Artificial intelligence (AI) has progressed from technical research to routine clinical use, reaching an inflection point where technological capabilities may exceed current regulatory and oversight frameworks. These systems are becoming more complex, progressing from narrow, task-specific algorithms to foundation models and early agentic prototypes. This progression has redistributed risk, responsibility, and clinical judgement, requiring radiologists and health care leaders to understand how policy choices affect patient safety and clinical innovation advancement. This Special Report provides a roadmap for aligning policy with clinical practice through a practical, lifecycle-based framework centered on patient safety. "Translational bialignment" is a concept which pairs regulatory science requirements (what AI systems deliver to clinicians and patients) with implementation science capabilities (what institutions provide to AI for safe deployment). This framework addresses the full AI lifecycle, from data stewardship and model development to validation, deployment, and monitoring, and articulates shared responsibilities for vendors, institutions, and clinicians grounded in trustworthy AI principles. The analysis focuses on U.S. regulatory frameworks, particularly Food and Drug Administration policies governing medical AI, with relevant highlights from international approaches. Concrete opportunities for radiologists to engage in policy formation, participate in oversight, and collaborate with industry and policymakers are provided to help shape a trustworthy and sustainable AI ecosystem. The alignment of policy, practice, and patient safety will enable medical AI to have a lasting impact on clinical care and public trust. The analysis and recommendations provided represent the authors' perspectives and do not necessarily reflect the official positions of the Radiological Society of North America.

In the French national breast cancer screening program, second reading is performed only for mammograms interpreted as negative (BI-RADS 1-2) at first reading, representing a unique screening workflow in Europe. This retrospective study assessed whether artificial intelligence (AI) could identify a subgroup of negative screening mammograms that could safely bypass second reading. A total of 55,589 screening mammograms from 42,419 women aged 50-74 years (January 2015-December 2019) initially classified as BI-RADS 1-2 were analyzed. Second-reading outcomes were compared with those of a commercial AI system using a predefined binary threshold (≥ 5). Among these examinations, 183 of 55,589 (0.33%) were recalled at second reading, yielding 12 cancers (positive predictive value, 6.6%; cancer detection rate, 0.22 per 1,000 examinations). AI classified 42,606 of 55,589 (76.6%) examinations as low risk (≤ 4) and 12,983 of 55,589 (23.3%) as non-low risk (≥ 5). One cancer was detected in the AI-low group (1 of 55,589 [0.002%]) compared with 11 in the AI-nonlow group (11 of 55,589 [0.020%]; P < .001). Interval cancer rates were higher in the AI-non-low group than in the AI-low group (2.16 vs 0.47 per 1,000 examinations). These findings suggest that excluding AI-low examinations from second reading could reduce workload by approximately 77% while focusing radiologist review on higher-risk cases; prospective validation is needed. AI triage could potentially reduce second-reading workload by approximately 77% in the French breast cancer screening program, despite a small but measurable risk of missed cancers and the need for governance.

