Citations

The field of radiology is experiencing rapid adoption of large language models (LLMs), yet their tendency to generate hallucinations (plausible but incorrect information) remains a significant barrier to trust. This comprehensive review evaluates emerging agentic artificial intelligence (AI) approaches, including multi-agent role-based systems, retrieval-augmented generation (RAG), and uncertainty quantification, to assess their potential for reducing hallucinations in radiology workflows. Evidence from 2024 to 2025 demonstrates that agentic AI can improve diagnostic accuracy and reduce error rates, though these methods remain computationally demanding and lack comprehensive clinical validation. Multi-agent frameworks enable cross-validation through role-based specialization and systematic workflow orchestration, while RAG strategies enhance accuracy by grounding responses in verified medical literature. Within multi-agent systems, uncertainty quantification enables agents to communicate confidence levels to one another, allowing them to appropriately weigh each other’s contributions during collaborative analysis. While multi-agent frameworks and RAG strategies show significant promise, practical deployment will require careful integration with human oversight, robust evaluation metrics tailored to medical imaging tasks, and regulatory adaptation to ensure safe clinical use in diverse patient populations and imaging modalities.

Abstract Glioblastoma, IDH-wildtype (GBM), and primary central nervous system lymphoma (PCNSL) are two aggressive brain tumors with distinct treatment strategies but overlapping imaging features on MRI. Accurate preoperative differentiation remains a clinical challenge, due to their overlapping imaging features, even when advanced MRI techniques are employed. This study explored deep learning approaches leveraging multi-sequence MRI to distinguish GBM from PCNSL. Models were developed utilizing 146 PCNSL cases, which were matched to146 GBM by age, sex and year of diagnosis. A 3D-DenseNet121 convolutional neural network was trained using two analytical strategies: (i) loss-based model selection and (ii) ensemble prediction using five-fold cross-validation. Model performance was evaluated on an independent validation set of 256 GBM and 37 PCNSL from the same institution as well as 36 PCNSL cases from an external institution. All MRIs were preprocessed using the FeTS pipeline. Using the independent validation cohort, area under the receiver operating characteristic curve (AUC) and 95% confidence intervals (CI) were computed overall and stratified by demographic characteristics. Using both T1 post-contrast (T1Gd) and T2-weighted images, the loss-based model achieved AUC=0.83 (95% CI: 0.77-0.88) and the ensemble model yielded an AUC=0.82 (95% CI: 0.76-0.89). Model performance was similar across males and females; however, model performance was better for subjects older than 50 years of age versus younger than 50. When using only T1Gd images, the loss-based model dropped to AUC=0.55 (95% CI: 0.48-0.63), whereas the ensemble model had AUC=0.79 (95% CI: 0.72-0.85). Models trained with only T2 images achieved AUCs of 0.67 (95% CI: 0.60-0.73) (loss-based) and 0.75 (95% CI: 0.69-0.82) (ensemble). These results highlight the added value of the multi-sequence input. Our results confirm the feasibility of using automated deep learning models to differentiate GBM from PCNSL. Notably, the model generalized well to external MRI data, emphasizing its potential utility as a diagnostic aid in real-world clinical workflows.

Abstract BACKGROUND The majority of pituitary neuroendocrine tumors (PitNETs) are effectively managed with medical therapy, surgery, and/or radiation therapy; however, a subset of pituitary adenomas exhibit aggressive behavior. Temozolomide is thought to be an effective treatment option for patients with PitNETs. METHODS We identified and retrospectively reviewed 11 patients with pituitary tumors treated with temozolomide. We then conducted a review of the literature, combining 114 previously published cases with our cohort. RESULTS 125 cases were identified with diagnoses of pituitary adenomas (n=45), aggressive pituitary tumors (n=61), and pituitary carcinoma (n=19) treated with temozolomide. 79 patients were male (68.1%) with a median age of 47.5 (range 9-72). A variety of tumor types were represented: 50 corticotroph adenomas (40.7%), 37 prolactinomas (30.1%), 22 hormone nonfunctioning adenomas (17.9%), 8 somatotroph adenomas (6.5%), 6 other functioning adenomas (4.9%), and 2 cases where the functional status was not mentioned. Median treatment duration was 9 cycles (n=91, 150-200 mg/m2 days 1-5, 28-day cycles; n=2, 75 mg/m2, days 1-21, 28-day cycles; n=1, 75 mg/m2 daily). 27 patients received concurrent therapy with a dopamine agonist/somatostatin analogue (n=10), other chemotherapy agent (n=9) or radiotherapy (n=8). Radiographic responses included stable disease (n=40), partial response (n=35), complete response (n=8), and progressive disease (n=30). Hormonal responses included complete response (n=13), partial response (n=32), and no response (n=16). 22 patients (47.8%; out of 46 patients where progression/recurrence was assessed) experienced progression/recurrence following temozolomide initiation with a median progression free survival (PFS) of 24 months. There was no significant association between MGMT methylation status and radiological or hormonal response. No unexpected toxicities were observed. CONCLUSION In this cohort, temozolomide was well-tolerated with a median PFS of 24 months. MGMT methylation status did not predict temozolomide response. Temozolomide is an option with a survival benefit for patients with PitNETs.

