Journal of Cardiovascular, Neurovascular & Stroke https://mycvns.com AI IN TELESTROKE: AN EARLY SINGLE-CENTER EXPERIENCE IN EVALUATING THE PERFORMANCE OF AI FOR DETECTION OF ICH, LVO, AND ASPECTS SCOREKamran U1*, Kanwal S1, Durrani M1, Yousaf A1, Abdul Qayum S2, Ur Rehman A3, Abbasi W3 1Department of Radiology, Rawalpindi Institute of Cardiology, Rawalpindi, Pakistan 2Department of Neurology, Rawalpindi Institute of Cardiology, Rawalpindi, Pakistan 3Department of Cardiology, Rawalpindi Institute of Cardiology, Rawalpindi, Pakistan *Corresponding author: Kamran Ummarah, Department of Radiology, Rawalpindi Institute of Cardiology, Rawalpindi, Pakistan Email: ummarahkamran1@gmail.com DOI: https://doi.org/10.32896/cvns.v8n1.1-10 Received: 10.11.2025 Revised: 01.02.2026 Accepted: 05.02.2026 Published: 31.03.2026 ABSTRACT Background Stroke represents a critical health crisis, causing significant global mortality and long-term disability. Acute ischemic stroke treatment focuses on timely interventions such as tPA administration and mechanical thrombectomy, but the narrow time window and specialist shortages pose challenges. Methods This retrospective study evaluated the efficacy of the Canon Automation Platform software in detecting intracranial hemorrhage (ICH), large vessel occlusion (LVO), and Alberta Stroke Program Early CT Scores (ASPECTS) in 20 stroke patients. Results AI demonstrated 100% sensitivity in identifying ICH and LVO, with high specificity and negative predictive value. While the AI matched radiologist findings perfectly in extreme ASPECTS categories (6 and 10), it showed variability in intermediate scores (specifically score 8). Conclusion These findings serve as a preliminary proof of concept. AI improves diagnostic speed and triage accuracy, though radiologist confirmation remains essential for intermediate ASPECTS and false-positive mitigation. Keywords: Artificial intelligence (AI); Deep learning (DL); Intracranial hemorrhage (ICH); Large vessel occlusion (LVO); Machine learning (ML) INTRODUCTION The A stroke represents a critical health crisis characterized by the obstruction of blood flow to the brain, depriving it of oxygen and essential nutrients [1]. This condition is a major cause of global mortality and long-term disability, underscoring the need for effective diagnostic and therapeutic strategies [2]. Treatment for acute ischemic stroke focuses on restoring blood flow through medications or procedures within strict time limits. Intravenous thrombolysis with alteplase is most effective within 4.5 hours of symptom onset, while endovascular treatment should occur within 6 hours for large vessel occlusions [3,4]. Due to the narrow diagnostic window and a shortage of specialized medical professionals, innovative approaches are essential. Teleneurology has emerged as a significant advancement, offering remote consultations that enhance recovery rates [5]. Artificial intelligence (AI), specifically through machine learning (ML) and deep learning (DL), presents promising solutions to high workloads and slow assessments [7,8]. The integration of AI, such as the computer-assisted Alberta Stroke Program Early CT Score (ASPECTS), has demonstrated high accuracy in detecting early ischemic changes [14,15]. This retrospective study aimed to assess the effectiveness of the Canon Automation platform in detecting ICH, LVO, and ASPECTS in a clinical telestroke setting. METHODOLOGY We conducted a retrospective analysis of patients presenting with stroke symptoms at the RIC Emergency Department between July 2023 and November 2023. We gathered data from clinical documents, imaging reports, and records from the AI software. We collected information about patients' age, gender and Alberta Stroke Program Early Computed Tomography Score (ASPECTS) for anterior circulation infarcts. The Alberta Stroke Program Early CT Score (ASPECTS) is a numerical rating system ranging from 0 to 10 that is utilized in assessing middle cerebral artery (MCA) stroke cases based on CT scan findings. This scoring method involves dividing the MCA vascular territory into segments, with one point deducted from the total score of 10 for each region affected at both the ganglionic and supraganglionic levels as shown in Table 1 and Figure 1. The inclusion criteria are as follows: 1.        