Angelos Sharobeam
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Angelos Sharobeam has a total of 57 co-authors. Here are the top co-authors by number of shared publications:
- Michael Valente
Department of Neurology, Monash Health, Clayton, Victoria, Australia
4 shared publications - Bernard Yan
Department of Neurology, Melbourne Brain Centre, The Royal Melbourne Hospital, The University of Melbourne, Parkville, Victoria, Australia
3 shared publications Andrew Bivard
Melbourne Brain Centre at Royal Melbourne Hospital, Melbourne, Australia; Department of Medicine, University of Melbourne, Melbourne, Australia
2 shared publications- Aravinda Sridhara Rao
Department of Electrical and Electronic Engineering, The University of Melbourne, Parkville, Melbourne, VIC, Australia
2 shared publications Bernard Yan
Melbourne Brain Centre, The Royal Melbourne Hospital, Parkville, Australia
2 shared publications- Freda Werdiger
Melbourne Brain Centre at Royal Melbourne Hospital, Melbourne, Australia; Department of Medicine, University of Melbourne, Melbourne, 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
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 5248-5252
Zero-shot Stroke Lesion Segmentation via CAM-guided Prompting of MedSAM2
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 5248-5252
Open AccessZero-shot Stroke Lesion Segmentation via CAM-guided Prompting of MedSAM2
- 8 December 2024
- Published by Springer Nature
- p. 104-120
EDAF: Early Detection of Atrial Fibrillation from Post-stroke Brain MRI
- 8 December 2024
- Published by Springer Nature
- p. 104-120
EDAF: Early Detection of Atrial Fibrillation from Post-stroke Brain MRI
- Open Access
- 29 May 2023
- Published by BMJ in Stroke and Vascular Neurology
- Vol. 9 (1) , 30-37
Early anticoagulation in patients with stroke and atrial fibrillation is associated with fewer ischaemic lesions at 1 month: the ATTUNE study
Background The optimal time to commence anticoagulation in patients with atrial fibrillation (AF) after ischaemic stroke or transient ischaemic attack (TIA) is unclear, with guidelines differing in recommendations. A limitation of previous studies is the focus on clinically overt stroke, rather than radiologically obvious diffusion-weighted imaging ischaemic lesions. We aimed to quantify silent ischaemic lesions and haemorrhages on MRI at 1 month in patients commenced on early (5 mL, regardless of anticoagulation timing. Conclusion Commencing anticoagulation <4 days after stroke or TIA is associated with fewer ischaemic lesions at 1 month in AF patients. There is no increased rate of haemorrhage with early anticoagulation. These results suggest that early anticoagulation after mild-to-moderate acute ischaemic stroke associated with AF might be safe, but randomised controlled studies are needed to inform clinical practice.
- 29 May 2023
- Published by BMJ in Stroke and Vascular Neurology
- Vol. 9 (1) , 30-37
Open AccessEarly anticoagulation in patients with stroke and atrial fibrillation is associated with fewer ischaemic lesions at 1 month: the ATTUNE study
Background The optimal time to commence anticoagulation in patients with atrial fibrillation (AF) after ischaemic stroke or transient ischaemic attack (TIA) is unclear, with guidelines differing in recommendations. A limitation of previous studies is the focus on clinically overt stroke, rather than radiologically obvious diffusion-weighted imaging ischaemic lesions. We aimed to quantify silent ischaemic lesions and haemorrhages on MRI at 1 month in patients commenced on early (5 mL, regardless of anticoagulation timing. Conclusion Commencing anticoagulation <4 days after stroke or TIA is associated with fewer ischaemic lesions at 1 month in AF patients. There is no increased rate of haemorrhage with early anticoagulation. These results suggest that early anticoagulation after mild-to-moderate acute ischaemic stroke associated with AF might be safe, but randomised controlled studies are needed to inform clinical practice.
