Citations

Abstract Despite growing interest in AI for pancreatic cyst management, no prior study has systematically investigated how lesion characteristics influence model behavior or provided a comprehensive explainability analysis in IPMN malignancy risk stratification. We present the first multi-center study integrating explainable AI with uncertainty quantification to evaluate how cyst type, size, and location affect deep learning performance in IPMN malignancy prediction. Our retrospective study analyzed 170 IPMNs from seven centers using a radiomics-deep learning fusion model. Cases were stratified by dysplasia grade, IPMN type, size, and location; with model interpretability assessed using SHAP, LIME, Grad-CAM, and uncertainty quantification. Overall accuracy was 67.1%, with model uncertainty lower for correct versus incorrect predictions (0.72 vs 0.78, p < 0.001). On subgroup analysis: Low-Risk lesion accuracy exceeded High-Risk (81.3% vs. 42.9%, p < 0.001), BD accuracy was inversely correlated with cyst size while MD remained stable, and body and whole-pancreas cysts stratified more accurately than tail cysts (77.8%, 76.6% vs. 47.8%). AI models face significant, previously uncharacterized performance drops when evaluating complex high-risk IPMNs. Our study shows that integrating uncertainty quantification can successfully flag unreliable predictions, enabling a selective-prediction framework where clinicians can confidently rely on AI for low-risk, low-uncertainty lesions while deferring high-uncertainty cases to expert review.

ABSTRACT Objective Falls are a leading cause of morbidity and loss of independence in adults. The association between overactive bladder (OAB) and fall risk has not been thoroughly investigated. This study evaluated the association between OAB and fall history in a nationally representative sample of US adults aged 20−69 years. Methods We performed a cross‐sectional analysis of adults aged 20−69 years from the 2021−2023 National Health and Nutrition Examination Survey (NHANES). OAB was classified using a composite symptom severity score that assessed urge urinary incontinence and nocturia. Fall history was assessed by self‐reporting (only in participants aged 20−69 years), where a recent fall was defined as experiencing at least one fall within the past year. Survey multivariable logistic regression was used to estimate adjusted odds ratios (aOR) for the association between fall history and OAB after controlling for sociodemographic and clinical covariates. Results Among 4118 participants, the population‐weighted prevalence of OAB was 19.6% (standard error [SE]: 1.1%), and the prevalence of a recent fall was 28.3% (SE: 1.3%). The prevalence of a recent fall was higher among those with OAB (39.7% [SE: 1.6%] vs. 25.7% [SE: 1.5%]). The prevalence of a recent fall increased progressively with OAB symptom severity (24.3% in those with no symptoms vs. 52.4% in those with severe symptoms). In the fully adjusted model controlling for sociodemographic and clinical covariates, OAB was independently associated with a 42% higher odds of experiencing a recent fall (aOR, 1.42; 95% CI, 1.10 to 1.83; p = 0.01). Conclusion OAB is independently associated with an increased risk of a recent fall in US adults aged 20‐69 years, with a strong positive relationship between OAB severity and fall risk. These findings suggest that screening for OAB may be a useful adjunct to fall risk assessment. Future studies are needed to determine whether OAB management reduces the risk of falls.

