Jason Vuong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jason Vuong has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Thanh G. Phan
Monash University
4 shared publications - Tharani Thirugnanachandran
Stroke and Ageing Research (STAR), Department of Medicine, School of Clinical Sciences at Monash Health, Melbourne, Victoria, Australia (T.T., H.M., T.P.).
3 shared publications - Angelos Sharobeam
The Royal Melbourne Hospital, Parkville, Victoria, Australia
2 shared publications Henry Ma
2 shared publications- Henry Ma
Centre for Neurological Sciences, Monash University, Melbourne, Australia
2 shared publications - Michael Valente
Department of Neurology, Monash Health, Clayton, Victoria, Australia
2 shared publications
Latest Publications
- Open Access
- 12 August 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Diagnostic Accuracy of a locally deployed Large Language Model Algorithm for Automated Code Stroke Pathway Identification in Emergency Department Triage Notes: Algorithm Development and Validation (Preprint)
BACKGROUND Delayed Code Stroke activation contributes to worse outcomes in acute stroke. Emergency Department (ED) triage notes contain free-text clinical information that could enable automated, real-time pathway activation. We evaluated the diagnostic accuracy of a multi-pass large language model (LLM) pipeline for identifying patients meeting Code Stroke criteria from ED triage notes. OBJECTIVE The aim of this study was to evaluate the diagnostic accuracy of a multi-pass large language model (LLM) pipeline for identifying patients meeting institutional Code Stroke criteria from ED triage notes and compare a single prompt to chain prompt performance. METHODS A retrospective cross-sectional study was conducted at Monash Medical Centre, Melbourne, Australia. De-identified triage notes from 3,023 ED presentations over a one-month period (September–October 2023) were analyzed. The pipeline applied sequential passes for translation, stroke symptom identification, mimic exclusion, baseline functional status, temporal window classification, and symptom resolution. Six locally deployed language models were evaluated. Performance was assessed against two reference standards: neurologist-labeled diagnosis and documented ED Code Stroke activation. Primary outcomes were sensitivity and specificity; secondary outcomes included PPV, NPV, and Gwet's AC1. Reliability of the neurologist reference standard was assessed by blinded independent re-review of a stratified random sample of 200 presentations by a second neurologist. RESULTS Of 3,023 presentations, 136 were neurologist-labeled positive. Agreement between the primary and a blinded second neurologist on a 200-note reliability sub-sample was almost perfect (raw agreement 95.0%, Cohen's κ 0.900, 95% CI 0.838–0.959). The cohort included 140 ED Code Stroke activations (median age 69, IQR 56–81 years), of whom 83 (59.2%) had confirmed stroke diagnosis. Sixteen patients (11.4%) underwent endovascular clot retrieval and 4 (2.9%) received thrombolysis. The best-performing model (Qwen 2.5 14B) achieved sensitivity 0.890 (95% CI 0.826–0.932), specificity 0.993 (0.989–0.996), PPV 0.858 (0.791–0.906), and NPV 0.995 (0.991–0.997). Pairwise McNemar testing demonstrated statistically superior overall accuracy for Qwen 2.5 14B over Llama 3.1 8B, Phi-4 14B, and Mistral 3 14B (all p<0.001 after Holm correction), with no significant difference detected versus Nemotron-Nano-12B-v2 or Qwen 3 14B. CONCLUSIONS A locally deployed language model demonstrates acceptable sensitivity and specificity for automated Code Stroke identification from free-text triage notes. Performance was comparable across the two best models, suggesting that capable open-weight models in this parameter range may be sufficient to proceed with ongoing internal testing and external validation. The pipeline operates without internet connectivity or model retraining on patient data, supporting feasibility for real-world ED integration. CLINICALTRIAL ACTRN12626000596303
- 12 August 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessDiagnostic Accuracy of a locally deployed Large Language Model Algorithm for Automated Code Stroke Pathway Identification in Emergency Department Triage Notes: Algorithm Development and Validation (Preprint)
BACKGROUND Delayed Code Stroke activation contributes to worse outcomes in acute stroke. Emergency Department (ED) triage notes contain free-text clinical information that could enable automated, real-time pathway activation. We evaluated the diagnostic accuracy of a multi-pass large language model (LLM) pipeline for identifying patients meeting Code Stroke criteria from ED triage notes. OBJECTIVE The aim of this study was to evaluate the diagnostic accuracy of a multi-pass large language model (LLM) pipeline for identifying patients meeting institutional Code Stroke criteria from ED triage notes and compare a single prompt to chain prompt performance. METHODS A retrospective cross-sectional study was conducted at Monash Medical Centre, Melbourne, Australia. De-identified