0.675 (95% CI: 0.671, 0.679), 0.603 (95% CI: 0.567, 0.722), and 0.701 (95% CI: 0.697, 0.706) for the U.S., UK, and Netherlands datasets, respectively. The model that was trained on the U.S. dataset yielded cross-national validation AUCs of 0.604 (95% CI: 0.560, 0.648) and 0.682 (95% CI: 0.607, 0.757) for the UK and Netherlands datasets. Conclusion Radiomic machine learning models were shown to have the potential to predict occult invasive cancer in women with DCIS across diverse settings. © RSNA, 2025

Background Evidence is limited on how radiologist interpretation, patient age, breast density, and family history influence interval breast cancer (IBC) risk. Purpose To investigate risk factors for IBC in women enrolled in a biennial population-based mammography screening program in Taiwan. Materials and Methods This retrospective study included women who underwent mammography in a population-based breast cancer screening program in Taiwan between January 1, 2004, and December 31, 2018. Standard two-view digital mammograms were interpreted by board-certified radiologists using Breast Imaging Reporting and Data System criteria. Performance measures included recall rate (RR), cancer detection rate (CDR), and positive predictive value (PPV); radiologist audit scores were derived from RR and PPV 1. IBC was defined as cancer diagnosed after a negative screening result and before the next 2-year screening. Multivariable Poison and logistic regression models with random intercepts for radiologist-facility pairs were used to estimate adjusted relative risks and odds ratios with 95% CIs. Results A total of 2 881 405 women (mean age, 56.1 years ± 6.6 [SD]; range, 45-69 years) underwent 6 592 768 mammographic examinations; 10 944 women were diagnosed with IBC. IBC risk per 1000 person-years was higher in women with a family history of breast cancer than in those without (1.41 [95% CI: 1.33, 1.51] vs 0.77 [95% CI: 0.75, 0.80]; < .001) and in women with extremely dense breasts compared with those in lower breast density categories (1.15 vs 0.88, 0.54, and 0.29; < .001). Women whose mammograms were interpreted by a radiologist with an audit score of 0-4 had fewer IBCs than those whose mammograms were interpreted by a radiologist with an audit score of 5 or greater (0.78 vs 1.23 per 1000 person-years; < .001). Women whose mammograms were interpreted by a radiologist with low RR and CDR had the highest proportion of IBCs among all cancers detected (53.87% [95% CI: 49.71, 57.98]; < .001). Among missed IBCs, asymmetry was the most common false-negative finding (22%). Conclusion High breast density, family history of breast cancer, and suboptimal radiologist performance were independently associated with increased IBC risk. These findings highlight the importance of performance auditing and targeted strategies to enhance early cancer detection in population screening. © RSNA, 2025

Background The potential of photon-counting detector (PCD) CT for rectal cancer (RC) staging remains, to the knowledge of the authors, unexplored. Purpose To determine the optimal virtual monoenergetic images (VMIs) and quantum iterative reconstruction (QIR) strength in PCD CT for RC T staging and to compare the diagnostic performance with that of MRI. Materials and Methods In this prospective study, participants with RC underwent preoperative contrast-enhanced abdominal PCD CT and rectal MRI between September and November 2024. Portal venous phase PCD CT images including VMIs at 40-70 keV in 10-keV increments and polychromatic images (referred to as T3D by the manufacturer) were reconstructed with five QIR levels (off and levels 1-4), and image quality was evaluated. Two junior radiologists and a senior radiologist independently performed T staging evaluations from T3D, optimal PCD CT images, and MRI scans. Diagnostic performance was compared across modalities using DeLong and McNemar tests. Results Seventy-two participants (mean age, 61 years ± 11.8 [SD]; 40 men) were included for image analysis; 51 underwent radical surgery for T staging analysis. VMIs at 40 keV with QIR level 4 (QIR-4) showed the best signal-to-noise ratio (12.1 ± 2.8), contrast-to-noise ratio (6.2 ± 2.6), and maximal Likert scores (median score, 5; IQR, 5-5) for tumor conspicuity, tumor margin clarity, and overall image quality versus other reconstructions (all < .001). Compared with T3D, optimal PCD CT improved T3-T4 diagnostic performance for the two junior radiologists (area under the receiver operating characteristic curve [AUC], 0.80 vs 0.64 [ = .03] and 0.82 vs 0.64 [ = .002], respectively) but not for the senior radiologist (AUC, 0.86 vs 0.80; = .13). Optimal PCD CT demonstrated comparable performance to MRI in overall T staging (accuracy, 65%-76% vs 67%-80%) and T3-T4 staging (AUC, 0.80-0.86 vs 0.78-0.88) across all readers (all > .05). Conclusion In participants with RC, portal venous abdominal PCD CT at 40 keV and with QIR-4 yielded the optimal image quality for RC T staging, showed enhanced T3-T4 diagnostic performance among less-experienced radiologists, and achieved performance that was comparable to that of MRI. © RSNA, 2025