- 25 April 2023
- Published by Elsevier in European Journal of Radiology
- Vol. 164, 110845
Automated occlusion detection for the diagnosis of acute ischemic stroke: A detailed performance review
- 25 April 2023
- Published by Elsevier in European Journal of Radiology
- Vol. 164, 110845
Automated occlusion detection for the diagnosis of acute ischemic stroke: A detailed performance review
- Open Access
- 17 March 2023
- Published by Elsevier in NeuroImage
- Vol. 271, 119985
Benchmark dataset for clot detection in ischemic stroke vessel-based imaging: CODEC-IV
- 17 March 2023
- Published by Elsevier in NeuroImage
- Vol. 271, 119985
Open AccessBenchmark dataset for clot detection in ischemic stroke vessel-based imaging: CODEC-IV
- 29 November 2021
- Published by Wolters Kluwer Health in Current Opinion in Neurology
- Vol. 35 (1) , 24-30
Advanced imaging in acute ischemic stroke: an updated guide to the hub-and-spoke hospitals
Purpose of review: The purpose of this review is to summarize the role of the hub-and-spoke system in acute stroke care, highlight the role of advanced imaging and discuss emerging concepts and trials relevant to the hub-and-spoke model. Recent findings: The advent of advanced stroke multimodal imaging has provided increased treatment options for patients, particularly in rural and regional areas. When used in the hub-and-spoke model, advanced imaging can help facilitate and triage transfers, appropriately select patients for acute therapy and treat patients who may otherwise be ineligible based on traditional time metrics. Recent, ongoing trials in this area may lead to an even greater range of patients being eligible for acute reperfusion therapy, including mild strokes and patients with large core infarct volumes. Summary: Integration of advanced imaging into a hub-and-spoke system, when complemented with other systems including telemedicine, improves access to acute stroke care for patients in regional and rural areas.
- 29 November 2021
- Published by Wolters Kluwer Health in Current Opinion in Neurology
- Vol. 35 (1) , 24-30
Advanced imaging in acute ischemic stroke: an updated guide to the hub-and-spoke hospitals
Purpose of review: The purpose of this review is to summarize the role of the hub-and-spoke system in acute stroke care, highlight the role of advanced imaging and discuss emerging concepts and trials relevant to the hub-and-spoke model. Recent findings: The advent of advanced stroke multimodal imaging has provided increased treatment options for patients, particularly in rural and regional areas. When used in the hub-and-spoke model, advanced imaging can help facilitate and triage transfers, appropriately select patients for acute therapy and treat patients who may otherwise be ineligible based on traditional time metrics. Recent, ongoing trials in this area may lead to an even greater range of patients being eligible for acute reperfusion therapy, including mild strokes and patients with large core infarct volumes. Summary: Integration of advanced imaging into a hub-and-spoke system, when complemented with other systems including telemedicine, improves access to acute stroke care for patients in regional and rural areas.
- Open Access
- 8 December 2020
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 11
Patterns of Infarction on MRI in Patients With Acute Ischemic Stroke and Cardio-Embolism: A Systematic Review and Meta-Analysis
Background: Cardioembolic strokes are common however atrial fibrillation, the most common cause, is often asymptomatic and difficult to detect. There is evidence that infarct topography and volume on magnetic resonance imaging may be associated with specific stroke etiologies.Aim: A systematic review and meta-analysis were undertaken to summarize the available evidence on the association between stroke etiology, infarct topography, and volume.Methods: A systematic review was conducted using Medline (OVID), Embase (OVID), and PubMed databases. Hand searches of the gray literature and of reference lists in relevant articles were also performed. A quality assessment was undertaken, based on the STROBE checklist. For each study, the number of patients with and without a CE source of stroke and infarct topography was collected and outcomes presented as odds ratios (OR) with 95% CI and p-values.Results: Four thousand eight hundred and seventy-three patients with ischemic stroke were included, of whom 1,559 were determined to have a CE source. Bilateral infarcts (OR 3.41; 95% CI 2.20–5.29; p < 0.0001) and multiple territory infarcts (OR 1.57; 95% CI 1.12–2.21; p = 0.009) were more common in patients with a CE source of stroke, than patients without a CE source. Lacunar infarcts (OR 0.49; 95% CI 0.31–0.80; p = 0.004) were more likely to occur in patients without a CE source. No significant difference between the frequency of multiple infarcts (OR 0.96; 95% CI 0.57–1.61; p = 0.87) anterior circulation (OR 1.45; 95% CI 0.83–2.53; p = 0.19) or posterior circulation infarcts (OR 1.06; 95% CI 0.72–1.57; p = 0.75), between the two groups were identified. Three out of four studies examining volume, found a significant association between increased infarct volume and CE source of stroke. A sensitivity analysis with cryptogenic and undetermined stroke sources assumed to be cardioembolic, did not alter the associations observed.Conclusion: The findings of this systematic review and meta-analysis are broadly consistent with previous literature and provide more robust evidence on the association between infarct topography, volume and stroke etiology. Our findings may assist with refining cardiac investigations for patients with cryptogenic stroke, based on infarct topography.