Importance Treatment for cognitive dysfunction due to postacute sequelae of long COVID (ie, symptoms of fatigue, malaise, weakness, confusion that persist beyond 12 weeks after an initial COVID infection) remains a significant unmet need. Objective To test evidence-based rehabilitation strategies for improving cognitive symptoms in persons with long COVID. Design, Setting, and Participants This was a 5-arm, multicenter, randomized clinical trial of 3 remotely delivered interventions conducted between August 17, 2023, and June 10, 2024. The study took place at 22 trial sites and included the screening of individuals with cognitive long COVID. Interventions Participants were randomized to 1 of 5 arms: adaptive computerized cognitive training (BrainHQ [Posit Science]), cognitive-behavioral rehabilitation involving both group and individual counseling sessions (PASC-Cognitive Recovery [PASC-CoRE]) paired with BrainHQ, and transcranial direct current stimulation (tDCS) paired with BrainHQ. Two comparator arms were included as follows: unstructured computer puzzles and games (active comparator) and sham tDCS paired with BrainHQ. The interventions occurred 5 times per week over 10 weeks. Main Outcomes and Measures Cognitive and behavioral in-person assessments were performed at baseline, midintervention, at the end of intervention, and 3 months after the end of the intervention. The primary outcome measure was the modified Everyday Cognition Scale 2 (ECog2) completed at the end of the intervention compared to the baseline visit based on participant self-report looking back over the prior 7 days. Results A total of 378 individuals were screened, from which there were 328 participants (median [IQR] age, 48.0 [37.0-58.0] years; 241 female [73.5%]; race: 15 Asian [4.6%], 47 Black [14.3%], and 235 White [71.6%]; ethnicity: 52 Hispanic [15.9%]). None of the 3 active interventions demonstrated benefits on the modified ECog2 in the intention-to-treat population by the end of the intervention period. The adjusted differences in mean change were 0.0 (95% CI, −0.2 to 0.2) for BrainHQ vs active comparator, 0.1 (95% CI, −0.1 to 0.3) for PASC-CoRE + BrainHQ vs active comparator, 0.0 (95% CI, −0.2 to 0.2) for tDCS-active + BrainHQ vs tDCS-sham + BrainHQ, and 0.1 (95% CI, −0.1 to 0.3) for PASC-CoRE + BrainHQ vs BrainHQ alone. Secondary participant-reported outcomes and neuropsychological tests showed no differential benefits for any treatment arm. All 5 arms demonstrated some improvements over time on the modified ECog2 and on secondary outcomes. There were no serious adverse events attributable to the interventions. Conclusions and Relevance This phase 2 randomized clinical trial failed to demonstrate differential benefits for online cognitive training, a structured cognitive rehabilitation program, and tDCS for cognitive long COVID. Trial Registration ClinicalTrials.gov Identifier: NCT05965739 .

