Spectral CT and photon-counting detector (PCD) CT are recent innovations in imaging that have gained increasing use as diagnostic tools. Both modalities leverage advanced detector technologies-PCD CT uses detectors that count individual x-ray photons and measure their energies, while spectral CT separates x-ray energies at dual or multiple levels. These approaches provide higher spatial resolution, improved contrast-to-noise ratio, and reduced noise and enable multienergy imaging from a single acquisition. These techniques enhance visualization of tissue characteristics and reduce artifacts compared with conventional energy-integrating detector CT. This review article introduces the fundamental principles of spectral CT and PCD CT, their advantages in neuro-oncologic applications, and their potential to improve diagnostic imaging through enhanced image quality and quantitative accuracy. Keywords: Physics, CT-Photon-counting, Pediatrics, CT-Spectral © RSNA, 2026.

Purpose To evaluate whether artificial intelligence (AI)-derived chamber volumetry from coronary artery calcium (CAC) CT improves heart failure (HF) risk prediction. Materials and Methods This retrospective study included asymptomatic patients without known cardiac disease undergoing CAC CT between 2010 and 2023. CAC CT images were analyzed using a validated AI model to calculate chamber volumes. HF events were identified based on International Classification of Diseases, Ninth and Tenth Revisions codes. Cox proportional hazard regression models incorporating chamber volumes, adjusted for Predicting Risk of cardiovascular disease EVENTs-Heart Failure (PREVENT-HF) score, were used to assess the association with HF. Time-dependent areas under the receiver operating characteristic curve (AUCs) at 3, 5, 8, and 10 years were calculated to compare the performance of volumetry, PREVENT-HF, CAC scoring, and their combination. Results A total of 5892 patients were included (mean age ± SD, 58.2 years ± 9.4; 3258 male). During a mean follow-up of 4 years ± 3, 377 patients (6.3%) developed HF. Larger left atrial, left ventricular, right atrial, and left ventricular myocardial volumes were associated with increased HF risk (P < .001). The composite, multivariable Cox regression model incorporating all chamber volumes, CAC score, and PREVENT-HF scores outperformed PREVENT-HF (AUC, 0.80 vs 0.76; ΔAUC, 0.04; P < .001) and CAC scores (AUC, 0.80 vs 0.70; ΔAUC, 0.10; P < .001) alone in predicting 10-year HF risk. A phase-volume interaction effect was identified, indicating that diastolic volumes were independently associated with higher HF risk than were systolic volumes (P < .001), adjusted for PREVENT-HF. Conclusion AI-derived cardiac chamber volumetry obtained from CAC CT improved HF risk prediction compared with PREVENT-HF or CAC scores alone. Keywords: Applications-CT, Deep Learning, Cardiac Supplemental material is available for this article. © RSNA, 2026.

Background Photon-counting CT (PCCT) can theoretically be used to quantify extracellular volume fraction (ECV) for liver fibrosis assessment, but its performance remains unclear. Purpose To determine the feasibility of PCCT-derived ECV for staging liver fibrosis by assessing its correlation with MR elastography-derived liver stiffness measurement (LSM) and comparing their diagnostic performance, using histopathologic findings as the reference standard. Materials and Methods Between July 2024 and July 2025, 157 participants with suspected hepatic malignancies were prospectively enrolled and underwent PCCT and MRI at Ruijin Hospital. Intraparticipant comparisons of ECV and LSM were performed, and their diagnostic performance in staging liver fibrosis was evaluated using histopathologic findings as the reference standard. Subgroup analyses of fibrosis staging using ECV were performed among participants with coexisting steatosis or inflammation and participants with a body mass index of 25 or higher. Correlation, receiver operating characteristic, and equivalence analyses were performed. Results Ultimately, 139 participants (mean age, 62 years ± 8 [SD]; 114 male) were included. ECV was strongly correlated with LSM (Spearman ρ = 0.84; P < .001) in an intraparticipant-level comparison. When histopathologic examination was used as the reference standard, the area under the receiver operating characteristic curve (AUC) values for ECV were 0.99 (95% CI: 0.98, 1.00), 0.98 (95% CI: 0.96, 1.00), and 0.98 (95% CI: 0.96, 1.00) for participants with fibrosis stages F2 or higher, F3 or higher, and F4, respectively. Equivalence analysis revealed comparable diagnostic performance between ECV and LSM (95% CI differences in AUCs: 0.005, 0.055 for stage F2 or higher; -0.015, 0.027 for stage F3 or higher; and -0.021, 0.012 for stage F4). With use of the cutoffs derived from the whole-group analysis, ECV showed excellent agreement with histopathologic stage overall and across each subgroup (all weighted κ coefficients ≥0.86; P < .001 for all). Conclusion PCCT-derived ECV was strongly correlated with MR elastography-derived LSM and showed equivalent, clinically feasible performance for liver fibrosis staging. © RSNA, 2026 Supplemental material is available for this article. See also the editorial by Wu and Shi in this issue.

