Citations

Abstract Background Integrating advanced artificial intelligence (AI) into clinical decision-support often requires the sharing of sensitive patient data with external services, raising privacy concerns. Homomorphic encryption (HE) allows computing directly on encrypted data, without revealing the underlying patient information. Objectives To develop a large language model (LLM)-assisted diagnosis framework while preserving patient privacy in the clinical text analysis, by leveraging HE and using rare disease (RD) diagnosis as a representative application. To demonstrate HE does not hinder the system performance. Materials and Methods Texts from patient histories and a RD knowledge base were embedded by LLMs into vectors, then encrypted using HE to obscure private information while retaining the semantic nuances. Diagnostic recommendations were generated by computing and ranking the similarities between the patient history and RD vectors in the encrypted space. The system was evaluated using 50 synthetic case reports (5 RDs, each with 10 reports). Results Applying HE did protect private information from reverse-embedding attacks. HE imposed little disruption to the diagnostic accuracy, with normalized discounted cumulative gains (nDCG) of 0.6108 ± 0.3412 (encrypted) versus 0.6083 ± 0.3415 (unencrypted). The accuracy and computational performance were tunable and consistent, as demonstrated across five different LLMs. Discussion Our privacy-preserving framework opens tremendous opportunities toward hosting and serving powerful AI solutions across institution boundaries, which would remove the need for local deidentification and incentivize users to access secure external decision-support services. Conclusions Integrating HE with LLM retrieval can promote the dissemination of nonredundant, high-capacity AI services by preserving both privacy and accuracy.

ABSTRACT Background Despite histologic remission (HR) being important in management of ulcerative colitis, the impact of endoscopic sampling on newer scoring systems remains uncertain. This study examines the outcome of applying PICaSSO Histologic Remission Index (PHRI) scoring on biopsies collected during routine clinical care and investigates the impact of endoscopic sampling variation on PHRI scoring and HR status. Methods A retrospective survey of UC endoscopic biopsies ( N = 128, 559 bites, 1–13 bites/biopsy) was evaluated using standard H&E stain and myeloperoxidase (MPO) immunohistochemistry to aid neutrophil identification. Convolutional neural networks (CNN), trained on colonic tissue architecture and cytomorphology features, were deployed alongside pathologist interpretations. In silico modeling of bite‐derived virtual biopsies ( N = 19,760) was investigated to determine the impact of sampling variation on PHRI scores and compared with ground truth clinical samples. Results PHRI MPO scoring displayed improved sensitivity and specificity for identifying neutrophils, compared with PHRI H&E . CNN‐derived PHRI scoring was most accurate for chronic active and chronic active with ulceration categories, with greatest variability observed in chronic inactive biopsies. Virtual patient biopsies comprised of < 5 endoscopic bites were impacted most by sampling permutations. Limited variability in PHRI scores was observed in samples consisting of ≥ 5 bites. Conclusions HR status (PHRI = 0) determination was slightly impacted by sampling. Biopsies consisting of < 5 endoscopic bites exhibited the most variability. Conservative sampling of 5–7 endoscopic bites may produce more accurate assessments of microscopic heterogeneity in clinical studies employing quantitative histologic measurements to assess disease activity and remission. Additionally, incorporating MPO increased the sensitivity and specificity of detecting active inflammation for PHRI scoring.