triage notes from 3,023 ED presentations over a one-month period (September–October 2023) were analyzed. The pipeline applied sequential passes for translation, stroke symptom identification, mimic exclusion, baseline functional status, temporal window classification, and symptom resolution. Six locally deployed language models were evaluated. Performance was assessed against two reference standards: neurologist-labeled diagnosis and documented ED Code Stroke activation. Primary outcomes were sensitivity and specificity; secondary outcomes included PPV, NPV, and Gwet's AC1. Reliability of the neurologist reference standard was assessed by blinded independent re-review of a stratified random sample of 200 presentations by a second neurologist. RESULTS Of 3,023 presentations, 136 were neurologist-labeled positive. Agreement between the primary and a blinded second neurologist on a 200-note reliability sub-sample was almost perfect (raw agreement 95.0%, Cohen's κ 0.900, 95% CI 0.838–0.959). The cohort included 140 ED Code Stroke activations (median age 69, IQR 56–81 years), of whom 83 (59.2%) had confirmed stroke diagnosis. Sixteen patients (11.4%) underwent endovascular clot retrieval and 4 (2.9%) received thrombolysis. The best-performing model (Qwen 2.5 14B) achieved sensitivity 0.890 (95% CI 0.826–0.932), specificity 0.993 (0.989–0.996), PPV 0.858 (0.791–0.906), and NPV 0.995 (0.991–0.997). Pairwise McNemar testing demonstrated statistically superior overall accuracy for Qwen 2.5 14B over Llama 3.1 8B, Phi-4 14B, and Mistral 3 14B (all p<0.001 after Holm correction), with no significant difference detected versus Nemotron-Nano-12B-v2 or Qwen 3 14B. CONCLUSIONS A locally deployed language model demonstrates acceptable sensitivity and specificity for automated Code Stroke identification from free-text triage notes. Performance was comparable across the two best models, suggesting that capable open-weight models in this parameter range may be sufficient to proceed with ongoing internal testing and external validation. The pipeline operates without internet connectivity or model retraining on patient data, supporting feasibility for real-world ED integration. CLINICALTRIAL ACTRN12626000596303
- Open Access
Diagnostic Accuracy of a locally deployed Large Language Model Algorithm for Automated Code Stroke Pathway Identification in Emergency Department Triage Notes
Background: Delayed Code Stroke activation contributes to worse outcomes in acute stroke. Emergency Department (ED) triage notes contain free-text clinical information that could enable automated, real-time pathway activation. We evaluated the diagnostic accuracy of a multi-pass large language model (LLM) pipeline for identifying patients meeting Code Stroke criteria from ED triage notes. Methods: A retrospective cross-sectional study was conducted at Monash Medical Centre, Melbourne, Australia. De-identified triage notes from 3,023 ED presentations over a one-month period (September–October 2023) were analysed. The pipeline applied sequential passes for translation, stroke symptom identification, mimic exclusion, baseline functional status, temporal window classification, and symptom resolution. Six locally deployed language models were evaluated. Performance was assessed against two reference standards: neurologist-labelled diagnosis and documented ED Code Stroke activation. Primary outcomes were sensitivity and specificity; secondary outcomes included PPV, NPV, and Gwet’s AC1. Reliability of the neurologist reference standard was assessed by blinded independent re-review of a stratified random sample of 200 presentations by a second neurologist. Results: Of 3,023 presentations, 136 were neurologist-labelled positive. Agreement between the primary and a blinded second neurologist on a 200-note reliability sub-sample was almost perfect (raw agreement 95.0%, Cohen’s κ 0.900, 95% CI 0.838–0.959). The cohort included 140 ED Code Stroke activations (median age 69, IQR 56–81 years), of whom 83 (59.2%) had confirmed stroke diagnosis. Sixteen patients (11.4%) underwent endovascular clot retrieval and 4 (2.9%) received thrombolysis. The best-performing model (Qwen 2.5 14B) achieved sensitivity 0.890 (95% CI 0.826–0.932), specificity 0.993 (0.989–0.996), PPV 0.858 (0.791–0.906), and NPV 0.995 (0.991–0.997). Pairwise McNemar testing demonstrated statistically superior overall accuracy for Qwen 2.5 14B over Llama 3.1 8B, Phi-4 14B, and Mistral 3 14B (all p<0.001 after Holm correction), with no significant difference detected versus Nemotron-Nano-12B-v2 or Qwen 3 14B. Conclusions: A locally deployed language model demonstrates acceptable sensitivity and specificity for automated Code Stroke identification from free-text triage notes. Performance was comparable across the two best models, suggesting that capable open-weight models in this parameter range may be sufficient to proceed with ongoing internal testing and external validation. The pipeline operates without internet connectivity or model retraining on patient data, supporting feasibility for real-world ED integration.