Background With the advent of immune checkpoint inhibitor-based combination therapy, treatment sequencing for advanced hepatocellular carcinoma (HCC) has become increasingly complex. The Barcelona Clinic Liver Cancer (BCLC) 2026 recommendations emphasize individualized decision-making. Purpose To compare clinical outcomes of hepatic arterial infusion chemotherapy (HAIC) and tyrosine kinase inhibitors (TKIs) in patients with advanced HCC after atezolizumab-bevacizumab (AB) failure, within the BCLC 2026 treatment paradigm. Materials and Methods This multicenter retrospective study included patients who received AB and subsequently received either HAIC or a TKI between March 2022 and August 2025. HAIC used a cisplatin-fluorouracil regimen, and TKIs were given at standard doses. Tumor response and progression-free survival (PFS) were assessed using contrast-enhanced multiphase CT or liver MRI according to routine clinical practice. Imaging was reviewed by radiologists blinded to treatment allocation. Inverse probability of treatment weighting (IPTW) was applied for age, sex, Eastern Cooperative Oncology Group performance status, Child-Pugh class, portal vein tumor thrombosis, and tumor burden based on the up-to-seven criteria. Results This study included 90 patients (mean age, 62 years ± 10.9 [SD]; 73 men; HAIC group, n = 51; TKI group, n = 39). Baseline tumor burden was higher in the HAIC group than in the TKI group (percentage of patients with tumors beyond the up-to-seven criteria: 88% [45 of 51] vs 64% [25 of 39]; P = .01). In unweighted analyses, median overall survival (OS) was similar between the HAIC and TKI groups (10.4 vs 6.4 months; P = .62), whereas PFS favored HAIC over TKIs (median, 5.3 vs 3.6 months; P = .008). Objective response rate and disease control rate were higher with HAIC than with TKIs (objective response rate: 35% [18 of 51] vs 5% [two of 39], P = .002; disease control rate: 67% [34 of 51] vs 26% [10 of 39], P < .001). After IPTW, HAIC remained associated with longer PFS than TKIs (median, 7.1 vs 3.4 months; P < .001), and OS remained similar between groups (median, 10.5 vs 6.3 months; P = .32). Conclusion In patients with advanced HCC, after AB failure, second-line HAIC yielded higher response rates and longer PFS than TKIs. © RSNA, 2026 Supplemental material is available for this article. See also the editorial by Deyirmendjian and Tang in this issue.

Background Clinical histories accompanying imaging orders guide protocol selection and diagnostic focus. However, they are often incomplete, potentially compromising diagnostic accuracy and workflow efficiency. Purpose To evaluate whether large language models (LLMs) can improve the clinical utility of provided imaging indications by leveraging clinical notes. Materials and Methods This retrospective study curated a dataset from deidentified electronic health records at the University of California San Francisco (January 2012 to August 2024), consisting of radiology reports with paired referring clinician-provided and radiologist-curated indications linked to clinical notes. The dataset was stratified across five body systems and five pathophysiologic categories to derive LLM selection and reader study internal test sets. For the reader study, 20 radiologists with 2-25 years of experience compared indications from the referring clinician, radiologist, and best-performing LLMs. Readers scored comprehensiveness, factuality, and conciseness and ranked indications for usefulness in protocoling, usefulness in interpretation, and overall ranking. Models and clinicians were compared using cumulative link mixed models with Tukey-adjusted post hoc comparisons. Results From 28 313 patients (mean age, 59 years ± 20.6 [SD]; 14 912 women), 250 examinations from 247 patients were sampled for the reader study. After nine exclusions, 241 examinations were analyzed, yielding 482 reader-examination evaluations. Indications from the best-performing proprietary (Claude 3.5 Sonnet; Anthropic) and open-source (Qwen 2.5-7B Instruct; Alibaba) LLM were rated as more comprehensive (Likert rating of 5: 37.14% and 28.42%, respectively; both P < .001) and factual (68.05% and 59.75%; both P < .001) than referring clinician indications. The proprietary LLM ranked most useful in protocoling (rank 1: 40.87%; all P < .001), useful in interpretation (44.61%; all P < .001), and overall ranking (44.19%, all P < .001). Comprehensiveness (65.77% of ratings; both P < .001) most strongly influenced overall rankings. Conclusion LLMs generated radiology-relevant indications from clinical notes that were more comprehensive and factual than clinician indications, and when generated by the proprietary LLM, were ranked most useful in protocoling and imaging interpretation. © RSNA, 2026 Supplemental material is available for this article. See also the editorial by Yilmaz and Cardoza-Ochoa in this issue.

Generative artificial intelligence (AI) has introduced new possibilities for assisting radiologists in diagnostic interpretation and workflow, particularly through automated report generation. Chest radiography has emerged as a leading use case for generative AI due to its high clinical volume, its diagnostic complexity, and the availability of large public datasets. This article describes a practical, experience-based approach to understanding and implementing generative AI for chest radiograph reporting. It outlines key concepts and the underlying architecture of generative models, followed by a clinical framework that positions AI as a draft assistant, highlighting both its utility and limitations. Through representative examples, the article demonstrates how generative AI can accelerate workflow, promote reporting consistency, and support trainee education, while also addressing known pitfalls such as hallucinations. It discusses implementation strategies, interface considerations, and safety checkpoints that support effective human-in-the-loop review. The article concludes with future directions that emphasize the importance of context-aware modeling, explainability, and multi-institutional evaluation. The aim is to equip radiologists with essential knowledge and insights to engage safely and effectively with generative AI tools in thoracic radiology practice, using chest radiography as a paradigm for broader integration across diagnostic imaging.