Erin W. Dickie
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Erin W. Dickie has a total of 1000 co-authors. Here are the top co-authors by number of shared publications:
- Aristotle N. Voineskos
Campbell Family Mental Health Research Institute, Centre for Addiction and Mental Health
16 shared publications Colin Hawco
Campbell Family Mental Health Research Institute, Centre for Addiction and Mental Health
9 shared publications- Tristan Glatard
CAMH
9 shared publications - Stephanie Hope Ameis
Campbell Family Mental Health Research Institute, Centre for Addiction and Mental Health
8 shared publications - Paul Thompson
Imaging Genetics Center, Mark and Mary Stevens Neuroimaging & Informatics Institute, Keck School of Medicine of the University of Southern California, Marina del Rey, California, USA
6 shared publications Anil K. Malhotra
Zucker Hillside Hospital, Glen Oaks, New York
5 shared publications
Latest Publications
- Open Access
- 27 October 2026
- Published by Center for Open Science in MetaArXiv
Neurobagel: building an international network for distributed data discovery
International data privacy regulations impede the pooling of research data for collaborative analysis. We introduce Neurobagel, a federated network enabling cohort discovery across locally governed, access-controlled datasets. Through intuitive graphical tools and a decentralized query infrastructure, Neurobagel facilitates harmonization, control, and discovery of data according to local regulations. Today, Neurobagel is deployed by consortia and data platforms in Europe, North America, Asia, and Australia, supporting diverse and evolving regulatory frameworks.
- 27 October 2026
- Published by Center for Open Science in MetaArXiv
Open AccessNeurobagel: building an international network for distributed data discovery
International data privacy regulations impede the pooling of research data for collaborative analysis. We introduce Neurobagel, a federated network enabling cohort discovery across locally governed, access-controlled datasets. Through intuitive graphical tools and a decentralized query infrastructure, Neurobagel facilitates harmonization, control, and discovery of data according to local regulations. Today, Neurobagel is deployed by consortia and data platforms in Europe, North America, Asia, and Australia, supporting diverse and evolving regulatory frameworks.
- Open Access
Evaluation of Elicit as an adjunct search, screening, and data extraction tool for systematic reviews
Background: Systematic reviews are critical for informing evidence-based medicine, policy, and future research. However, conducting a systematic review is labour- and cost-intensive and error-prone. Elicit is an AI tool that claims to replicate the systematic review process. Methods: Our study investigated Elicit’s ability to replicate a systematic review by comparing its results with those of an existing systematic review recently published by our research group. We calculated accuracy, precision, sensitivity, error rate, and person-time savings at each stage of the systematic review. Results: At search, Elicit had poor overlap, finding 158 of the 5285 studies. For abstract and full-text screening, Elicit had moderate sensitivity and accuracy but poor precision, with a skew toward false positives. Elicit’s data extraction tool was promising, with sentence-specific citations, and provided helpful information for 85% of values even when it did not find exact matches. Elicit offered person-time savings, taking 3.16 hours to complete the full systematic review process. Conclusions: Elicit shows promise as an adjunct screening and extraction AI tool, but in its current form, it is not ready as a standalone tool to replace the systematic review process.
Open AccessEvaluation of Elicit as an adjunct search, screening, and data extraction tool for systematic reviews
Background: Systematic reviews are critical for informing evidence-based medicine, policy, and future research. However, conducting a systematic review is labour- and cost-intensive and error-prone. Elicit is an AI tool that claims to replicate the systematic review process. Methods: Our study investigated Elicit’s ability to replicate a systematic review by comparing its results with those of an existing systematic review recently published by our research group. We calculated accuracy, precision, sensitivity, error rate, and person-time savings at each stage of the systematic review. Results: At search, Elicit had poor overlap, finding 158 of the 5285 studies. For abstract and full-text screening, Elicit had moderate sensitivity and accuracy but poor precision, with a skew toward false positives. Elicit’s data extraction tool was promising, with sentence-specific citations, and provided helpful information for 85% of values even when it did not find exact matches. Elicit offered person-time savings, taking 3.16 hours to complete the full systematic review process. Conclusions: Elicit shows promise as an adjunct screening and extraction AI tool, but in its current form, it is not ready as a standalone tool to replace the systematic review process.