Background Emphysema progression is spatially heterogeneous; individuals with chronic obstructive pulmonary disease (COPD) may show similar overall declines in lung density but different regional patterns of disease progression. However, there are currently no tools available to predict lobe-specific density changes to quantify and forecast emphysema progression over time. Purpose To develop and evaluate a deep learning prognostic model using chest CT scans to help predict lobar lung density decline and assess lobar emphysema progression in participants with COPD. Materials and Methods In this prospective study, 5823 participants with and without COPD who underwent serial chest CT were included from the Genetic Epidemiology of Chronic Obstructive Pulmonary Disease (COPDGene) and Evaluation of COPD Longitudinally to Identify Predictive Surrogate End-points (ECLIPSE) studies. A foundational model was trained on CT scans from a subset of 2136 participants from the COPDGene training set to identify local radiographic features indicative of emphysema progression. The lobe-based model was designed to leverage a global attention mechanism to capture lobar tissue destruction and was evaluated in an internal test set of 2718 participants from the COPDGene study and an external test set of 969 participants from the ECLIPSE study. CT scans were harmonized to a 20 HU noise profile, consistent with the COPDGene high-dose protocol. Predicted annualized change in volume-adjusted lung density (ΔALD, g/L per year) values were compared with observed values using root mean squared error and Pearson correlation (r), with agreement assessed using Bland-Altman plots. Results This study included 5823 participants (mean age, 60 years ± 8.6 [SD]; 3003 men). Model predictions of ΔALD showed a positive correlation with observed lobar decline in the COPDGene internal test set (mean Pearson correlation coefficient across lobes, r = 0.433 ± 0.054) and in the ECLIPSE external test set (mean Person r across lobes = 0.471 ± 0.036). The model performed similarly in identification of lobes with accelerated emphysema progression based on predicted ΔALD values in both the COPDGene internal test set (mean lobar area under the receiver operating characteristic curve [AUC], 0.70 ± 0.013) and the ECLIPSE external test set (mean AUC, 0.70 ± 0.009) across ΔALD risk values. Conclusion A CT-based prognostic model of lobe density decline and emphysema progression was developed and evaluated in participants with COPD at chest CT, and model predictions showed positive correlations with observed values. ClinicalTrials.gov registration nos. NCT00608764 and NCT0029255210 © The Author(s) 2026. Published by the Radiological Society of North America under a CC BY 4.0 license. Supplemental material is available for this article. See also the editorial by Marrocchio and Lynch in this issue.

Background The randomized ESCAPE-MeVO trial revealed no benefit of endovascular thrombectomy (EVT) plus usual care in participants with acute stroke due to medium-vessel occlusion (MeVO), a heterogeneous patient population. Identifying subgroups most likely to benefit from EVT with the use of baseline imaging remains critical. Purpose To assess the associations of baseline imaging characteristics with clinical outcomes and intravenous thrombolysis (IVT) and EVT effects in participants with MeVO stroke. Materials and Methods In this post hoc analysis of the ESCAPE-MeVO trial (April 2022 to June 2024), head CT, CT angiography (CTA), and CT perfusion baseline imaging characteristics, including occlusion site and collateral status, were assessed in a central blinded core laboratory review. The primary outcome was the 90-day modified Rankin Scale (mRS) score. Outcomes stratified by imaging characteristics were compared between the EVT and control groups. Effect size estimates for associations of imaging features with outcomes were obtained using logistic regression. Multiplicative interaction terms determined whether imaging characteristics modified the effects of IVT or EVT on outcomes. Results A total of 529 ESCAPE-MeVO participants were included (median age, 75 years [IQR, 64-82 years]; 284 male). Most MeVOs (442 of 522 [84.7%]) were in the middle cerebral artery territory. Collateral status at CTA was good in 129 of 517 participants (25.0%), moderate in 276 (53.4%), and poor in 111 (21.5%). Collateral status was the imaging variable associated with the primary outcome in the multivariable analysis (adjusted common odds ratio, 0.82 [95% CI: 0.67, 0.998]). The probability of EVT achieving an adjusted mRS score of 0 or 1 was lower with EVT (25%) in participants with poor collaterals (vs 37% in the control arm) and higher in those with good collaterals (47% in the EVT arm vs 36% in the control arm), although the difference was not statistically significant (P[interaction] = .30). Conclusion Collateral status at CTA may have a weak association with clinical outcomes among patients with MeVO, warranting further investigation. Clinical trial registration no. NCT05151172 © RSNA, 2026 Supplemental material is available for this article. See also the editorial by Bilgin and Kallmes in this issue.