OBJECTIVE While endonasal transsphenoidal surgery (ETS) is the standard treatment for nonfunctioning pituitary adenomas (NFPAs), the postoperative recovery rates of secondary adrenal insufficiency and their predictive factors remain poorly understood. The authors investigated the relationship between the tumor resection rate and adrenal function recovery in patients with NFPAs. METHODS This retrospective study included 450 patients with NFPAs who underwent ETS between January 2013 and November 2023. The rates of preoperative hypopituitarism and postoperative recovery rates of secondary adrenal insufficiency and their associated factors were analyzed. Recovery was defined as normalization of serum cortisol levels without replacement therapy for at least 24 hours, accompanied by symptom resolution. The tumor resection rate was calculated as the percentage of the tumor volume resected. RESULTS The median follow-up period for the entire cohort was 52 months (range 1–131 months). A total of 246 (55%) patients had dysfunction in one or more pituitary axes. Fifty-three (47%) of 114 patients with preoperative adrenal insufficiency showed postoperative adrenal insufficiency recovery. The cumulative adrenal insufficiency recovery rates were 12% at 1 month, 19% at 3 months, 32% at 6 months, and 41% at 12 months. Among the 114 patients with preoperative adrenal insufficiency, the factors associated with postoperative adrenal insufficiency recovery were investigated. Patients with smaller tumor sizes of < 20 mm (log-rank test, p = 0.001) and those with tumor resection rates of more than 85% (log-rank test, p = 0.032) demonstrated better postoperative adrenal insufficiency improvement. Bivariate Cox proportional hazards analysis demonstrated that smaller tumor sizes (continuous, HR 0.97, 95% CI 0.93–0.98; p = 0.043) and higher tumor resection rates (HR 1.03, 95% CI 1.01–1.07; p = 0.001) were significantly associated with better postoperative adrenal insufficiency recovery. Multivariable analysis confirmed that only higher tumor resection rates were significantly associated with better postoperative adrenal insufficiency recovery (HR 1.03, 95% CI 1.01–1.06; p = 0.019). CONCLUSIONS Nearly half of the patients with preoperative adrenal insufficiency achieved recovery after ETS for NFPAs. The enhanced extent of tumor removal was an independent factor associated with better postoperative recovery, indicating that optimizing safe maximal resection procedures may be crucial in enhancing adrenal function.

Introduction: Carotid artery intraplaque hemorrhage (IPH) is increasingly recognized as a predictive marker of cerebrovascular events. Despite widespread utilization of contrast-enhanced CT angiography (CTA) for the evaluation of carotid atherosclerosis, diagnosis of IPH on CTA is challenging, limiting risk stratification capabilities. We evaluated convolutional neural networks (CNN) for detection of histology-proven IPH in carotid CTA. Methods: The study included carotid endarterectomy patients from our center who had preoperative CTA and available plaque specimens from the Mayo Clinic Carotid Artery Atherosclerosis Trial (MCAT) Biobank. A validated segmentation model localized common and internal carotid arteries (CCA/ICA). CCA bifurcation served as landmark after a large language model, Llama3.0-70B, verified culprit plaque proximity to the bifurcation from CTA reports. A 3×3×4cm box around ipsilateral CCA bifurcation was sampled from CTA volumes and images were labeled IPH by histology study of carotid specimen (Fig 1). Given limited cases with concurrent imaging and histology, we employed a robust cross-validation strategy: 22 cases (20%) held-out for testing, with remaining data 86 cases used for 5-fold cross-validation (stratified by IPH presence). The five resulting models independently generated IPH probability scores, and ensembled predictions calculated by three methods to enhance generalizability: Majority and weighted voting based on binary IPH class and soft voting based on IPH probability scores. Results: Of 418 patients with histological analysis, 130 had CTA studies available. After excluding 22 cases lacking histological consensus, 108 cases were analyzed. In the 22 test cases the mean age was 68.1±8.5 years and 59% were male; In the cross-validation folds age ranged from 67.1±6.9 to 75.3±7.6 years 56-82% were male. Individual models' performance on the test set showed mean ROC-AUC of 0.79 (95% CI: ±0.05) and average precision of 0.85 (±0.05) (Fig 2). The ensemble approach achieved superior performance with soft voting demonstrated 82% ROC-AUC, 77% accuracy, 79% precision, 85% sensitivity, and 82% F1-score for IPH detection, outperforming majority voting and weighted voting (Fig 3). Conclusion: Automated IPH detection in widely available CTA imaging can turn it to a powerful stroke risk assessment tool, supporting personalized prevention strategies and surgical decision-making, ultimately reducing cerebrovascular morbidity and mortality.

The COVID-19 pandemic caused unprecedented disruption of social networks and negatively impacted social determinants of health (SDoH, e.g., economic stability, mental health, health care access) among African Americans. The FAITH! (Fostering African-American Improvement in Total Health!) Program, an academic-community partnership with African American churches, pivoted to pandemic-related impacts on health behaviors of African Americans. This exploratory, cross-sectional study examined associations between multilevel factors influencing difficulty maintaining a healthy lifestyle during the early pandemic phase, including mental/emotional health and COVID-19 hardships (e.g., job/food/housing insecurity, paying utilities). An online survey was distributed via FAITH!-affiliated churches and social media. Logistic regression (odds ratios [ORs], adjusted for age/gender) and associated 95% confidence intervals [CIs]) examined associations between difficulty maintaining a healthy lifestyle and multilevel factors. Greater than half of respondents reported difficulty maintaining a healthy lifestyle (N = 169, 54% [91/169]). Adjusted odds of having difficulty maintaining a healthy lifestyle were greater for those reporting mental/emotional health issues (high-stress OR 4.0; 95% CI [1.9,8.9], p < .001); occasional symptoms of depression (OR 3.7; 95% CI [01.9,7.7]; p < .001); anger (OR 2.5; 95% CI [1.2,5.4]; p = .044), and anxiety (OR 2.8; 95% CI [1.4,6.0]; p = .008)). COVID-19 hardships of job insecurity (OR 1.4.; 95% CI [0.7, 2,8]; p = .015), difficulty paying rent (OR 4.4; 95% CI [2.1,9.9]; p < .001) or difficulty paying for food/utilities (OR 5.4; 95% CI [2.6,11.7]; p < .001) were associated with difficulty maintaining a healthy lifestyle. Our study revealed the negative impact of mental/emotional health and COVID-19 hardships on maintaining a healthy lifestyle among African Americans. This could inform SDoH-focused lifestyle interventions during future public health crises.