Open AccessDiagnostic Accuracy of a locally deployed Large Language Model Algorithm for Automated Code Stroke Pathway Identification in Emergency Department Triage Notes
Background: Delayed Code Stroke activation contributes to worse outcomes in acute stroke. Emergency Department (ED) triage notes contain free-text clinical information that could enable automated, real-time pathway activation. We evaluated the diagnostic accuracy of a multi-pass large language model (LLM) pipeline for identifying patients meeting Code Stroke criteria from ED triage notes. Methods: A retrospective cross-sectional study was conducted at Monash Medical Centre, Melbourne, Australia. De-identified triage notes from 3,023 ED presentations over a one-month period (September–October 2023) were analysed. The pipeline applied sequential passes for translation, stroke symptom identification, mimic exclusion, baseline functional status, temporal window classification, and symptom resolution. Six locally deployed language models were evaluated. Performance was assessed against two reference standards: neurologist-labelled diagnosis and documented ED Code Stroke activation. Primary outcomes were sensitivity and specificity; secondary outcomes included PPV, NPV, and Gwet’s AC1. Reliability of the neurologist reference standard was assessed by blinded independent re-review of a stratified random sample of 200 presentations by a second neurologist. Results: Of 3,023 presentations, 136 were neurologist-labelled positive. Agreement between the primary and a blinded second neurologist on a 200-note reliability sub-sample was almost perfect (raw agreement 95.0%, Cohen’s κ 0.900, 95% CI 0.838–0.959). The cohort included 140 ED Code Stroke activations (median age 69, IQR 56–81 years), of whom 83 (59.2%) had confirmed stroke diagnosis. Sixteen patients (11.4%) underwent endovascular clot retrieval and 4 (2.9%) received thrombolysis. The best-performing model (Qwen 2.5 14B) achieved sensitivity 0.890 (95% CI 0.826–0.932), specificity 0.993 (0.989–0.996), PPV 0.858 (0.791–0.906), and NPV 0.995 (0.991–0.997). Pairwise McNemar testing demonstrated statistically superior overall accuracy for Qwen 2.5 14B over Llama 3.1 8B, Phi-4 14B, and Mistral 3 14B (all p<0.001 after Holm correction), with no significant difference detected versus Nemotron-Nano-12B-v2 or Qwen 3 14B. Conclusions: A locally deployed language model demonstrates acceptable sensitivity and specificity for automated Code Stroke identification from free-text triage notes. Performance was comparable across the two best models, suggesting that capable open-weight models in this parameter range may be sufficient to proceed with ongoing internal testing and external validation. The pipeline operates without internet connectivity or model retraining on patient data, supporting feasibility for real-world ED integration.
