Publications

Continual learning (CL) enables animals to learn new tasks without erasing prior knowledge. CL in artificial neural networks (NNs) is challenging due to catastrophic forgetting, where new learning degrades performance on older tasks. While various techniques exist to mitigate forgetting, theoretical insights into when and why CL fails in NNs are lacking. Here, we present a statistical-mechanics theory of CL in deep, wide NNs, which characterizes the network's input-output mapping as it learns a sequence of tasks. It gives rise to order parameters (OPs) that capture how task relations and network architecture influence forgetting and anterograde interference, as verified by numerical evaluations. For networks with a shared readout for all tasks (single-head CL), the relevant-feature and rule similarity between tasks, respectively measured by two OPs, are sufficient to predict a wide range of CL behaviors on classic benchmark tasks. In addition, the theory predicts that increasing the network depth can effectively reduce interference between tasks, thereby lowering forgetting. For networks with task-specific readouts (multihead CL), the theory identifies a phase transition where CL performance shifts dramatically as tasks become less similar, as measured by another task-similarity OP. While forgetting is relatively mild compared to single-head CL across all tasks, sufficiently low similarity leads to catastrophic anterograde interference, where the network retains old tasks and interpolates new training data perfectly but completely fails to generalize new learning. Our results delineate important factors affecting CL performance and offer theoretical insights into common heuristics for mitigation of forgetting.

Aldehydes accumulating in response to reduced fatty acid oxidation in tumor-infiltrating lymphocytes damage mitochondria and drive T cell exhaustion.

BACKGROUND: Professionalism is a core competency in graduate medical education, yet research examining specialty-specific professionalism perceptions between trainees and faculty remains limited, particularly regarding the influence of role and institutional culture on these perceptions. This study examined how anesthesiology trainees and attendings perceive unprofessional behavior and whether these perceptions differ based on participant characteristics.

METHODS: A multi-site cross-sectional survey was conducted at five anesthesiology residency programs from February to March 2024. Participants rated degree of unprofessionalism on19 workplace vignettes depicting potentially unprofessional behaviors using a 7-point Likert scale. Vignettes were categorized into five themes: Verbal, Supervision, Quality, Time, and Engagement. Proportional odds models examined differences in ratings based on role (trainee vs. attending), adjusting for gender, race, underrepresented status, and institution.

RESULTS: Among 369 respondents (153 trainees, 216 attendings; 35.9% response rate), perceptions varied by scenario and participant characteristics. Six vignettes were more consistently rated as unprofessional (>80% unprofessional ratings), while four showed higher variability (<50% unprofessional ratings). Significant institutional differences were observed in five vignettes (Odds ratios [ORs] <0.14 or >3.7, p < 0.0001 to 0.027). Age influenced ratings of five vignettes (ORs = 0.75, 1.68, 1.63, 1.35 and 1.31 respectively, p <0.0001 to 0.027), while gender, race and underrepresented status showed no significant differences. After adjusting for demographics, trainees and attendings differed significantly in their ratings of 10 vignettes (p <0.0001 to 0.033). Attendings rated nine scenarios as more unprofessional than trainees (ORs ranging from 0.26 to 0.50), while trainees rated only one scenario as more unprofessional than attendings (OR = 2.01).

CONCLUSIONS: Perceptions of unprofessional behavior among anesthesiology professionals vary significantly by role and institution. These findings underscore the importance of context-sensitive approaches to professionalism education that acknowledge diverse perspectives and institutional cultures while maintaining core professional standards.

BACKGROUND AND OBJECTIVES: Cerebral adrenoleukodystrophy (CALD) is a common manifestation of adrenoleukodystrophy (ALD) in men. Early detection of CALD lesions through MRI screening is critical to allow for therapeutic action preventing severe disability and death. While the frequency of brain MRI monitoring has been addressed by international recommendations, no consensus currently exists regarding which MRI sequences should be used in a real-world setting for screening and follow-up of CALD lesions. The aim of this study was to establish guidelines for the MRI protocol in clinical practice and to identify priority sequences for research use, thereby promoting intercenter harmonization.

METHODS: A modified Delphi procedure was used to achieve consensus on MRI protocols for ALD screening, lesion monitoring, and research applications among experts with experience in brain imaging in ALD. Questionnaires allowed experts to indicate whether they considered sequences as core, optional, or research, or to express agreement (5-point scale ranging from completely disagree to completely agree) with specific statements. Topics where no agreement was reached were discussed during online consensus meetings.