- Open Access
- 21 September 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Data Work as Care Work: A Scoping Review of Education, Training, and Professional Development for Computational Scientists in Health (Preprint)
BACKGROUND Artificial intelligence (AI) is reshaping health care, yet educational efforts have focused largely on preparing clinicians to work with AI rather than on preparing the computational scientists who design these systems to work within health systems. Framing this data work as care work foregrounds its ethical and clinical consequences. OBJECTIVE To map the literature on education, training, and professional development for computational scientists in health, including the competencies taught, engagement with clinicians and patients, attention to ethics, and evaluation of outcomes. METHODS We followed the Joanna Briggs Institute (JBI) methodology for scoping reviews and reported according to PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews); the protocol was registered on the Open Science Framework.Ovid MEDLINE, APA PsycINFO, ERIC (Education Resources Information Center), Scopus, and IEEE Xplore were searched for published sources from 2015 to September 15, 2026. Eligibility was defined using the Population, Concept, and Context framework. After calibration on a dual-screened subset (Cohen κ=0.61), a single reviewer completed screening and charting in Covidence. Data were summarized using descriptive statistics and exploratory thematic analysis. RESULTS Of 7,242 records screened, 44 studies were included. Half were conducted in the United States (22/44, 50%), and informatics journals were the most common venue (15/44, 34%). Studies primarily described curricula (15/44, 34%), identified competencies (12/44, 27%), described experiential learning (10/44, 23%), or analyzed the workforce (4/44, 9%). Health domain competencies were described in 40 (91%) studies, and 35 (80%) characterized their programs as interdisciplinary or multidisciplinary. However, structured exchange with clinical knowledge holders was less common: clinical exposure (12/44, 27%), cross-faculty delivery (8/44, 18%), and patient involvement (5/44, 11%). Ethics was referenced in 30 (68%) studies, yet only 4 (9%) described a dedicated ethics course, and responsible AI accounted for 12 (5%) of 248 extracted competencies. Thirteen (30%) studies acknowledged algorithmic bias; 6 (14%) described educational activities to address it. Communication deficits were noted in 28 (64%) studies, but only 8 (18%) described corresponding curricular content. We identified 78 distinct job titles. Only 14 (32%) studies evaluated learning outcomes, predominantly through learner self-report; none assessed changes in practice. CONCLUSIONS Educational effort is asymmetric: clinicians are increasingly taught to understand AI, whereas the computational scientists who build it are seldom taught to understand health care. Ethics and equity are endorsed as competencies but rarely resourced as curriculum. Educational investment should be rebalanced toward the computational workforce, through a shared core competency framework - spanning health system literacy, communication and collaboration with clinicians and patients, responsible AI, and project and change management - that is differentiated by role, delivered through structured clinical and patient engagement, and evaluated beyond learner satisfaction. CLINICALTRIAL Registration: Open Science Framework; https://osf.io/8nyd9
- 21 September 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessData Work as Care Work: A Scoping Review of Education, Training, and Professional Development for Computational Scientists in Health (Preprint)
BACKGROUND Artificial intelligence (AI) is reshaping health care, yet educational efforts have focused largely on preparing clinicians to work with AI rather than on preparing the computational scientists who design these systems to work within health systems. Framing this data work as care work foregrounds its ethical and clinical consequences. OBJECTIVE To map the literature on education, training, and professional development for computational scientists in health, including the competencies taught, engagement with clinicians and patients, attention to ethics, and evaluation of outcomes. METHODS We followed the Joanna Briggs Institute (JBI) methodology for scoping reviews and reported according to PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews); the protocol was registered on the Open Science Framework.Ovid MEDLINE, APA PsycINFO, ERIC (Education Resources Information Center), Scopus, and IEEE Xplore were searched for published sources from 2015 to September 15, 2026. Eligibility was defined using the Population, Concept, and Context framework. After calibration on a dual-screened subset (Cohen κ=0.61), a single reviewer completed screening and charting in Covidence. Data were summarized using descriptive statistics and exploratory thematic analysis. RESULTS Of 7,242 records screened, 44 studies were included. Half were conducted in the United States (22/44, 50%), and informatics journals were the most common venue (15/44, 34%). Studies primarily described curricula (15/44, 34%), identified competencies (12/44, 27%), described experiential learning (10/44, 23%), or analyzed the workforce (4/44, 9%). Health domain competencies were described in 40 (91%) studies, and 35 (80%) characterized their programs as interdisciplinary or multidisciplinary. However, structured exchange with clinical knowledge holders was less common: clinical exposure (12/44, 27%), cross-faculty delivery (8/44, 18%), and patient involvement (5/44, 11%). Ethics was referenced in 30 (68%) studies, yet only 4 (9%) described a dedicated ethics course, and responsible AI accounted for 12 (5%) of 248 extracted competencies. Thirteen (30%) studies acknowledged algorithmic bias; 6 (14%) described educational activities to address it. Communication deficits were noted in 28 (64%) studies, but only 8 (18%) described corresponding curricular content. We identified 78 distinct job titles. Only 14 (32%) studies evaluated learning outcomes, predominantly through learner self-report; none assessed changes in practice. CONCLUSIONS Educational effort is asymmetric: clinicians are increasingly taught to understand AI, whereas the computational scientists who build it are seldom taught to understand health care. Ethics and equity are endorsed as competencies but rarely resourced as curriculum. Educational investment should be rebalanced toward the computational workforce, through a shared core competency framework - spanning health system literacy, communication and collaboration with clinicians and patients, responsible AI, and project and change management - that is differentiated by role, delivered through structured clinical and patient engagement, and evaluated beyond learner satisfaction. CLINICALTRIAL Registration: Open Science Framework; https://osf.io/8nyd9
- Open Access
- 29 August 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Brain-cognition relationships and treatment outcome in treatment-resistant late-life depression
Older adults with treatment-resistant depression are at significant risk for cognitive impairment. However, the relationship between treatment response and cognitive function in this population is not well-established. We examined neural correlates of executive and memory function, and their relationship with remission to pharmacological treatment in a biomarker study embedded within a clinical trial for late-life treatment-resistant depression. Among participants who completed neuroimaging, better cognition was associated with lower connectivity between components of the default mode and the frontoparietal networks and within the frontoparietal network. Using diffusion imaging data, lower tract integrity in a distributed set of tracts was associated with poorer processing speed. Additionally, gray matter structure was positively associated with cognition. Greater education was also associated with better cognition. Ongoing treatment resistance was predicted by poorer cognition and gray matter structure in cross-validated regularized logistic models. These results identify distinct cross-sectional associations between neural circuits and cognitive function in treatment-resistant late-life depression, and build a predictive model for antidepressant medication response.
- 29 August 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessBrain-cognition relationships and treatment outcome in treatment-resistant late-life depression
Older adults with treatment-resistant depression are at significant risk for cognitive impairment. However, the relationship between treatment response and cognitive function in this population is not well-established. We examined neural correlates of executive and memory function, and their relationship with remission to pharmacological treatment in a biomarker study embedded within a clinical trial for late-life treatment-resistant depression. Among participants who completed neuroimaging, better cognition was associated with lower connectivity between components of the default mode and the frontoparietal networks and within the frontoparietal network. Using diffusion imaging data, lower tract integrity in a distributed set of tracts was associated with poorer processing speed. Additionally, gray matter structure was positively associated with cognition. Greater education was also associated with better cognition. Ongoing treatment resistance was predicted by poorer cognition and gray matter structure in cross-validated regularized logistic models. These results identify distinct cross-sectional associations between neural circuits and cognitive function in treatment-resistant late-life depression, and build a predictive model for antidepressant medication response.
- Open Access
Untrained Convolutional Neural Networks as Feature Extractors for Structural MRI
We demonstrate that features extracted from structural MRI using un-CNN, an untrained convolutional neural network, achieve predictive performance comparable to or exceeding that of state-of-the-art pretrained foundation models across three structural MRI datasets and three downstream tasks. Un-CNN extends a classical 3D CNN architecture with multi-channel inputs, a hierarchical encoder with multi-scale feature aggregation, and covariance pooling. Untrained CNNs circumvent several key limitations of trained models, including high computational cost and memory requirements, the need to distribute large model weights, risks of data leakage, and challenges in reproducibility.