- Open Access
- 18 July 2025
- Published by S. Karger AG in Cerebrovascular Diseases
- Vol. 55 (3) , 311-317
The Anatomy of Infarcts Causing Hemianopia and Quadrantanopia in Posterior Cerebral Artery Stroke
Introduction: Following posterior cerebral artery stroke, infarcts causing quadrantanopia are thought to be smaller than those causing homonymous hemianopia. We investigated whether these two presentations were different due to the varying involvement of the striate, extrastriate cortex, or geniculocalcarine tract. Methods: Patients with unilateral posterior cerebral artery infarcts on magnetic resonance imaging and visual field defects, as identified by automated visual field perimetry, were included. Infarcts were manually segmented and registered to a standard brain template to facilitate comparison. Infarct volume and infarct involvement with geniculocalcarine fiber tracts were calculated in patients with hemianopia and quadrantanopia. Results: There were twenty-two patients: 15 patients with homonymous hemianopia (median age 68 [interquartile range, 55–76 years old]) and 7 with superior quadrantanopia (median age 40 [interquartile range, 30–56 years old]). Infarct volume significantly differed between the two groups: hemianopia 34.6 mL (interquartile range, 21.6–56.3 mL) versus superior quadrantanopia 15.5 mL (interquartile range, 7.2–24.1 mL), p = 0.026. There was significantly greater involvement of infarct with the geniculocalcarine tracts in the hemianopia group, 5.7 mL (interquartile range, 2.3–8.2 mL) than the superior quadrantanopia group, 2.0 mL (interquartile range, 1.2–2.7 mL), p = 0.042. Infarct involvement with cuneus and the calcarine cortex also significantly differed between the two groups (p ≤ 0.01). Conclusion: In posterior cerebral artery infarction, sparing of the superior calcarine cortex, cuneus, and superior paraventricular geniculocalcarine tract differentiated superior quadrantanopia from homonymous hemianopia.
- 18 July 2025
- Published by S. Karger AG in Cerebrovascular Diseases
- Vol. 55 (3) , 311-317
Open AccessThe Anatomy of Infarcts Causing Hemianopia and Quadrantanopia in Posterior Cerebral Artery Stroke
Introduction: Following posterior cerebral artery stroke, infarcts causing quadrantanopia are thought to be smaller than those causing homonymous hemianopia. We investigated whether these two presentations were different due to the varying involvement of the striate, extrastriate cortex, or geniculocalcarine tract. Methods: Patients with unilateral posterior cerebral artery infarcts on magnetic resonance imaging and visual field defects, as identified by automated visual field perimetry, were included. Infarcts were manually segmented and registered to a standard brain template to facilitate comparison. Infarct volume and infarct involvement with geniculocalcarine fiber tracts were calculated in patients with hemianopia and quadrantanopia. Results: There were twenty-two patients: 15 patients with homonymous hemianopia (median age 68 [interquartile range, 55–76 years old]) and 7 with superior quadrantanopia (median age 40 [interquartile range, 30–56 years old]). Infarct volume significantly differed between the two groups: hemianopia 34.6 mL (interquartile range, 21.6–56.3 mL) versus superior quadrantanopia 15.5 mL (interquartile range, 7.2–24.1 mL), p = 0.026. There was significantly greater involvement of infarct with the geniculocalcarine tracts in the hemianopia group, 5.7 mL (interquartile range, 2.3–8.2 mL) than the superior quadrantanopia group, 2.0 mL (interquartile range, 1.2–2.7 mL), p = 0.042. Infarct involvement with cuneus and the calcarine cortex also significantly differed between the two groups (p ≤ 0.01). Conclusion: In posterior cerebral artery infarction, sparing of the superior calcarine cortex, cuneus, and superior paraventricular geniculocalcarine tract differentiated superior quadrantanopia from homonymous hemianopia.
- 1 June 2025
- Published by Wolters Kluwer Health in Stroke
- Vol. 56 (6) , 1642-1645
STAY-HOME: Endovascular Thrombectomy for Patients With Prestroke Disability
Thrombectomy significantly improves clinical outcomes in patients with large vessel occlusion. However, the foundational trials used restrictive criteria, excluding patients with prestroke disability and included a few elderly patients (>80 years old). In addition, results from registries demonstrated conflicting results. We propose that there is equipoise for the elderly and prestroke disability to undergo thrombectomy and that trials are urgently required to guide treatment.
- 1 June 2025
- Published by Wolters Kluwer Health in Stroke
- Vol. 56 (6) , 1642-1645
STAY-HOME: Endovascular Thrombectomy for Patients With Prestroke Disability
Thrombectomy significantly improves clinical outcomes in patients with large vessel occlusion. However, the foundational trials used restrictive criteria, excluding patients with prestroke disability and included a few elderly patients (>80 years old). In addition, results from registries demonstrated conflicting results. We propose that there is equipoise for the elderly and prestroke disability to undergo thrombectomy and that trials are urgently required to guide treatment.