Abstract Background Genetic testing is increasingly used in presurgical evaluation, but the yield of resection across germline genetic epilepsies remains uncertain. Methods We conducted a systematic review of MEDLINE (PubMed) and Scopus and added cases from three institutional cohorts and the Pediatric Epilepsy Research Consortium (PERC) databases, including individuals with pathogenic/likely pathogenic germline variants besides tuberous sclerosis and neurofibromatosis who underwent resection or laser ablation. Etiologies were grouped into biologically informed categories (GATORopathies, vascular, overgrowth, CNVs, channelopathies, synaptopathies, other). Primary outcome was seizure freedom (Engel I) at last follow-up. Group comparisons used Fisher’s exact and Kruskal–Wallis tests (α=0.05; Bonferroni when applicable). Prespecified sensitivity analyses stratified by lesional status, excluded GATORopathies, and restricted to literature-only cases. Results We included 223 literature cases (64 studies), 35 institutional cases, and 11 PERC cases (n=269). Median follow-up was 24 months (IQR 12–48.0). Seizure freedom was achieved in: vascular disorders 14/19 (73.6%), GATORopathies 79/120 (67.5%), CNVs 18/31 (66.7%), overgrowth 7/13 (53.8%), other 16/25 (41.7%), channelopathies 13/43 (33.3%), and synaptopathies 4/18 (22.2%) (overall p<0.001). Among cases with known imaging, 208/253 (82.2%) had epileptogenic MRI lesions, including 64% of channelopathies and 80% of the synaptopathies. In univariate contrasts (each category vs all others), odds of seizure freedom were higher for vascular disorders (2.66-fold, 95% CI 0.87–9.72) and GATORopathies (2.46-fold, 95% CI 1.46–4.18), and lower for synaptopathies (∼4.2-fold lower, OR 0.24, 95% CI 0.05– 0.78) and channelopathies (∼4.5-fold lower, OR 0.22, 95% CI 0.09–0.48). Direction and magnitude were consistent across prespecified sensitivity analyses (lesional-only, literature-only, exclusion of GATORopathies). Conclusions Resective surgeries can be effective in germline genetic epilepsies, but outcomes vary by pathway. Disorders with discrete, lesional substrates (GATORopathies, vascular) show the highest likelihood of seizure freedom, whereas channelopathies and synaptopathies, despite the presence of MRI lesions, have substantially lower yields even in lesional cases. Prospective, genotype-aware surgical registries with standardized reporting (EEG, imaging, pathology) and time-to-event outcomes are needed to refine the selection of surgical candidates and quantify seizure and non-seizure-related outcomes.

Abstract The purpose of the current study was to identify the influence of individual characteristics on differences in body fat percentage between air displacement plethysmography and dual-energy X-ray absorptiometry (change in body fat percentage) in 360 collegiate student-athletes. Body fat percentage was obtained from air displacement plethysmography and dual-energy X-ray absorptiometry. A 2 (sex)×3 (Brozek body fat percentage, Siri body fat percentage, and dual-energy x-ray absorptiometry body fat percentage) analysis of variance was used to evaluate the difference between the body fat percentages obtained from air displacement plethysmography and dual-energy x-ray absorptiometry in men and women. The agreement between air displacement plethysmography and dual-energy X-ray absorptiometry was evaluated via Bland–Altman plot and linear regression. Multiple linear regression was used to identify the influence of sex, race, height, and body mass on the change in body fat percentages. Dual-energy X-ray absorptiometry yielded greater body fat percentages in both men (4.8±2.8; p < 0.001) and women (2.2±2.9; p < 0.001) compared to air displacement plethysmography measurements. Proportional and systematic biases were observed between air displacement plethysmography and dual-energy X-ray absorptiometry. Race, height, and mass were significant predictors of the body fat percentage difference between air displacement plethysmography and dual-energy X-ray absorptiometry (R 2=0.445, p<0.001, and standard error of the estimate=2.527). The present study suggests that air displacement plethysmography underestimates body fat percentages when compared to dual-energy X-ray absorptiometry and this underestimation may not be the same for every athlete due to their individual characteristics.

