Biology subjects

Winstein, C.

Publications and source records attributed to Winstein, C..

4 recordsLinked to original sources

bioRxiv2026

CALM-VLM: CALIBRATION AND SELECTIVE PREDICTION IN VISION-LANGUAGE MODELS FOR RELIABLE BRAIN MRI CLASSIFICATION

AO_SCPLOWBSTRACTC_SCPLOWRecent advances in vision-language models (VLMs) have demonstrated strong multimodal capabilities for medical image analysis. However, their confidence in diagnostic predictions is often unclear, limiting adoption in clinical settings. We introduce CALM-VLM (CALibration Mechanism for Vision-Language Models), which integrates confidence calibration and selective prediction into a generative 3D VLM. To create CALM-VLM, we fine-tuned the Med3DVLM architecture for Alzheimers disease (AD) and stroke classification as initial test cases. To improve reliability, we incorporated temperature scaling on the VLMs generative outputs. The calibrated model then selectively abstained from predictions when uncertain; this also improved its diagnostic accuracy. Experiments across multi-site MRI datasets, from 10 countries worldwide, show that CALM-VLM improved confidence relative to uncalibrated VLMs. Coverage-adjusted test receiver-operator characteristic curve-area under the curve (ROC-AUC) increased by 5% to 13% for both diagnostic tasks across independent test sets. Our calibrated VLM achieved a test ROC-AUC of 0.951 for AD classification and 0.905 for stroke classification. These findings highlight the importance of calibrated, uncertainty-aware VLMs for trustworthy neuroimaging AI.

neuroscience↗

bioRxiv2019

Predictors of arm non-use in chronic stroke: a preliminary investigation

BackgroundThe phenomenon of non-use after stroke is characterized by failure to use the contralesional arm despite adequate capacity. It has been suggested that non-use is a consequence of the greater effort and/or attention required to use the affected limb, but such accounts have not been tested, and we have poor understanding of the characteristics of individuals who exhibit non-use.\n\nObjectiveWe aimed to provide preliminary evidence regarding the demographic, neuropsychological, and psychological factors that may influence non-use in chronic stroke.\n\nMethodsTwenty chronic stroke survivors (10 left and 10 right hemisphere stroke) with mild-to-moderate sensory-motor impairment on the upper extremity Fugl-Meyer (UEFM) were assessed with the Actual Amount of Use Test (AAUT), which measures the disparity between amount of use and quality of movement in \"spontaneous\" versus \"forced\" conditions. Participants were als assessed with measures of limb apraxia, spatial neglect, attention/arousal, and self-efficacy (confidence in arm movement). Using stepwise multiple regression, we determined whether demographic information and/or performance on these measures predicted AAUT non-use scores.\n\nResultsScores on the UEFM as well as attention/arousal and self-efficacy predicted the degree of non-use. Attention/arousal predicted non-use above and beyond UEFM.\n\nConclusionsGiven the complexity of the non-use phenomenon, it follows that a combination of impairment, attention/arousal, and perceived confidence predicted non-use behavior. That a measure of attention/arousal predicted non-use behavior above and beyond sensory-motor functioning highlights the importance of motivated engagement to drive use of the paretic limb. Larger-scale studies incorporating additional measures (e.g., mental health, lesion volume and white matter connectivity, pain, motivation) will be important for future investigations.

neuroscience↗

bioRxiv2018

Dosage Matters: A Randomized Controlled Trial of Rehabilitation Dose in the Chronic Phase after Stroke

Background and PurposeFor stroke rehabilitation, task-specific training in animal models and human rehabilitation trials is considered important to trigger inherent neuroplasticity, promote motor learning, and functional recovery. Little is known, however, about what constitutes an effective dosage of therapy.\n\nMethodsThis is a parallel group, four arm, single blind, phase I, randomized control trial of four dosages of upper extremity therapy delivered in an outpatient setting during the chronic phase after stroke. Participants were randomized into groups that varied in total dosage of therapy (i.e., 0, 15, 30, or 60 hours). Seven hundred and four participants were assessed for eligibility, 50 were eligible to enroll, 45 were randomized, 44 participated and 41 completed the study. Planned primary analyses used linear mixed effects regression to model baseline to post-intervention changes in the Motor Activity Log-Quality of Movement rating (MALQ) and the Wolf Motor Function Test (WMFT) time score as a function of therapy dosage. A series of hierarchical models were constructed using the MALQ and WMFT.\n\nResultsWe observed a significant dose response curve: the greater the dosage of training, the greater the change in MALQ, with the dose by week slope parameter of 0.0045 ({Delta}MAL/hour/week; p = 0.0011; 95% CI = [0.0019; 0.0071]). Over the 3 weeks of therapy, this corresponds to a gain of 0.81 in MALQ for the 60 hour dose.\n\nConclusionsFor mild-to-moderately impaired stroke survivors, the dosage of a patient-centered, task specific motor therapy was shown to systematically influence the gain in quality of arm use in the natural environment, but not functional capacity as measured in the laboratory. We highlight the importance of recovery outcomes that capture arm use vs. functional capacity.\n\nClinical Trial RegistrationURL: http://www.clinicaltrials.gov. Unique identifier: NCT 01749358

clinical trials↗

bioRxiv2017

The Anatomical Tracings of Lesions After Stroke (ATLAS) Dataset - Release 1.1

Stroke is the leading cause of adult disability worldwide, with up to two-thirds of individuals experiencing long-term disabilities. Large-scale neuroimaging studies have shown promise in identifying robust biomarkers (e.g., measures of brain structure) of long-term stroke recovery following rehabilitation. However, analyzing large rehabilitation-related datasets is problematic due to barriers in accurate stroke lesion segmentation. Manually-traced lesions are currently the gold standard for lesion segmentation on T1-weighted MRIs, but are labor intensive and require anatomical expertise. While algorithms have been developed to automate this process, the results often lack accuracy. Newer algorithms that employ machine-learning techniques are promising, yet these require large training datasets to optimize performance. Here we present ATLAS (Anatomical Tracings of Lesions After Stroke), an open-source dataset of 304 T1-weighted MRIs with manually segmented lesions and metadata. This large, diverse dataset can be used to train and test lesion segmentation algorithms and provides a standardized dataset for comparing the performance of different segmentation methods. We hope ATLAS release 1.1 will be a useful resource to assess and improve the accuracy of current lesion segmentation methods.

neuroscience↗