A 66-year-old man with a history of hypertension presented with left-sided jaw jerking, repetitive movements, aphasia, and dysarthria. He was afebrile and in no acute distress, with mild hypertension (153/81 mm Hg). Routine laboratory studies were unremarkable. Brain MRI showed multiple nonenhancing supratentorial cortical and subcortical hyperintensities on T2-weighted fluid-attenuated inversion recovery (FLAIR) images, which were nonspecific for infectious, inflammatory, vascular, or neoplastic etiologies. CT of the chest, abdomen, and pelvis and MRI of the cervical and thoracic spine were unremarkable. CT angiography of the head and neck showed no areas of stenosis or occlusion. Extensive infectious, autoimmune, and paraneoplastic serologic tests were negative, including serum acetylcholinesterase, antinuclear antibody, erythrocyte sedimentation rate, C-reactive protein, aquaporin 4 and myelin oligodendrocyte glycoprotein (MOG) antibodies, treponemal testing, Tropheryma whipplei, human immunodeficiency virus, tick-borne disease panel, and COVID-19. Lumbar puncture showed 2% neutrophils (reference range, <6%), with normal protein and glucose levels. Cerebrospinal fluid (CSF) studies, including protein electrophoresis, JC virus, herpes simplex virus polymerase chain reaction, MOG antibodies, cytologic examination, and a paraneoplastic panel (Table), were negative. The patient was treated with nicardipine and levetiracetam, with clinical improvement. A presumptive clinical diagnosis of posterior reversible encephalopathy syndrome was made, with plans for short-interval surveillance imaging and possibly brain biopsy, which the patient initially declined. He was subsequently lost to follow-up. Three years later, he experienced several months of progressive mood alterations followed by the development of psychosis and two generalized tonic-clonic seizures. At readmittance, he was again afebrile and in no acute distress, with mild hypertension (150/70 mm Hg). His neurologic examination was remarkable only for symmetric hyperreflexia. Repeat laboratory testing, including human immunodeficiency virus and hepatitis serologic tests, remained negative. Lumbar puncture showed 1% neutrophils, with normal protein and glucose levels and absent oligoclonal bands. MOG antibodies and an autoimmune encephalitis panel (Table) were negative. Repeat brain MRI was performed.

Background Radiologic diagnosis accuracy and efficiency depend on effective metacognitive processes (eg, ongoing confidence monitoring and regulatory decisions). Identifying pitfalls in metacognitive processes is essential for improving diagnostic accuracy and workflow. Purpose To examine how the metacognitive bias termed the hard-easy effect and temporal thinking dynamics manifest in radiologic decision-making and differ with expertise levels. Materials and Methods In this metacognitive analysis, radiologists (at residency year 2 or later) and nonradiology medical professionals reviewed bone radiographs. They determined whether bone lesions were present or absent, which were compared with a reference standard based on CT or MRI findings validated by a panel of two board-certified musculoskeletal radiologists. For each image, participants rated confidence (50% [ie, a guess] to 100%) and recommended the next step (submitting, ordering additional imaging, and consulting a senior colleague). Success, confidence, overconfidence, response time (RT), resolution (confidence-accuracy correlation), efficiency, control sensitivity (confidence-decision correlation), and RT association with confidence were determined. Results Among 28 radiologists and 50 nonradiologists (2806 decisions), the task was challenging, and overconfidence was common (14.73%). Radiologists demonstrated higher success (73.59% [989 of 1344] vs 64.71% [946 of 1462]; P < .001) and confidence (85.12% vs 81.24%; P = .01), less overconfidence (overconfidence of 11.52% vs 16.52%; P = .03), and better resolution (γ = 0.37 vs 0.23; P = .04) than nonradiologists. In both groups, confidence predicted subsequent actions (γ = 0.32 vs 0.53; P = .05). A consistent hard-easy effect emerged: confidence was well-calibrated for easy cases but showed substantial overconfidence for difficult cases (>40%). Fast decisions were accurate and well-calibrated, whereas prolonged deliberation predicted mistakes and increased overconfidence. Lesion-present diagnoses were associated with lower confidence and more deferrals to seniors. Conclusion Metacognitive analysis of radiologic decision-making exposed the hard-easy effect and detrimental temporal dynamics. Although radiologic expertise increased success and confidence in monitoring accuracy, radiologists and nonradiologists were susceptible to overconfidence, particularly in challenging lengthy cases. © RSNA, 2026 Supplemental material is available for this article. See also the article by Krupinski in this issue.