Abstract Background Intracranial mesenchymal tumors, FET::CREB fusion-positive (ICMT), show fusions involving FET RNA-binding protein family genes (EWSR1 or FUS) and CREB family of transcription factors (ATF1, CREB1, or CREM). The methylation signature(s), gene expression characteristics, and clinical behavior of this important tumor type require further characterization. Methods We study the methylation profiles of 81 ICMT cases (61 newly profiled cases and 20 cases from publicly available sources). Clinicopathologic and genomic data were recorded for each case when available. Results ICMT showed a relatively distinct methylation signature compared to related tumors. Among the 65 cases where fusion types were documented, the identified fusions included EWSR1::ATF1 (25 cases), EWSR1::CREB1 (12 cases), EWSR1::CREM (21 cases), FUS::CREM (3 cases), and SMARCA2::CREM (4 cases). We confirmed the prior description of 2 distinct subgroups of ICMT (subclasses A and B). The majority of the cases belonged to subclass A (n = 69; 85%), which showed a higher median age compared to subclass B patients (26 years vs. 15 years). Subclass B cases (n = 12; 15%) showed shorter progression-free survival (P &lt; 0.01). Gene expression analysis of ICMT showed key overexpressed markers in ICMT, with significant CREM overexpression regardless of fusion type, when compared to either meningioma alone or a larger group of CNS tumors. Conclusions This work provides further characterization of ICMT as an important CNS mesenchymal neoplasm that is prone to tumor recurrence, showing 2 prognostically relevant methylation subclasses and warranting diagnostic distinction from other epigenetically and histologically related tumors. ICMTs show substantial overexpression of the CREM gene, independent of fusion type.

Abstract Isolated/idiopathic rapid eye movement sleep behavior disorder (iRBD) is a usually prodromal manifestation of neurodegenerative disorders with α-synuclein pathology: Parkinson’s disease (PD), dementia with Lewy bodies (DLB), and multiple system atrophy (MSA). Clinical trials in the iRBD population face substantial barriers: limited access to well-characterized cohorts, inconsistent assessment protocols across centers, and the absence of validated biomarkers of disease burden. The North American Prodromal Synucleinopathy (NAPS) Consortium was established to address these challenges and facilitate clinical trials for neuroprotective therapies targeting synucleinopathy at the earliest known stages. In this multi-site, longitudinal, observational study, nine academic centers across North America will enroll and follow over 500 individuals with iRBD from existing sleep centers. Sixty control participants, matched for age, sex, and race will also be recruited. A harmonized protocol—including a standardized clinical battery assessing motor, cognitive, autonomic, psychiatric, sensory, and sleep function; structured diagnostic adjudication; biospecimen collection; and centralized analysis of both polysomnography and neuroimaging data—is outlined herein and reflects NAPS Stage 2. Each participants completes these assessments annually, and are replaced in the event of phenoconversion. By unifying assessments and expanding geographic reach, NAPS lays the groundwork for efficient, well-powered clinical trials designed to delay or prevent progression of iRBD to overt PD, DLB, or MSA—ultimately enabling earlier, more effective therapeutic intervention for neurodegenerative disease. Registered at clinicaltrials.gov (NCT05826457).

Metabolic dysfunction-associated steatohepatitis (MASH) is a global health care burden. Appropriate large animal models mimicking the main MASH characteristics of steatosis, inflammation, and hepatocyte damage alongside progression of fibrosis are imperative for robust preclinical studies, especially in the field of hepatobiliary surgery. The present study aimed to characterize a novel model of steatohepatitis in fumarylacetoacetate hydroxylase-deficient ( FAH -/- ) pigs. FAH -/- pigs were generated using a CRISPR/Cas9 system. The animals received a choline-deficient, L-amino acid-defined high-fat diet and subtherapeutic doses of nitisinone and were followed for 3 months. Liver biopsies were obtained at baseline and every month during treatment. Steatosis, inflammation, and fibrosis were assessed by conventional histological scoring systems and by an artificial intelligence (AI) model. Serum liver parameters were analyzed every 2 weeks. Steatosis increased significantly throughout the study, with severe steatosis observed as early as 2 months into treatment. The inflammation score was increased in all animals after 3 months of treatment, whereas the AI-based CD45+ cell count showed region-specific trends in the portal and lobular areas. Collagen content and the corresponding fibrosis stage showed an increase over the 3-month period; however, the difference was not significant. Serum liver parameters did not show any relevant elevations during the study. In summary, we successfully developed and characterized a novel model of steatohepatitis in the FAH -/- pig within 3 months. Further studies with prolonged observation time and/or cycling of nitisinone administration are needed to evaluate whether progressive fibrosis and cirrhosis could be achieved with this model.