Patients with an activated telestroke code. 2.        Completion of CTA head and neck scans. 3.        Analysis via the automated AI software. 4.        Confirmation of findings by radiologists. The imaging data were processed using the Canon Automation Platform, utilizing the ICH, LVO, and ASPECTS modules. These DL-based algorithms analyze non-contrast CT and CTA scans to flag life-threatening neurovascular conditions. LVO specifically targets occlusions in the Internal Carotid Artery (ICA) and the M1/M2 segments of the Middle Cerebral Artery (MCA). ASPECTS automatically segments the 10 defined regions of the MCA territory to calculate a score based on tissue density changes. RESULTS AND DISCUSSION Out of 20 patients, 14 were male and 6 were female (Mean age: 43.11 ± 20.12 years). Significant co-morbidities included diabetes mellitus (79.7%) and hypertension (66.1%). The AI system demonstrated 100% sensitivity for both ICH and LVO, correctly identifying all cases confirmed by radiologists. For ICH, specificity was 86.7% (2 false positives). For LVO, specificity was 88.2%. Both categories yielded a Negative Predictive Value (NPV) of 100%, confirming the software’s reliability as a rule-out tool. The AI showed 100% agreement with radiologists for extreme scores of 6 and 10. However, for the intermediate score of 8, the AI correctly identified only one out of four cases identified by radiologists, indicating lower performance and variability in intermediate ischemic change detection, illustrated in Table 2 and Figures 2-4. The Diagnosing LVO is crucial for mechanical thrombectomy eligibility. While literature reports AI sensitivity between 86% and 97.5%, our study using Canon Automation Platform,achieved 100% sensitivity. This high sensitivity is vital for mobilizing stroke teams promptly [17,18]. However, our findings regarding ASPECTS indicate that AI performance fluctuates. The drop in accuracy for Score 8 suggests that subtle "gray-sign" changes—such as early loss of the insular ribbon—remain difficult for algorithms to distinguish from normal patient variation or chronic leukoencephalopathy [19]. With a sample size of N=20, these findings serve as a preliminary proof of concept. While results are encouraging, a single case can significantly distort percentages in a small cohort. Future multi-center studies with larger datasets are required to validate these performance metrics. Additionally, infrastructure obstacles such as internet interoperability remain a challenge for real-time AI integration. CONCLUSION AI technology in stroke imaging could significantly change future diagnostics. Our observations suggest that automated software like Canon Automation Platform is highly beneficial for rapid triage and LVO detection, especially for non-expert staff. While AI does not currently substitute for an experienced radiologist, its evolution into deep learning suggests it will become an increasingly robust partner in precision medicine. DATA AVAILABILITY Further information regarding the data used for this work can be obtained upon reasonable request. ACKNOWLEDGEMENT We extend our sincere appreciation to our esteemed colleagues in neurology and cardiology for their unwavering support and invaluable guidance throughout this telestroke journey. FUNDING: This work received no external funding. CONFLICT OF INTEREST: The authors have no conflicts of interest to declare and is in agreement with the contents of the manuscript. REFERENCES: Mackay J, Mensah GA. The atlas of heart disease and stroke. Geneva: World Health Organization; 2004. Benjamin EJ, Muntner P, Alonso A, Bittencourt MS, Callaway CW, Carson AP, et al. Heart disease and stroke statistics—2019 update: a report from the American Heart Association. Circulation. 