- 8 December 2020
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 11
Open AccessPatterns of Infarction on MRI in Patients With Acute Ischemic Stroke and Cardio-Embolism: A Systematic Review and Meta-Analysis
Background: Cardioembolic strokes are common however atrial fibrillation, the most common cause, is often asymptomatic and difficult to detect. There is evidence that infarct topography and volume on magnetic resonance imaging may be associated with specific stroke etiologies.Aim: A systematic review and meta-analysis were undertaken to summarize the available evidence on the association between stroke etiology, infarct topography, and volume.Methods: A systematic review was conducted using Medline (OVID), Embase (OVID), and PubMed databases. Hand searches of the gray literature and of reference lists in relevant articles were also performed. A quality assessment was undertaken, based on the STROBE checklist. For each study, the number of patients with and without a CE source of stroke and infarct topography was collected and outcomes presented as odds ratios (OR) with 95% CI and p-values.Results: Four thousand eight hundred and seventy-three patients with ischemic stroke were included, of whom 1,559 were determined to have a CE source. Bilateral infarcts (OR 3.41; 95% CI 2.20–5.29; p < 0.0001) and multiple territory infarcts (OR 1.57; 95% CI 1.12–2.21; p = 0.009) were more common in patients with a CE source of stroke, than patients without a CE source. Lacunar infarcts (OR 0.49; 95% CI 0.31–0.80; p = 0.004) were more likely to occur in patients without a CE source. No significant difference between the frequency of multiple infarcts (OR 0.96; 95% CI 0.57–1.61; p = 0.87) anterior circulation (OR 1.45; 95% CI 0.83–2.53; p = 0.19) or posterior circulation infarcts (OR 1.06; 95% CI 0.72–1.57; p = 0.75), between the two groups were identified. Three out of four studies examining volume, found a significant association between increased infarct volume and CE source of stroke. A sensitivity analysis with cryptogenic and undetermined stroke sources assumed to be cardioembolic, did not alter the associations observed.Conclusion: The findings of this systematic review and meta-analysis are broadly consistent with previous literature and provide more robust evidence on the association between infarct topography, volume and stroke etiology. Our findings may assist with refining cardiac investigations for patients with cryptogenic stroke, based on infarct topography.
- 27 October 2020
- Published by Wiley in Internal Medicine Journal
- Vol. 50 (10) , 1274-1277
Anticoagulant prescribing in patients with ischaemic stroke: what has changed over a decade?
A study at our institution conducted 10 years ago identified that warfarin was under‐utilised in patients with atrial fibrillation. We replicated this study and compared the results. There was a significant increase in the proportion of patients appropriately anticoagulated (63% vs 9%, P < 0.001). Eighty‐seven percent of eligible patients were appropriately anticoagulated on discharge (86% previously). Seventy‐two percent of treated patients were prescribed a direct oral anticoagulant. Of these, 10% were discharged with a sub‐therapeutic dose without obvious explanation.
- 27 October 2020
- Published by Wiley in Internal Medicine Journal
- Vol. 50 (10) , 1274-1277
Anticoagulant prescribing in patients with ischaemic stroke: what has changed over a decade?
A study at our institution conducted 10 years ago identified that warfarin was under‐utilised in patients with atrial fibrillation. We replicated this study and compared the results. There was a significant increase in the proportion of patients appropriately anticoagulated (63% vs 9%, P < 0.001). Eighty‐seven percent of eligible patients were appropriately anticoagulated on discharge (86% previously). Seventy‐two percent of treated patients were prescribed a direct oral anticoagulant. Of these, 10% were discharged with a sub‐therapeutic dose without obvious explanation.