Abstract Object Transdermal biosensors may provide an alternative to conventional blood-based biomarker measurement. Our purpose was to determine the binary correlation between transdermal (Infrasensor™; RCE, Inc, Carlsbad, CA) and conventional blood-based measurements. Methods This was a secondary analysis from a previously published observational cardiac troponin I (cTnI) study performed to establish the upper reference level of cTnI, at 10 US hospitals. After obtaining informed consent, 2 cohorts of patients were enrolled: 1) those who completed a health assessment questionnaire and appeared healthy, and 2) those with a known elevated cTnI per the local hospital standard assay. All blood lab analyses were performed at the University of Maryland Medical Center, Baltimore, MD. Normal was defined as cTnI <53.48 ng/L (male) or 34.11 ng/L (female) using the Siemens Atellica IM assay (Siemens Medical Solutions, Mountain View, CA), NT-proBNP <450 pg/mL (>75 years) or <124 pg/mL (<75 years), creatinine >1.17 mg/dL (male) or >0.95 mg/dL (female), and HbA1c <6.4%. The Infrasensor was placed on the patient’s wrist for measurement and blood drawn for analysis at approximately the same time. Results Of 840 enrolled patients, the median (IQR) age was 46 (30,57), 416 (49.5%) were female, 10.36% Hispanic, 6.7% Asian, 12.9% African American, and 69.1% White. Elevated lab tests were 102 hscTnI’s, 156 NTproBNP’s, 37 HbA1C’s, and 163 creatinine’s. Significant binary correlations were found between all transdermal signals and the corresponding lab blood levels Conclusion Infrasensor transcutaneous measurement demonstrates similar results as that obtained from blood testing in the central laboratory. Capsule The Infrasensor (RCE, Inc, Carlsbad, CA, USA) is rapid point of care transcutaneous biomarker measurement device. This study evaluated its ability to provide qualitative results for troponin I, NTproBNP, creatinine, and HbA1c levels in 840 patients. Significant correlations were found between all transdermal signals and the corresponding binary lab blood levels.

Abstract Objective High-frequency oscillations (HFOs) are highly specific biomarkers of epileptogenic tissue, yet their noninvasive localization remains challenging due to their brief duration, low amplitude, and poor signal-to-noise ratio. Here, we introduce TH-DeepSIF, a transformer-based deep learning framework trained on biologically realistic neural mass model simulations, to robustly perform HFO source imaging from scalp EEG. Methods TH-DeepSIF was evaluated in simulated single- and dual-source HFO scenarios, where it was tasked with recovering both the spatial location and temporal dynamics of HFO generators under increasing spatiotemporal complexity. We further validated TH-DeepSIF in 25 patients with drug-resistant epilepsy by comparing EEG source imaged HFO sources against surgical resection regions. Results TH-DeepSIF accurately recovered both the spatial location and temporal dynamics of simulated HFO generators, achieving low localization error and strong waveform correspondence with ground truth. TH-DeepSIF localization for pathological HFOs (pHFOs, or spike ripples) demonstrated strong agreement with surgical resection regions, achieving a median localization error of 11.9 mm and specificity of 0.896. Compared with all HFOs (aHFOs), pHFO-based source imaging showed significantly stronger spatial correspondence with resection regions, significantly smaller localization error, and higher precision, sensitivity, geometric mean, and F1 score. Significance These findings demonstrate that TH-DeepSIF provides a robust, data-driven framework for noninvasive HFO source imaging with improved anatomical specificity and enhanced clinical utility for presurgical evaluation using scalp EEG. Moreover, they show that pathological HFOs (spike ripples )—rather than general HFOs—serve as robust EEG biomarkers for accurate localization of the epileptogenic zone. Key Points A fully data-driven, parameter-free framework for noninvasive HFO source imaging using scalp EEG. Pathological HFO source maps exhibit strong spatial concordance with both the surgical resection region and the seizure onset zone. Source imaging based on pathological HFOs consistently outperforms imaging based on all detected HFOs.