Background Radiology staging reports (ie, oncologic reports) are written for referring physicians using complex medical terminology. Large language models (LLMs) show promise for simplifying medical text for patient use, but controlled studies evaluating the impact of LLM simplification on patients' comprehension of radiology reports are lacking. Purpose To evaluate whether LLM-based simplification of oncologic CT reports improves patients' cognitive workload, text comprehension, perception, and reading time. Materials and Methods This prospective, controlled, open-label, quasi-randomized trial enrolled 200 adults with cancer who underwent routine CT restaging. Between April and May 2025, participants were alternately assigned to receive either standard CT reports (100 participants) or LLM-simplified versions created using Llama 3.3 70B (Meta) with mandatory radiologist review (100 participants). The primary outcomes were participant-reported scores on nine seven-point Likert scale items, and composite scores, in the domains of cognitive workload, text comprehension, and report perception, as well as reading time. Secondary outcomes included readability metrics and independent radiologist assessments of report errors, usefulness, and quality. Statistical analyses included logistic regression adjusted for participant characteristics. Results Among the 200 participants (mean age, 64 years ± 14 [SD]; 112 male participants), simplified reports reduced the median reading time from 7 minutes to 2 minutes ( < .001). Participants who received simplified reports reported lower cognitive workload (adjusted odds ratio [OR], 0.18 [95% CI: 0.13, 0.25]), better comprehension (adjusted OR, 13.28 [95% CI: 9.31, 18.93]), and better perception of report usefulness (adjusted OR, 5.46 [95% CI: 3.55, 8.38]) than did those who received standard reports (all < .001). Simplification improved report readability (mean Flesch-Kincaid Grade Level, 8.89 ± 0.93 vs 13.69 ± 1.13; < .001). Radiologist review revealed factual errors in 6% (moderate, 2%; severe, 4%), content omissions in 7% (minor, 2%; moderate, 1%; severe, 4%), and inappropriate additions in 3% (minor, 1%; moderate, 2%) of simplified reports. Conclusion LLM simplification of oncologic CT reports improved patient comprehension and reduced reading burden. However, clinically relevant errors were identified. © RSNA, 2025

Background Despite transarterial chemoembolization (TACE) serving as the first-line treatment for patients with intermediate stage B hepatocellular carcinoma (HCC), complete response rates are generally below 27%. Purpose To compare the efficacy and safety of an anhydrous cisplatin suspension in ethiodized oil-based TACE versus a conventional aqueous cisplatin emulsion in the treatment of participants with HCC. Materials and Methods In this prospective, multicenter, randomized controlled trial conducted from September 2016 to February 2023, participants from three hospitals in Hong Kong were randomized to an experimental or control group in a 1:1 ratio. The experimental group received an anhydrous cisplatin suspension (4 mL of ethiodized oil with 20 mg of cisplatin powder). The control group received a conventional aqueous cisplatin emulsion. TACE was performed within 4 weeks after randomization in two to three treatments, 2 months apart. Response was assessed with CT and digital subtraction angiography during subsequent TACE. Primary end points were complete tumor response and severe adverse events. Secondary end points included progression-free survival and overall survival (OS). Survival outcomes were compared using the log-rank test and hazard ratios with 95% CIs. Results A total of 77 participants were included (median age, 68 years; IQR, 64-75 years; 59 men). At 6 months, the complete tumor response rate was higher in the suspension group (90% [35 of 39 participants]) compared with the emulsion group (47% [18 of 38 participants]; < .001). Serious adverse events were similar in the suspension group (2.3% [three of 128 procedures]) and emulsion group (5.2% [eight of 153 procedures]; = .21). Median progression-free survival was higher in the suspension group (21.1 months; 95% CI:14.3, 38.9) compared with the emulsion group (10.4 months; 95% CI: 7.3, 13.4) (hazard ratio, 0.35; < .001). Median OS was higher in the suspension group (53.3 months; 95% CI: 40.5, not reached) than in the emulsion group (36.0 months; 95% CI: 25.7, 46.6) (hazard ratio, 0.32; = .004). Conclusion Ethiodized oil-based TACE using an anhydrous cisplatin suspension resulted in better complete tumor response, progression-free survival, and OS rates compared with the conventional aqueous cisplatin emulsion. ClinicalTrials.gov identifier NCT03268499 © RSNA, 2025