2019;139(10):e56–e528. Turc G, Bhogal P, Fischer U, Khatri P, Lobotesis K, Mazighi M, et al. European Stroke Organisation (ESO)–European Society for Minimally Invasive Neurological Therapy (ESMINT) guidelines on mechanical thrombectomy in acute ischemic stroke. J Neurointerv Surg. 2023;15(8):e2. Powers WJ, Rabinstein AA, Ackerson T, Adeoye OM, Bambakidis NC, Becker K, et al. Guidelines for the early management of patients with acute ischemic stroke: 2019 update to the 2018 guidelines. Stroke. 2019;50(12):e344–e418. Baratloo A, Rahimpour L, Abushouk AI, Safari S, Lee CW, Abdalvand A. Effects of telestroke on thrombolysis times and outcomes: a meta-analysis. Prehosp Emerg Care. 2018;22(4):479–484. Koenig MA, editor. Telemedicine in the ICU. Cham: Springer; 2019. Krittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. J Am Coll Cardiol. 2017;69(21):2657–2664. Lee JG, Jun S, Cho YW, Lee H, Kim GB, Seo JB, et al. Deep learning in medical imaging: general overview. Korean J Radiol. 2017;18(4):570–584. Erickson BJ, Korfiatis P, Akkus Z, Kline TL. Machine learning for medical imaging. Radiographics. 2017;37(2):505–515. Hamburg MA, Collins FS. The path to personalized medicine. N Engl J Med. 2010;363(4):301–304. Ovbiagele B, Nguyen-Huynh MN. Stroke epidemiology: advancing our understanding of disease mechanism and therapy. Neurotherapeutics. 2011;8(3):319–329. Donkor ES. Stroke in the 21st century: a snapshot of the burden, epidemiology, and quality of life. Stroke Res Treat. 2018;2018:3238165. Lee EJ, Kim YH, Kim N, Kang DW. Deep into the brain: artificial intelligence in stroke imaging. J Stroke. 2017;19(3):277–285. Chawla M, Sharma S, Sivaswamy J, Kishore LT. A method for automatic detection and classification of stroke from brain CT images. In: 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Conf Proc IEEE Eng Med Biol Soc. 2009;2009:3581–3584. Saeys Y, Inza I, Larrañaga P. A review of feature selection techniques in bioinformatics. Bioinformatics. 2007;23(19):2507–2517. Tang FH, Ng DK, Chow DH. An image feature approach for computer-aided detection of ischemic stroke. Comput Biol Med. 2011;41(7):529–536. Chatterjee A, Somayaji NR, Kabakis IM. AI detection of large vessel occlusion. Stroke. 2019;50(Suppl 1):WP428. Rodrigues G, Mory B, Bentley P, Chapman M, Barnes A, Halse O, et al. Automated large artery occlusion detection in stroke imaging. Cerebrovasc Dis. 2022;51(2):161–168. Guberina N, Dietrich U, Radbruch A, Seker F, Ringelstein A, Zimmermann J, et al. Detection of early infarction signs with machine learning-based ASPECTS. Neuroradiology. 2018;60(9):911–921. FIGURE LEGENDS: Figure 1: Basal Ganglia and supra ganglionic levels for ASPECT scoring. Figure 2: Left MCA occlusion correctly identified by AI software. Figure 3: Presence of ICH in right basal ganglia correctly identified by AI software. Figure 4: Absence of Bleed and ASPECT score correctly identified by AI software. TABLE LEGEND: Table 1: Alberta Stroke Program Early CT Score (ASPECTS) ranging from 0 to 10 for assessing middle cerebral artery (MCA). Region involved PointsGanglionic level:Caudate1putamen1Anterior or posterior limb of Internal capsule1Insular cortex1M1: cortex corresponding to frontal operculum1M2: cortex corresponding to anterior temporal lobe1M3: cortex corresponding to posterior temporal lobe1Supraganglionic level:M4: MCA territory superior to M11M5:  MCA territory superior to M21M6: MCA territory superior to M31   Table 2: Performance Metrics of AI Systems for ICH, LVO, and ASPECT Scores.   MetricICH - AI SystemLVO - AI SystemASPECT Scores - AI System(NA, 6, 10)ASPECT Scores - AI System (8)ASPECT Scores - AI System (9)Sensitivity100% (5/5)100% (3/3)100%N/AN/ASpecificity86.7% (13/15)88.2% (15/17)100%N/AN/APositive Predictive Value (PPV)71.4% (5/7)60.0% (3/5)100%50% (2/4)100% (3/3)Negative Predictive Value (NPV)100% (13/13)100% (15/15)100%N/AN/A 10