Abstract Background Ventricular arrhythmias (VAs) are a proximate mechanism of sudden cardiac death, yet national patterns in place of death (POD) and their determinants remain sparsely described. We quantified 25-year trends and factors associated with POD among UAs decedents in the United States. Methods We analyzed CDC WONDER Multiple Cause of Death data (1999–2024) for adults ≥25 years with ventricular arrhythmias (ICD-10 I47.2, I49.0) as underlying cause. POD was categorized as inpatient, outpatient/emergency department (ED), home, hospice/nursing, or other/unknown. Covariates included age, sex, race, Hispanic origin, and urbanization. We calculated Annual and Average Annual Percent Changes (AAPCs and APC) using Age-Adjusted Mortality Rates (AAMRs), and fit multinomial logistic regression (reference = inpatient) to obtain adjusted odds ratios (ORs, 95% CIs). Results Among 433,988 ventricular arrhythmia (VA) deaths, POD was inpatient 62.0%, outpatient/ED 18.1%, home 11.1%, hospice/nursing 5.8%, other 3.1%. Inpatient deaths increased from 57.8% (1999) to 66.1% (2024). AAMRs declined sharply from 13.3 per 100,000 in 1999 to 6.3–6.5 during 2010–2019, then rose to 7.4 in 2021 and fell to 6.8 in 2024. In home vs inpatient: ≥85 years, medium/small metropolitan counties and rural counties had higher odds of VA deaths, whereas younger age groups, females, Black, American Indian, Asian/Pacific Islander individuals and Hispanic individuals had lower odds. In outpatient/ED vs inpatient: 25–44 years, 45–64 years, males, Black, American Indian and Asian/Pacific Islander individuals had higher odds, whereas ≥85 years and females had lower odds. In hospice/nursing facilities: ≥85 years, females, Whites, non-Hispanic individuals, medium/small metropolitan counties and rural counties had higher odds of VA deaths, whereas younger age groups, Black, American Indian, Asian/Pacific Islander individuals and Hispanic individuals had markedly lower odds. Conclusion From 1999–2024, VA deaths shifted toward hospitals. Persistent disparities by age, sex, race/ethnicity, and rurality highlight the need to expand equitable advance care planning and device deactivation discussions.

Objectives: Smoking is a key risk factor for pancreatitis, contributing to pathogenesis and disease progression. Data regarding second-hand smoking, or passive smoking (PS) exposure, are lacking. We therefore aimed to assess the impact of PS exposure across the pancreatitis spectrum. Methods: We analyzed baseline data from the PROCEED study, a multicenter study in the United States including patients with acute (AP), recurrent acute (RAP), and chronic (CP) pancreatitis. Participants detailed their individual smoking history and PS exposure, which was compared between AP, RAP, and CP subgroups. Participant factors and clinical characteristics were compared by level of individual and PS exposure. Results: Among 1369 participants (190 AP, 498 RAP, 681 CP), 346 (25.3%) were current and 409 (29.9%) were former smokers. The CP subgroup had the highest proportion of current smokers (36.7% CP vs 15.1% RAP, 11.1% AP, P < 0.001). PS exposure was also significantly higher in the CP subgroup (68.6% CP vs 58.2% RAP, 55.3% AP, P < 0.001). A minority (16.3%) with CP reported no smoking exposure (by self or passive) while nearly half smoked ≥20 pack-years. The mean PS duration was significantly higher in CP participants (21.8 y) compared with those with AP (17.8 y) or RAP (18.9 y) ( P < 0.0166). Conclusions: In this multicenter study, we affirmed an association with smoking use and intensity with CP-related complications. For the first time, we report a high prevalence of prior exposure to passive smoking in all pancreatitis subtypes, which requires further study to understand the impact on disease outcomes.