The increasing integration of artificial intelligence (AI) tools into radiology has created an urgent need for clear, practical guidance on how to evaluate them. These tools, including computer-assisted detection and triage devices, hold promise for improving accuracy, efficiency, and workflow. However, their adoption into clinical practice requires rigorous evaluation to ensure safety, generalizability, and clinical value. Radiology has a strong foundation in diagnostic test assessment, and AI models represent an extension of this tradition, with new considerations in evaluation strategy, performance measurement, and study design. This article provides a structured primer on evaluating AI models across their development and deployment life cycle. It outlines key principles for internal and external testing, highlights performance metrics tailored to AI outputs-including classification, detection, segmentation, and continuous measures-and describes how to assess clinical impact with multireader multicase studies. Practical examples from radiology research, as well as updated reporting standards, are incorporated throughout. By translating core statistical concepts into radiology-specific guidance, this article aims to support radiologists, researchers, and reviewers in conducting and interpreting high-quality AI evaluation studies.

A 64-year-old man was brought to the emergency department after being found unresponsive at home, with family reporting progressive confusion following heavy alcohol intake the previous day. On arrival, he was nonverbal and unresponsive to stimuli. Vital signs showed a normal heart rate and blood pressure, with oxygen saturation measured at 80%. Arterial blood gas analysis revealed a pH of 6.86 (reference range, 7.35-7.45), bicarbonate level of 5.4 mmol/L (reference range, 20-28 mmol/L), and anion gap of 28 mmol/L (reference range, 4-12 mmol/L). Lactate level was elevated, at 7.8 mmol/L (reference range, 1-2 mmol/L). Creatinine level was 1.45 mg/dL (128 µmol/L) (reference range, 0.7-1.3 mg/dL [62-115 µmol/L]). Blood glucose level was 282 mg/dL (15.7 mmol/L) (reference range, 70-99 mg/dL [3.9-5.5 mmol/L]), and white blood cell count was 18.8 × 10/L (reference range, 4.5 × 10/L to 11 × 10/L). Liver function test results were within normal limits. Results of initial noncontrast head CT and CT angiography of the head performed at an outside institution were interpreted as normal. Brain MRI was performed without intravenous contrast material approximately 8 hours after initial presentation. Brain MRI included axial diffusion-weighted imaging (Figs 1-3), T2-weighted fluid-attenuated inversion recovery (FLAIR) imaging (Fig 4), and susceptibility-weighted imaging (Fig 5).

A 17-year-old boy presented to the outpatient department with left-sided breast enlargement for 2 years. He also had multiple long bone fractures after trivial trauma over the past 8 years, along with intermittent episodes of streaky hemoptysis. He was treated with bisphosphonates, with marginal relief of symptoms. At the index presentation, as part of preanesthetic checkup for elective surgery for gynecomastia, laboratory investigations revealed thrombocytopenia, with a platelet count of 100 000/µL (reference range, 150 000-400 000/µL), and a normal hormonal profile. Chest radiography was performed, followed by diagnostic thoracentesis, which yielded minimal hemorrhagic fluid. Subsequently, contrast-enhanced CT and contrast-enhanced MRI were performed. Echocardiography showed diffuse pericardial thickening.