OBJECTIVE Nonfunctioning pituitary adenomas (NFPAs) often present with associated mild hyperprolactinemia, described in literature as "stalk effect." The presence of preoperative hyperprolactinemia caused by tumor compression of the pituitary stalk and endocrinological outcomes has been noted; however, a detailed investigation on the impact on perioperative endocrinological outcome is lacking. The authors aimed to determine whether elevated preoperative serum prolactin concentrations are associated with increased risk of perioperative pituitary dysfunction in patients with NFPAs undergoing endonasal transsphenoidal surgery (ETS). METHODS The authors conducted a single-center retrospective cohort study of patients who underwent ETS for NFPAs between January 2013 and April 2023. Patients were divided into two groups based on preoperative prolactin concentrations: normal or hyperprolactinemic (prolactin values above upper limit of reference range, adjusted for sex). Outcomes included perioperative endocrinological function and surgical outcomes. Multivariable analysis was conducted to identify factors associated with perioperative endocrinological function outcomes. RESULTS Of 244 patients (140 [57%] males; median age 59 years) with a median follow-up of 65 months, 131 (54%) had hyperprolactinemia. Compared with patients with normal preoperative prolactin levels (n = 113), the hyperprolactinemia group had similar tumor size (median diameter 2.5 cm vs 2.3 cm, p = 0.102) but higher rates of cavernous sinus invasion (99% vs 91%, p = 0.035), pituitary stalk deviation (95% vs 74%, p = 0.001), stalk bending (14% vs 1%, p = 0.001), preoperative secondary hypothyroidism (39% vs 22%, p = 0.005), and adrenal insufficiency (AI; 34% vs 22%, p = 0.048). Postoperatively, the hyperprolactinemia group showed higher rates of new secondary hypothyroidism (16% vs 7%, p = 0.048) and new permanent AI (16% vs 6%, p = 0.046). Multivariable analysis revealed preoperative high prolactin as an independent predictor of preoperative secondary hypothyroidism (OR 1.83, 95% CI 1.17–3.44; p = 0.046) and its postoperative deterioration (OR 2.65, 95% CI 1.02–7.69; p = 0.042). Maximum tumor diameter was also a significant predictor of various endocrinological dysfunctions both pre- and postoperatively. CONCLUSIONS Preoperative hyperprolactinemia in patients with NFPA is associated with higher risks of specific perioperative endocrinological dysfunction and pronounced pituitary stalk deviation. These findings highlight the utility of preoperative prolactin levels as a biomarker for predicting postoperative endocrinological outcomes, potentially enhancing risk assessment and guiding personalized perioperative management.

Radiology is undergoing a paradigm shift from traditional single-function AI systems to sophisticated multi-agent networks capable of autonomous reasoning, coordinated decision-making, and adaptive workflow management. These agentic AI systems move beyond simple pattern recognition to encompass complex radiological workflows including image analysis, report generation, clinical communication, and care coordination. While multi-agent radiological AI promises enhanced diagnostic accuracy, improved workflow efficiency, and reduced physician burden, it simultaneously amplifies the long-standing “black box” problem. Traditional explainable AI methods, which are adequate for understanding isolated diagnostic predictions, fail when applied to multi-step reasoning processes involving multiple specialized agents coordinating across imaging interpretation, clinical correlation, and treatment planning. This paper examines how agentic AI systems in radiology create “compound opacity” layers of inscrutability from agent interactions and distributed decision-making processes. We analyze the autonomy–transparency paradox specific to radiological practice, where increasing AI capability directly conflicts with interpretability requirements essential for clinical trust and regulatory oversight. Through examination of emerging multi-agent radiological workflows, we propose frameworks for responsible implementation that preserve both diagnostic innovation and the fundamental principles of medical transparency and accountability.