RESULTS: Thirty experts from 9 countries participated and agreed that the core screening protocol for ALD in adults and children should include at least 3D T1-weighted, spin-echo T2-weighted, 3D fluid-attenuated inversion recovery, and diffusion-weighted imaging (DWI). Postcontrast T1-weighted imaging should be performed systematically in specific clinical scenarios. Experts supported using DWI alongside the Loes score and postcontrast imaging to assess lesion progression. A research protocol was defined, prioritizing diffusion tensor imaging, MR perfusion, and quantitative volumetric analyses.

DISCUSSION: This international project harmonizes the ALD MRI protocol, thus offering a practical framework to screen and monitor lesions, which will improve clinical decision making. It also identifies MRI sequences that should be prioritized in future research. Future research on MRI in ALD should focus on topics where no consensus has yet been reached in this project.

OBJECTIVE: This study aimed to identify the scope of VCE responsibilities as well as areas of improvement for the VCE role.

DESIGN: An 11-item free response/multiple-choice survey was distributed to prospective participants. Quantitative data was analyzed using descriptive statistics (mean and standard deviation) and qualitative methods were used to analyze free response data.

SETTING: Participants completed the survey electronically.

PARTICIPANTS: Current vice chairs of education from across the United States were recruited.

RESULTS: Twenty-five of 60 identified VCE (42% response rate) completed the survey. Responders held the VCE position for 4.7±3.2 years. The majority of respondents reported that they oversaw all educational activities in their department. Twenty-two respondents (88%) indicated that their job responsibilities were appropriate, while two (8%) felt their roles were not well-defined. Four VCE (16%) desired more control over the departmental budget for education-related activities, while two (8%) felt that their role was undermined by their Department Chairperson. Oversight and coordination of departmental education activities was the most frequently reported value for the VCE role. Eighteen respondents (72%) received compensation for their VCE role, but 7 (28%) did not.

CONCLUSIONS: The results of this survey provide insight into the responsibilities and perceived value of the VCE role in surgery. This survey also identified areas of concern which merit intradepartmental examination in order to improve the effectiveness of the VCE role at a local level.

OBJECTIVE: The objective of this study was to evaluate the impact of operator training level, specifically comparing Emergency Medicine (EM) attending physicians and residents, on the analgesic efficacy of ultrasound-guided nerve blocks (UGNBs) performed in the emergency department (ED).

METHODS: This is a secondary analysis of the National Ultrasound-Guided Nerve (NURVE) Block Registry, involving 11 U.S. EDs from January 1, 2022, to December 31, 2023. Adult patients undergoing UGNBs for acute pain or procedural analgesia were included, totaling 1595 procedures after exclusion of incomplete post-procedural pain scores. The primary outcome was percent pain reduction, with >50% defined as clinically meaningful and > 75% as substantial analgesia. Subgroup analyses were performed by operator experience and block type.

RESULTS: Attendings achieved clinically meaningful pain reduction in 80.7% of cases versus 63.4% for residents, and substantial reduction in 68.1% vs 47.7% respectively (p < 0.001). This difference persisted at the highest experience level (>20 prior blocks: 82.3% vs 71.0%, p = 0.0007) and was observed across block types, reaching significance for erector spinae plane blocks (79.6% vs 63.6%, p = 0.01). Complications were rare (0.13%), with both events in resident-performed blocks.

CONCLUSION: UGNBs performed by attendings were associated with greater analgesic success compared with those by residents, yet both groups achieved high rates of clinically meaningful pain reduction with very low complication rates. These results underscore the role of experience in UGNB efficacy while supporting the safety and effectiveness of supervised resident performance in the ED.