Background Randomized trials have shown that endovascular thrombectomy (EVT) benefits patients with large vessel occlusion; however, its role in medium vessel occlusion remains unclear. Purpose To compare the efficacy and safety of EVT versus standard medical management in medium vessel occlusion stroke. Materials and Methods This multicenter, retrospective study enrolled patients with acute ischemic stroke from an occlusion of M3 or M4 segments of the middle cerebral artery, anterior cerebral artery, or posterior cerebral artery from 25 stroke centers (September 2019 to September 2024). The primary end point was an ordinal shift in 90-day modified Rankin Scale (mRS) scores. Safety outcomes included 90-day mortality and symptomatic intracranial hemorrhage at 24 hours. Results A total of 1075 patients were included in the study (median age, 69 years; IQR, 59-76 years; 637 men; 529 treated with EVT and 546 treated with standard medical management; median baseline National Institutes of Health Stroke Scale [NIHSS], 10 [IQR, 6-12]). In the primary analysis using inverse probability of treatment weighting, there was an improvement in 90-day mRS distribution (adjusted common odds ratio, 1.38; 95% CI: 1.18, 1.61; < .001) favoring EVT. Rates of mRS scores of 0-1 (43.7% [231 of 529 patients] vs 36.1% [197 of 546 patients]; adjusted risk ratio [RR], 1.61; 95% CI: 1.33, 1.96; < .001) and mRS scores of 0-2 (60.9% [322 of 529 patients] vs 53.5% [292 of 546 patients]; adjusted RR, 1.39; 95% CI: 1.14, 1.69; = .001) were higher with EVT. Between EVT and standard medical management groups, there was no difference in 90-day mortality (8.32% [44 of 529 patients] vs 8.97% [49 of 546 patients]; adjusted RR, 0.74; 95% CI: 0.51, 1.12; = .17) or symptomatic intracranial hemorrhage (11.5% [61 of 529 patients] vs 10.4% [57 of 546 patients]; adjusted RR, 1.24; 95% CI: 0.92, 1.66; = .16). Subgroup analyses indicated EVT benefit in patients with an NIHSS score of 6 or higher (adjusted RR, 1.62; 95% CI: 1.37, 1.92), but not in those with an NIHSS score of 1-5 (adjusted RR, 0.79; 95% CI: 0.58, 1.19; value for interaction < .001). Conclusion Compared with standard medical management, EVT was associated with better outcomes in patients with acute medium vessel occlusion stroke, particularly those with more severe symptoms, without increasing symptomatic intracranial hemorrhage or 90-day mortality. Chinese Clinical Trial Registry no. ChiCTR2500096954 © The Author(s) 2025. Published by the Radiological Society of North America under a CC BY 4.0 license.

Background The T2-weighted fluid-attenuated inversion recovery (FLAIR) sequence is part of the routine brain MRI protocol. In regions with strong inhomogeneities in the static magnetic (B) field and/or the radiofrequency (RF) field, inadequate magnetization inversion results in artifacts that may mimic or obscure pathologic features. Purpose To reduce artifacts on FLAIR images by using an optimized inversion pulse that is robust to inhomogeneities in both the B and RF fields. Materials and Methods In this prospective study, a FLAIR inversion pulse was designed using optimal control. FLAIR and FLAIR with controlled inversion (C-FLAIR) images were acquired at 3 T in a phantom designed to exhibit strong inhomogeneities in B and in 14 participants (mean age, 36.1 years ± 11.5 [SD]; nine male participants) enrolled between October 2024 and August 2025 at a single academic medical center: nine healthy participants, two with relapsing-remitting multiple sclerosis, one with persistent concussion symptoms, and two with asymptomatic white matter hyperintensities. In the phantom, water signal suppression was assessed visually. In human participants, cerebrospinal fluid signal suppression, presence of artifacts, and visibility of multiple sclerosis lesions and white matter hyperintensities were assessed visually by a radiologist. In eight healthy volunteers, mean signal-to-noise ratio (SNR) and mean contrast-to-noise ratio (CNR) were computed for FLAIR and C-FLAIR, with differences between the sequences evaluated using the Student test. Results C-FLAIR exhibited nearly perfect inversion in the presence of inhomogeneities in the B field, resulting in the removal of artifactual signal. The image contrast of demyelinating multiple sclerosis lesions and white matter hyperintensities was identical for the optimized pulse and the conventional pulse. There was no evidence of a difference in mean SNR (26.5 ± 3.5 for FLAIR vs 26.4 ± 4.7 for C-FLAIR) or mean CNR (18.5 ± 2.1 for FLAIR vs 17.9 ± 2.4 for C-FLAIR) between sequences. C-FLAIR had 13.8% higher specific absorption rate (0.033 vs 0.029 W/kg) and 6.8% higher time-averaged RF (0.78 vs 0.73 µT). Conclusion C-FLAIR with robust RF inversion showed practical elimination of artifacts caused by incomplete inversion. © RSNA, 2025 See also the editorial by Casselman and Bowen in this issue.