Background The Fontan procedure for single-ventricle congenital heart disease enables survival into adulthood but causes Fontan-associated liver disease (FALD), characterized by progressive hepatic fibrosis and hepatic nodules. Contrast-enhanced US (CEUS) is a proven method to evaluate hepatic nodules, but CEUS findings of FALD nodules have not been systematically described. Purpose To evaluate hepatic CEUS findings and assess interobserver agreement in CEUS interpretations in participants with FALD. Materials and Methods This single-center prospective study included participants with FALD who underwent hepatic CEUS between June 2018 and December 2023. Imaging was evaluated blindly by two CEUS-specialized physicians. Clinical variables were collected via chart review, and biopsy diagnoses were obtained from pathologic examination. Interobserver variability in CEUS findings and diagnoses was quantified using Cohen κ, reported as means and standard errors. Results This study included 32 participants (mean age, 30 years ± 8.7 [SD]; 19 male). Thirty-five hepatic nodules were evaluated, including 20 (57%) focal nodular hyperplasia-like lesions, seven (20%) regenerative nodules, five (14%) hemangiomas, and three (9%) possible malignancies. Interobserver agreement was 100% (κ = 1; no disagreements observed) for CEUS-guided nodule diagnoses. Interobserver agreement was also 100% for CEUS findings, except arterial phase hepatic enhancement pattern (mean κ = 0.92; standard error = 0.07) and lesion echogenicity (mean κ = 0.55; standard error = 0.13). For five biopsied FALD nodules, CEUS diagnosis agreed with pathologic diagnosis, except for one diagnosis of possible malignancy that was biopsy-proven benign adenoma. Washout was seen in 38% (11 of 29) of nodules evaluated with multiphase abdominal CT or MRI. However, at CEUS, only 14% (four of 29) exhibited washout. Conclusion In participants with FALD, there was excellent interobserver agreement in the evaluation of FALD nodules at hepatic CEUS, and CEUS diagnoses mostly agreed with pathologic diagnoses in the subset of cases with biopsies, including for malignant lesions and lesions requiring follow-up. © RSNA, 2026 See also the editorial by Barr in this issue.

History A 16-year-old girl presented to the emergency department with a 1-month history of right lower back pain. She was originally seen at an urgent care clinic, where a lumbar radiograph was reportedly normal, and she was prescribed naproxen. However, she came to the emergency department once pain had escalated to 10 out of 10. Over the past month, the patient's right flank had increased in size and "hardened" to touch. She reported no trauma or other inciting incident. She had a prior history of laparoscopic appendectomy for perforated appendicitis more than 6 years earlier at an outside hospital. A prior contrast-enhanced abdominal CT scan, obtained 2 weeks after the appendectomy, was available for review (Fig 1). At evaluation, the patient was febrile and mildly tachycardic (104 beats per minute). Examination of the right flank revealed a 4-cm area of erythema and tenderness with 10-cm surrounding induration. She had leukocytosis (white blood cell count, 12 × 10[9]/L; reference range, <9.44 × 10[9]/L) and elevated C-reactive protein (23 mg/L; reference range, <9 mg/L). US (Fig 2) followed by contrast-enhanced MRI of the abdomen (Fig 3) was performed. Based on the imaging findings, the patient underwent a percutaneous drainage procedure. All cultures, including aerobic, anaerobic, fungal, and acid fast, were negative, with the caveat that the patient had been started on intravenous antibiotics before the procedure. She was discharged and prescribed 10 days of amoxicillin and clavulanate to complete antibiotic therapy, and the drain was removed 6 weeks later. One month after drain removal, the patient returned to the emergency department with return of severe flank pain, erythema, and purulent drainage. The patient was afebrile with a white blood cell count of 9.8 × 10[9]/L. Contrast-enhanced abdominal CT was performed (Fig 4).