Open AccessUntrained Convolutional Neural Networks as Feature Extractors for Structural MRI
We demonstrate that features extracted from structural MRI using un-CNN, an untrained convolutional neural network, achieve predictive performance comparable to or exceeding that of state-of-the-art pretrained foundation models across three structural MRI datasets and three downstream tasks. Un-CNN extends a classical 3D CNN architecture with multi-channel inputs, a hierarchical encoder with multi-scale feature aggregation, and covariance pooling. Untrained CNNs circumvent several key limitations of trained models, including high computational cost and memory requirements, the need to distribute large model weights, risks of data leakage, and challenges in reproducibility.
- Open Access
Nipoppy: A framework for standardizing neuroimaging studies to facilitate international derived-data sharing
Neuroimaging data management and processing are tedious and error-prone, prompting reproducibility concerns. Globally, studies with heterogeneous infrastructure and governance policies lead to eclectic data processing and sharing, necessitating standardization of data workflows to ensure reusability and comparability of multi-centric datasets. The Nipoppy neuroinformatics framework facilitates such standardization by combining specification, protocol, and software to manage study-level data workflows. With its adoption, researchers can share standardized, derived datasets enabling efficient, reproducible, and inclusive research.
Open AccessNipoppy: A framework for standardizing neuroimaging studies to facilitate international derived-data sharing
Neuroimaging data management and processing are tedious and error-prone, prompting reproducibility concerns. Globally, studies with heterogeneous infrastructure and governance policies lead to eclectic data processing and sharing, necessitating standardization of data workflows to ensure reusability and comparability of multi-centric datasets. The Nipoppy neuroinformatics framework facilitates such standardization by combining specification, protocol, and software to manage study-level data workflows. With its adoption, researchers can share standardized, derived datasets enabling efficient, reproducible, and inclusive research.
- Open Access
A Comparative Evaluation of Structural MRI Foundation Models for Age, Sex, and Body-Mass Index Predictions
Foundation models pre-trained on large neuroimaging datasets offer a promising approach to overcome the limited sample sizes typical of clinical imaging studies, yet their generalization across diverse populations remains unclear. We present the first systematic benchmark of four publicly available structural MRI foundation models: AnatCL, BrainIAC, 3D-Neuro-SimCLR, and SwinBrain. Using T1-weighted MRIs from the Parkinson's Progression Markers Initiative (PPMI), Healthy Brain Network (HBN), and Nathan Kline Institute (NKI) datasets, we evaluate these models on sex classification, brain age prediction, and body mass index prediction, comparing against models trained from FreeSurfer-derived cortical thickness and cortical surface area features. Submitted models are evaluated using a standardized frozen feature probing framework across downstream tasks (sex classification, age prediction, and BMI prediction). The evaluation methods are available in BrainFMBench, a living benchmark for structural brain MRI foundation models hosted on GitHub, where new models can be added through pull requests. Although some foundation models outperformed FreeSurfer on particular tasks and datasets, 3D-Neuro-SimCLR and AnatCL outperformed the baselines overall, with 3D-Neuro-SimCLR demonstrating the most consistent performance (with the notable exception of HBN sex classification). The remaining models did not consistently outperform the baselines, indicating that the advantage of learned representations over morphometric features did not generalize reliably for these models. In addition, cross-model feature correlation analysis reveals that foundation model representations correlate differently with traditional cortical measurements. These findings position structural MRI foundation models, particularly 3D-Neuro-SimCLR and AnatCL, as promising avenues to boost the performance of predictive models in neuroimaging.