Background Current CT ancillary features (AFs) have limitations, as several AFs are more evident at MRI. Precontrast low attenuation may serve as a potential AF in the Liver Imaging Reporting and Data System (LI-RADS). Purpose To evaluate the diagnostic value of precontrast low attenuation at CT as an additional AF for diagnosing hepatocellular carcinoma (HCC) and to assess its impact on LI-RADS diagnostic performance. Materials and Methods This retrospective study included adults at risk of HCC who underwent multiphase dynamic liver CT before hepatic resection or liver transplant at a tertiary referral facility between January and December 2022. Two radiologists assessed the presence of major features and AFs for each hepatic observation on the basis of LI-RADs categories, as follows: LR-3, intermediate probability of malignancy; LR-4, probably HCC; and LR-5, definitely HCC. Each lesion was assigned a LI-RADS category twice: first, by using AFs only (LI-RADS category with AFs only) and next, by using AFs with precontrast low attenuation (LI-RADS category with AFs and precontrast low attenuation). Precontrast low attenuation was defined as an attenuation of the target hepatic observation lower than that of the liver parenchyma based on visual assessment. Histopathologic analysis and clinical assessment were used as reference standards. The diagnostic performance of the two LI-RADS strategies was compared using generalized estimating equations. Results A total of 194 patients (mean age, 59 years ± 10 [SD]; 159 men) with 328 hepatic observations were included: 187 (57.0%) HCCs, 26 (7.9%) non-HCC malignancies, and 115 (35.1%) benign lesions. Precontrast low attenuation was associated with HCCs, yielding a diagnostic odds ratio of 9.1 (95% CI: 4.9, 16.6; < .001). Adding precontrast low attenuation upgraded 20 observations (17 HCCs and three dysplastic nodules) from LR-3 to LR-4, increasing the proportion of HCCs in LR-4 from 64.7% (22 of 34) to 72.2% (39 of 54). Compared with LR with AFs only, LR with AFs and precontrast low attenuation had a higher sensitivity (88.6% [163 of 187] vs 79.9% [146 of 187]; < .001), with no evidence of a difference in specificity (82.7% [121 of 141] vs 85.0% [124 of 141]; = .06). Conclusion Applying precontrast low attenuation at CT as an additional AF increased the proportion of HCCs in LR-4 and improved the sensitivity of LI-RADS for diagnosing HCC. © RSNA, 2025

Foundation models (FMs) represent a transformative advancement in artificial intelligence (AI), with growing applications in medical imaging. These models leverage self-attention mechanisms and are capable of processing multimodal data, such as images, text, audio, and video, across multiple scales. Although FMs require large datasets for initial training, they can be adapted to specific medical imaging tasks using smaller labeled datasets through techniques such as transfer learning, fine-tuning, prompt engineering, few-shot learning, and zero-shot learning, making them especially valuable in data-scarce settings. Many FMs also incorporate generative AI capabilities that support the creation of synthetic medical images to further address annotation limitations. Current applications span various imaging modalities in radiology, where FMs have shown potential to improve diagnostic accuracy and streamline workflows. However, clinical integration remains challenging due to issues such as limited interpretability, potential bias, privacy concerns, regulatory constraints, high computational costs, and domain shifts between training data and real-world clinical environments. Addressing these barriers will require coordinated efforts among technical developers, health care providers, and regulatory bodies. This review explores the evolving role of FMs and generative AI in radiology, highlighting recent research advances, clinical applications, and the key challenges that must be addressed for responsible deployment.

Thyroid and parathyroid ablation have gained substantial attention as promising minimally invasive alternatives to surgery. Despite growing evidence supporting their use, standardized ablation protocols remain scarce. This practical guide focuses on the essential aspects of thyroid and parathyroid ablation, including practitioner qualifications, standardized procedures, critical techniques, complication prevention, and evaluation. These areas are emphasized due to the anatomic proximity of the thyroid and parathyroid glands, which results in procedural similarities. Practitioners need proper qualifications before performing thyroid and parathyroid nodule ablation. Doctors interested in conducting these procedures should possess foundational knowledge and undergo prior training. Patient selection should involve multidisciplinary consultations to carefully weigh the advantages and disadvantages of different treatment options. Preprocedural preparations, including anesthetic and technical considerations, are discussed to enhance treatment efficacy and optimize patient safety. The hydrodissection technique is recommended to minimize complications and prevent tissue adhesion. Detailed descriptions of ablation techniques are provided to facilitate understanding and application. Contrast-enhanced US assessment is performed immediately after ablation to evaluate its effectiveness. Management of complications and follow-up practices are also addressed. By outlining the clinical implementation of this technology, this practical guide aims to standardize thyroid and parathyroid ablation, ensuring efficacy and reducing risks.