ABSTRACT Background A randomized phase II screening trial of gemcitabine, nab‐paclitaxel, and cisplatin with a medically supervised ketogenic diet (MSKD) versus usual diet (non‐MSKD) was conducted in patients with treatment‐naive metastatic pancreatic ductal adenocarcinoma (PDAC). Methods Patients with untreated metastatic PDAC were randomized 1:1 to MSKD or non‐MSKD while receiving gemcitabine, nab‐paclitaxel, and cisplatin on days 1 and 8 of a 21‐day cycle. The MSKD was guided by a remote health care team and daily ketone (beta‐hydroxybutyrate) levels, with goal beta‐hydroxybutyrate of 0.5 to 3.0 mM. The primary endpoint was progression‐free survival (PFS) using a one‐sided alpha level of 0.20. Secondary endpoints included overall survival (OS), safety, and quality of life (QOL). Changes in microbiome were an exploratory endpoint. Findings Overall, there were 32 evaluable patients. In the MSKD arm, 15 of 16 patients achieved nutritional ketosis; the median proportion of days in ketosis was 39.4%. The median PFS was 8.5 months in MSKD patients and 6.2 months in non‐MSKD patients: hazard ratio, 0.53 (95% CI, 0.21–1.37); one‐sided p = .096. The median OS was 13.7 months with MSKD and 10.2 months in the non‐MSKD arm: hazard ratio, 0.58 (95% CI, 0.25–1.37); one‐sided p = .107). All MSKD‐related adverse events were grade 1‐2. There were no significant differences in grade ≥3 chemotherapy‐related adverse events between the arms. MSKD patients had no decline in QOL and had significant enrichment of beneficial taxa in the microbiome ( p < .05, log‐fold change ≥2). Conclusions The MSKD is feasible in patients with PDAC and, although not powered for definitive outcomes, shows trends in improved PFS and OS when combined with gemcitabine, nab‐paclitaxel, and cisplatin, without added toxicity or detriment to QOL. Larger studies are required to confirm these findings and establish the value of the MSKD in pancreatic cancer treatment.

Abstract Medulloblastoma (MB) is one of the most prevalent malignant brain tumors in children, with tremendous cognitive and neuroendocrine disability among survivors. Group 3 (G3) MBs have poor overall survival at &lt; 50%, higher frequency of metastasis, and no targeted therapies. Amplification of MYC and activation of TGFβ signaling are frequent in G3MB. Remarkably, some tumors have no reported mutations, suggesting roles for epigenetic mechanisms in driving disease. We have generated novel humanized models for G3MB by transducing neuroepithelial stem cells (NESC) derived from human induced pluripotent stem (iPSC) cells with TGFβ effectors alone and/or in combination with MYC, prioritizing combinations observed in patients. We determined that both MYC and TGFβ effectors drove tumor formation in vivo with the combination driving more aggressive tumors and higher frequency of metastasis. NESC-derived tumors clustered with human G3/G4MB, indicating our models recapitulated human disease. To decipher mechanisms of resistance, we integrated CUT&RUN for MYC genomic localization and histone PTMs with RNA-sequencing and discovered MYC-bound neural developmental genes repressed in MYC and TGFβR1-driven lines. Further interrogation identified JARID1B, a MYC binding partner, as a regulator of many of these repressed neural genes. We postulate that deregulation of MYC and context-specific targeting to the chromatin template culminates in transcriptional alterations that promote aggressive G3MB. To further investigating therapeutic vulnerabilities, we have performed a in vivo whole genome CRISPRi screen to reveal important drivers of malignancy, one of the first of it’s kind in brain tumors. We are functionally validating key candidate sgRNAs that were differentially significant in primary tumors and may also have key roles in driving metastasis. Excitingly, many sgRNAs identified from the screen were chromatin modifiers and may be important co-factors of MYC-mediated regulation on the epigenome. Collectively, our studies provide a critical framework for understanding mechanisms of resistance and the epigenetic landscape of MYC- and TGFβ-driven MB.

OBJECTIVE Electrical stimulation mapping is a widely used technique to determine functional localization for medically refractory epilepsy. Sites where stimulation interferes with naming are often focal regions of the frontal and temporal cortex. The extent to which these crucial sites remain in the same location over time in an individual adult patient has not yet been established. The aim of this study was to determine whether cortical naming sites identified using stimulation mapping are stable in their anatomical location over time in adult patients with medically refractory epilepsy. METHODS Twenty-two patients who underwent electrical stimulation mapping for medically refractory epilepsy during surgical interventions separated by more than 1 year between 1967 and 2005 were included. A median of 8.35 years elapsed between mappings. The mean age at the first operation was 27.7 (range 10–39) years. Fourteen patients were female. Mapping occurred under two different conditions: intraoperatively in procedures conducted under local anesthesia or extraoperatively through implanted grid electrodes. A Bayesian hierarchical model of language site locations across repeated interventions was used to assess the stability of locations of stimulation-evoked interference in language naming. RESULTS Sites where electrical stimulation interferes with language naming were separated by a median of 0.6 cm between the 2 mappings. Eighty-six percent of the mapped sites related to language naming at the second operation were within 1.5 cm of a site identified at the first operation, 61% within 1 cm, and 36% within 0.5 cm. However, in 2 patients, none of the identified language naming sites at the second operation were within 1.5 cm of the sites from the first operation. CONCLUSIONS This unique, long-term series of neurosurgical mappings reveals that language naming sites in the cortex of adult patients with epilepsy show substantial long-term stability over many years. However, rare relocation of these sites does occur in some patients over many years.