- Open Access
- 31 July 2023
- Published by Frontiers Media SA in Frontiers in Neuroscience
- Vol. 17, 1237920
Corrigendum: Phase I trial outcome of amnion cell therapy in patients with ischemic stroke (I-ACT)
Corrigendum: Phase I trial outcome of amnion cell therapy in patients with ischemic stroke (I-ACT)
- 31 July 2023
- Published by Frontiers Media SA in Frontiers in Neuroscience
- Vol. 17, 1237920
Open AccessCorrigendum: Phase I trial outcome of amnion cell therapy in patients with ischemic stroke (I-ACT)
Corrigendum: Phase I trial outcome of amnion cell therapy in patients with ischemic stroke (I-ACT)
- Open Access
- 30 September 2021
- Published by S. Karger AG in Cerebrovascular Diseases
- Vol. 51 (2) , 248-258
Topographic Evolution of Anterior Cerebral Artery Infarction and Its Impact on Motor Impairment
Introduction: Motor deficit is common following anterior cerebral artery (ACA) stroke. This study aimed to determine the impact on the motor outcome, given the location of descending corticofugal fiber tracts (from the primary motor cortex [M1], dorsal and ventral premotor area [PMdv], and supplementary motor area [SMA]) and the regional variations in collateral support of the ACA territory. Methods: Patients with ACA vessel occlusion were included. Disruption to corticofugal fibers was inferred by overlap of tracts with a lesion on computed tomography perfusion at the onset and on magnetic resonance imaging (MRI) poststroke. The motor outcome was defined by dichotomized and combined National Institute of Health Stroke Scale (NIHSS) sub-scores for the arm and leg. Multivariate hierarchical partitioning was used to analyze the proportional contribution of the corticofugal fibers to the motor outcome. Results: Forty-seven patients with a median age of 77.5 (interquartile range 68.0–84.5) years were studied. At the stroke onset, 96% of patients showed evidence of motor deficit on the NIHSS, and the proportional contribution of the corticofugal fibers to motor deficit was M1-33%, SMA-33%, and PMdv-33%. By day 7, motor deficit was present in Conclusion: Improvements in the motor outcome seen in ACA stroke may be due to the relative sparing of M1 fiber tracts from infarction. This may occur as a consequence of the posterior location of M1 fiber tracts and the evolving topography of ACA stroke due to the compensatory capacity of leptomeningeal anastomoses.
- 30 September 2021
- Published by S. Karger AG in Cerebrovascular Diseases
- Vol. 51 (2) , 248-258
Open AccessTopographic Evolution of Anterior Cerebral Artery Infarction and Its Impact on Motor Impairment
Introduction: Motor deficit is common following anterior cerebral artery (ACA) stroke. This study aimed to determine the impact on the motor outcome, given the location of descending corticofugal fiber tracts (from the primary motor cortex [M1], dorsal and ventral premotor area [PMdv], and supplementary motor area [SMA]) and the regional variations in collateral support of the ACA territory. Methods: Patients with ACA vessel occlusion were included. Disruption to corticofugal fibers was inferred by overlap of tracts with a lesion on computed tomography perfusion at the onset and on magnetic resonance imaging (MRI) poststroke. The motor outcome was defined by dichotomized and combined National Institute of Health Stroke Scale (NIHSS) sub-scores for the arm and leg. Multivariate hierarchical partitioning was used to analyze the proportional contribution of the corticofugal fibers to the motor outcome. Results: Forty-seven patients with a median age of 77.5 (interquartile range 68.0–84.5) years were studied. At the stroke onset, 96% of patients showed evidence of motor deficit on the NIHSS, and the proportional contribution of the corticofugal fibers to motor deficit was M1-33%, SMA-33%, and PMdv-33%. By day 7, motor deficit was present in Conclusion: Improvements in the motor outcome seen in ACA stroke may be due to the relative sparing of M1 fiber tracts from infarction. This may occur as a consequence of the posterior location of M1 fiber tracts and the evolving topography of ACA stroke due to the compensatory capacity of leptomeningeal anastomoses.