Diffusion MRI (dMRI) tractography is an advanced technique that uniquely enables in vivo mapping of brain fiber pathways. Traditional methods rely on tissue modeling to estimate fiber orientations for streamline propagation, which are computationally intensive and remain sensitive to noise and artifacts. Recent deep learning-based approaches enable data-driven fiber tracking by directly mapping dMRI signals to orientations, demonstrating both improved efficiency and accuracy. However, existing methods typically operate by either leveraging local signal information or learning global dependencies along streamlines. This paper presents DDTracking, a deep generative framework for tractography. One key innovation is the reformulation of streamline propagation as a conditional denoising diffusion process. To the best of our knowledge, this is the first work to apply diffusion models for fiber tracking. Our network architecture incorporates two new designs, including: (1) a dual-pathway encoding scheme that extracts complementary local spatial features and global temporal context, and (2) a conditional diffusion model module that integrates the spatiotemporal features to predict propagation orientations. All components are trained jointly in an end-to-end manner without any pretraining. In this way, DDTracking can capture fine-scale structural details at each point while ensuring long-range consistency across the entire streamline. We conduct a comprehensive evaluation across diverse datasets, including both synthetic and clinical data. Experiments demonstrate that DDTracking outperforms traditional model-based and state-of-the-art deep learning-based methods in terms of tracking accuracy and computational efficiency. Furthermore, our results highlight DDTracking's high generalizability across heterogeneous datasets, spanning varying health conditions, age groups, imaging protocols, and scanner types. Code is available at: https://github.com/yishengpoxiao/DDTracking.git.

BACKGROUND: Sleep disturbance, which is a common symptom in Long COVID, promotes a pro-inflammatory state and dysregulates lipid-derived specialized pro-resolving mediators (SPMs), presumably contributing to chronic unresolved inflammation. This study aimed to investigate the role of sleep disturbance in inflammatory resolution in Long COVID.

METHODS: We studied 39 participants (30F/9M, age range 22-68 years), including 31 individuals with Long COVID and 8 SARS-CoV-2-infected controls, who did not develop Long COVID. The study consisted of a 14-day at-home phase followed by a 1-day (24-h) in-laboratory stay. Sleep disturbance was assessed using PROMIS Sleep Disturbance T-scores. During the in-laboratory stay, a fasting morning blood sample was taken for assessment of lipid mediators. Data were analyzed using generalized linear mixed models.

RESULTS: Participants with Long COVID reported higher sleep disturbance than controls (p<.001). Pro-inflammatory lipid pathways were upregulated in Long COVID compared to control, as indicated by higher prostaglandin E2 (PGE2) levels (p<.05). Long COVID participants with high sleep disturbance (PROMIS Sleep Disturbance T-score ≥60) had lower SPM levels, including the precursor of D-series resolvins 17-hydroxydocosahexaenoic acid (17-HDHA), 17R/S-resolvin D1 (17R/S-RvD1), 15R-lipoxin B4 (15R-LXB4), and protectin D1n-3 DPA (PD1n-3 DPA) than those with low sleep disturbance (T-score <60) (p<.05).

CONCLUSIONS: This study suggests that sleep disturbance may contribute to chronic inflammation in Long COVID by compromising certain inflammatory resolution pathways. Promoting inflammatory resolution physiology in particular in those individuals with Long COVID suffering from sleep disturbance may serve as a mechanistic target to mitigate inflammation and symptom burden in Long COVID.

TRIAL REGISTRATION: ClinicalTrials.gov NCT05606211.

Midena et al.1 employ a nanoengineered 3D "nichoid" substrate that mechanically supports CD34+ hematopoietic stem and progenitor cells (HSPCs) during ex vivo manipulation, reducing culture-associated stress and improving engraftment and polyclonal output after gene editing or lentiviral gene addition. The work spotlights mechanobiology as a manufacturing lever for improving HSPC gene therapies.

BACKGROUND: Serum neurofilament light chain (sNfL) and glial fibrillary acidic protein (sGFAP) are promising biomarkers for Multiple Sclerosis (MS) disease activity. There is less known about their association with the symptomatic phenotypes such as depression and mental health outcomes.

OBJECTIVES: To investigate the association between sNfL and sGFAP and Patient-Reported Outcome (PRO) measures for depression and overall mental health in individuals with MS (iwMS).

METHODS: Participants completed the Center for Epidemiological Studies Depression Scale (CESD) and MS Quality of Life-54 (MSQOL-54)- at the time of the blood draw. Linear regression was used to estimate the association between the PRO measures as the outcome and the log-transformed biomarkers as the predictor. The association between baseline biomarkers and longitudinal change in PROs was estimated using linear mixed-effect models.

RESULTS: Cross-sectional analysis showed a significant correlation between sNFL and CES-D (p = 0.035) and MSQOL-54 Mental Health Composite (MHC) (p = 0.003) scores. This association remained statistically significant after adjusting for sex, age, EDSS and MS treatment. Neither cross-sectional nor longitudinal analysis of sGFAP levels showed significant correlation with PROs scores.

CONCLUSION: Serum NfL is associated with depression and overall mental health scores in iwMS. We did not find a significant relationship between sGFAP and PRO measures.