Oversight of Artificial Intelligence in Human Subjects Research (AI HSR) presents unique challenges. These challenges arise from both the non-linear and iterative nature of AI development, as well as from the way AI shifts risk from individual research subjects to larger populations affected by AI-driven decisions and data handling. Traditional Institutional Review Boards (IRBs) often struggle to keep pace with these changes, which can lead to gaps in risk assessment and delays in the review process. There is a growing need for transparent, repeatable methods to manage AI risk in healthcare. This paper introduces the Three-Stage Framework, a risk-based oversight model designed to align ethical and regulatory review with an AI project’s stage of maturity and potential human impact. By aligning the level and timing of IRB review with the types of risks present at each stage of AI system development, the framework supports appropriate regulatory pathways and documents expectations while maintaining effective protection of human subjects. Through gradual, stage-appropriate documentation, the approach supports responsible and adaptive innovation while preparing AI systems for safe and ethical use across biomedical, social, behavioral, and educational domains. This approach prepares AI systems for safe and ethical use, accelerates compliant research, and helps IRBs and institutions maintain trustworthiness while protecting human subjects.

OBJECTIVES: To evaluate and compare the ability of three popular open-source artificial intelligence platforms to diagnose common trauma-related fractures using radiologic imaging. METHODS: Design: Retrospective diagnostic performance comparison study. Setting: Publicly accessible online radiologic imaging databases. Patient Selection Criteria: Five common orthopedic trauma fractures were assessed: ankle, tibial plateau, intertrochanteric, femoral neck, and humerus. Radiographs and computed tomography (CT) images were collected. Images were randomly selected from confirmed diagnoses on Radiopaedia.org. Outcome Measures and Comparisons: ChatGPT 5, Grok 3, and Claude 4.5 Sonnet were queried with each image. Diagnostic accuracy, sensitivity, specificity, positive and negative predictive values, and performance by modality (X-ray vs. CT) were assessed. The reference standard was the expert-verified diagnosis provided by Radiopaedia.org, limited to cases labeled with a “diagnosis certain” tag. RESULTS: Each model was provided with 30 radiographs and 20 CT images whenever possible. ChatGPT 5, Grok 3, and Claude 4.5 Sonnet accurately diagnosed diseased images in 26.8%, 18.8%, and 22.4% of cases, respectively. By fracture type, ChatGPT 5 demonstrated the highest correct classification rates for ankle (10%), femoral neck (38%), humerus (40%), and tibial plateau (44%) fractures. Grok 3 demonstrated the highest correct classification rate for intertrochanteric fractures (6%). Overall sensitivities were 0.267, 0.187, and 0.223 for ChatGPT 5, Grok 3, and Claude 4.5 Sonnet, respectively. ChatGPT 5 and Grok 3 outperformed Claude 4.5 Sonnet (both P < 0.001). No modality-based performance differences were observed for any model. CONCLUSIONS: Among the publicly available large language models evaluated for radiologic interpretation of orthopedic trauma imaging, ChatGPT 5 demonstrated the highest overall diagnostic accuracy, followed by Claude 4.5 Sonnet and Grok 3. Despite relative variation between the models, overall diagnostic accuracy for fracture detection was low across all platforms (<27%). In their baseline forms, these publicly accessible large language models are not recommended for radiologic imaging interpretation. LEVEL OF EVIDENCE: Diagnostic Level III. See Instructions for Authors for a complete description of levels of evidence.