Open AccessA Comparative Evaluation of Structural MRI Foundation Models for Age, Sex, and Body-Mass Index Predictions
Foundation models pre-trained on large neuroimaging datasets offer a promising approach to overcome the limited sample sizes typical of clinical imaging studies, yet their generalization across diverse populations remains unclear. We present the first systematic benchmark of four publicly available structural MRI foundation models: AnatCL, BrainIAC, 3D-Neuro-SimCLR, and SwinBrain. Using T1-weighted MRIs from the Parkinson's Progression Markers Initiative (PPMI), Healthy Brain Network (HBN), and Nathan Kline Institute (NKI) datasets, we evaluate these models on sex classification, brain age prediction, and body mass index prediction, comparing against models trained from FreeSurfer-derived cortical thickness and cortical surface area features. Submitted models are evaluated using a standardized frozen feature probing framework across downstream tasks (sex classification, age prediction, and BMI prediction). The evaluation methods are available in BrainFMBench, a living benchmark for structural brain MRI foundation models hosted on GitHub, where new models can be added through pull requests. Although some foundation models outperformed FreeSurfer on particular tasks and datasets, 3D-Neuro-SimCLR and AnatCL outperformed the baselines overall, with 3D-Neuro-SimCLR demonstrating the most consistent performance (with the notable exception of HBN sex classification). The remaining models did not consistently outperform the baselines, indicating that the advantage of learned representations over morphometric features did not generalize reliably for these models. In addition, cross-model feature correlation analysis reveals that foundation model representations correlate differently with traditional cortical measurements. These findings position structural MRI foundation models, particularly 3D-Neuro-SimCLR and AnatCL, as promising avenues to boost the performance of predictive models in neuroimaging.
- Open Access
- 11 May 2026
- Published by Center for Open Science in MetaArXiv
Neurobagel: building an international network for distributed data discovery
International data privacy regulations impede the pooling of research data for collaborative analysis. We introduce Neurobagel, a federated network enabling cohort discovery across locally governed, access-controlled datasets. Through intuitive graphical tools and a decentralized query infrastructure, Neurobagel facilitates harmonization, control, and discovery of data according to local regulations. Today, Neurobagel is deployed by consortia and data platforms in Europe, North America, Asia, and Australia, supporting diverse and evolving regulatory frameworks.
- 11 May 2026
- Published by Center for Open Science in MetaArXiv
Open AccessNeurobagel: building an international network for distributed data discovery
International data privacy regulations impede the pooling of research data for collaborative analysis. We introduce Neurobagel, a federated network enabling cohort discovery across locally governed, access-controlled datasets. Through intuitive graphical tools and a decentralized query infrastructure, Neurobagel facilitates harmonization, control, and discovery of data according to local regulations. Today, Neurobagel is deployed by consortia and data platforms in Europe, North America, Asia, and Australia, supporting diverse and evolving regulatory frameworks.
- Open Access
- 8 May 2026
- Published by Public Library of Science (PLoS) in PLOS Mental Health
- Vol. 3 (5) , e0000605
An evaluation of the clinician-facing research dashboards from the Toronto Adolescent and Youth (TAY) Cohort Study in mental health care
Dashboards that integrate patient research data for clinical use can streamline information sharing and support clinical care, yet their application in mental health care is underexplored. This study evaluated the clinician-facing dashboards developed from the Toronto Adolescent & Youth (TAY) Cohort Study at the Centre for Addiction and Mental Health (CAMH), guided by the Reach, Effectiveness, Adoption, Implementation, and Maintenance (RE-AIM) framework. Dashboard reach was assessed through quantitative analyses, while semi-structured interviews with seven clinicians were conducted to explore perceptions of effectiveness, adoption, implementation, and maintenance. Quantitative analyses demonstrate that dashboards are completed for 69% of participants. Following the initial implementation phase, the average time from participant consent to dashboard completion decreases to six months. Qualitative findings suggest that dashboards can serve as a supplementary information source that may aid in supporting clinical decision-making and the integration of patient-reported research data into care. Clinicians also identified areas for improvement, including delays in dashboard completion, difficulty locating dashboards, and inefficient dashboard completion notifications. Clinicians suggested that addressing these areas through improved accessibility, timely data availability, and aligned communication strategies can increase uptake and sustainability. These findings emphasize the value of engaging end users and conducting ongoing evaluation to optimize the integration of research data into clinical workflows using dashboards to enhance mental health care. Future research should further examine factors influencing dashboard use across mental health care settings.
- 8 May 2026
- Published by Public Library of Science (PLoS) in PLOS Mental Health
- Vol. 3 (5) , e0000605
Open AccessAn evaluation of the clinician-facing research dashboards from the Toronto Adolescent and Youth (TAY) Cohort Study in mental health care
Dashboards that integrate patient research data for clinical use can streamline information sharing and support clinical care, yet their application in mental health care is underexplored. This study evaluated the clinician-facing dashboards developed from the Toronto Adolescent & Youth (TAY) Cohort Study at the Centre for Addiction and Mental Health (CAMH), guided by the Reach, Effectiveness, Adoption, Implementation, and Maintenance (RE-AIM) framework. Dashboard reach was assessed through quantitative analyses, while semi-structured interviews with seven clinicians were conducted to explore perceptions of effectiveness, adoption, implementation, and maintenance. Quantitative analyses demonstrate that dashboards are completed for 69% of participants. Following the initial implementation phase, the average time from participant consent to dashboard completion decreases to six months. Qualitative findings suggest that dashboards can serve as a supplementary information source that may aid in supporting clinical decision-making and the integration of patient-reported research data into care. Clinicians also identified areas for improvement, including delays in dashboard completion, difficulty locating dashboards, and inefficient dashboard completion notifications. Clinicians suggested that addressing these areas through improved accessibility, timely data availability, and aligned communication strategies can increase uptake and sustainability. These findings emphasize the value of engaging end users and conducting ongoing evaluation to optimize the integration of research data into clinical workflows using dashboards to enhance mental health care. Future research should further examine factors influencing dashboard use across mental health care settings.
- Open Access
Individualized cortical gradient and network topology reveal symptom-linked disruptions and neurobiological subtypes in schizophrenia
Schizophrenia is often conceptualized as a brain network disorder, yet the organizational principles and heterogeneity underlying widespread cortical abnormalities remain poorly understood. Leveraging multisite MRI data from 3,958 individuals diagnosed with schizophrenia and 5,489 neurotypical individuals, we studied the cortical organization and its subtyping by analyzing individualized cortical network similarity. We used eigenvector decompositions to study spatial patterning of the gradients and graph theory to study small–world topology. Individuals with schizophrenia showed widespread alterations of gradient loadings, which followed inferior–superior and frontal–temporal axes. Alterations in small–world topology were localized in key network hubs, including the insula and anterior cingulate cortex. Brain–symptom association analyses identified a latent dimension linking disorganization symptoms to topological alterations. Finally, clustering cortical alterations identified two robust subtypes, characterized by divergent anterior cingulate (S1) versus temporoparietal (S2) thickness differences aligned with the intrinsic gradient-topology patterns. Both subtypes were present early in the illness and stable across disease stages and age groups. These findings reveal systematic disruptions of cortical organization in schizophrenia, providing a network-level framework for macroscale brain organization and inter-individual heterogeneity.
Open AccessIndividualized cortical gradient and network topology reveal symptom-linked disruptions and neurobiological subtypes in schizophrenia
Schizophrenia is often conceptualized as a brain network disorder, yet the organizational principles and heterogeneity underlying widespread cortical abnormalities remain poorly understood. Leveraging multisite MRI data from 3,958 individuals diagnosed with schizophrenia and 5,489 neurotypical individuals, we studied the cortical organization and its subtyping by analyzing individualized cortical network similarity. We used eigenvector decompositions to study spatial patterning of the gradients and graph theory to study small–world topology. Individuals with schizophrenia showed widespread alterations of gradient loadings, which followed inferior–superior and frontal–temporal axes. Alterations in small–world topology were localized in key network hubs, including the insula and anterior cingulate cortex. Brain–symptom association analyses identified a latent dimension linking disorganization symptoms to topological alterations. Finally, clustering cortical alterations identified two robust subtypes, characterized by divergent anterior cingulate (S1) versus temporoparietal (S2) thickness differences aligned with the intrinsic gradient-topology patterns. Both subtypes were present early in the illness and stable across disease stages and age groups. These findings reveal systematic disruptions of cortical organization in schizophrenia, providing a network-level framework for macroscale brain organization and inter-individual heterogeneity.