Abstract
Introduction The mantra “time is brain” has led to significant efforts to expedite tPA administration time and initiation of mechanical thrombectomy for the treatment of ischemic stroke. Evidence supporting surgical intervention for intracerebral hemorrhage (ICH) remains elusive. Numerous clinical trials have had negative overall results and others are still on-going. Most investigators and trial protocols have agreed that intervention should occur rapidly to decrease the risk of hematoma expansion and reduce perihematomal edema, however no trial has directly studied this question. Artificial intelligence applications have been shown to improve ischemic stroke workflow metrics, both decreasing transfer times from outside hospitals and rapidly alerting the interventional teams. We aimed to determine whether the implementation of an ICH detection algorithm that provides immediate active notification to provider cell phones would improve hemorrhagic stroke workflow at our institution.
Methods A retrospective review was performed of patients presenting between January 2018 and March 2022 who suffered a spontaneous ICH and for whom the neurosurgical service was consulted for possible surgical intervention. Stroke workflow metrics were compared pre- and post-implementation of the VizAI (Viz.ai, San Francisco, California, USA) smartphone application. Additional demographic, clinical, and radiographic information was all collected.
Results 188 adult patients were identified during the study period. Time between identification of ICH to neurosurgical team notification was reduced by 50 minutes after the implementation of VizAI (p<0.002). The number increases to 57 minutes when hemorrhages not identified by the ICH algorithm were excluded.
Discussion Active notification of the neurosurgical team by an artificial intelligence application significantly reduces the time from hemorrhage identification to surgical evaluation. Further studies are needed to evaluate whether this results in a clinical benefit.
Introduction
In the United States, nearly 800,000 people suffer from strokes annually1. The vast majority of these strokes are ischemic, however over ten percent are hemorrhagic stroke. ICH represents a significant acute and long-term cost on the health care system2-4. While mechanical thrombectomy is well-established as the standard of care for large vessel occlusion (LVO) ischemic strokes, surgical intervention for intracerebral hemorrhage is debated in the literature and randomized controlled trials are still investigating the question5,6. However, there is mounting evidence that the mantra “time is brain” is also applicable to intracerebral hemorrhage7.
Hematoma expansion occurs in over 70% of patients during the first 24 hours following symptom onset and may lead to increased morbidity and mortality8. Clinically significant hematoma expansion typically occurs early in the patient’s clinical course9. Multiple risk factors have been identified for hematoma expansion, including an angiographic spot sign, coagulopathy, and initial hemorrhage volume10. Interventions have targeted these risk factors and have included surgical evacuation, tranexamic acid or recombinant Factor VII administration, and aggressive blood pressure control11-16.
The use of artificial intelligence to identify LVOs in ischemic stroke has been established at numerous centers around the country and has shown benefit in workflow metrics17. Automated ICH detection using Viz Artificial Intelligence (Viz.ai, San Francisco, California, USA) is a newer feature that was implemented at our center in 2021. There is preliminary evidence that it is an effective tool for identifying ICH and alerting providers18.
We performed a retrospective review of patients with ICH for whom the neurosurgical service was consulted for possible surgical intervention to determine whether Viz.ai detection improved the speed at which these patients were evaluated.
Methods
A retrospective chart review of patients with spontaneous non-traumatic intracerebral hemorrhage who were evaluated for surgical evacuation by the neurosurgical service between January 2018 and March 2022 at a regional Comprehensive Stroke Center was undertaken. Patients under <18 years of age were excluded.
Neurosurgical team notification data was obtained from the electronic medical record and from the Viz.AI smartphone application which was first implemented in 2021 across the neurosurgical residency program. Viz.AI was applied to all non-contrast head CT scans performed at our tertiary center. Demographic and clinical information, stroke workflow metrics, and radiographic hemorrhage data was collected. Hemorrhage volume was calculated using the ABC/2 method. All statistical analyses were performed using SPSS Statistics for Windows, Version 28.0. (IBM Corp., Armonk, NY).
Results
188 patients were identified with spontaneous intracerebral hemorrhages. 100 of these patients were identified prior to the implementation of Viz.AI. Afterwards, an additional 88 patients were identified, of which 73(83%) were positively identified by the artificial intelligence algorithm. Overall, both groups had very similar baseline demographics. The mean age of patients in the AI alert group was 68.6 versus 68.9 (p = 0.88); 60% of patients in the alert group were male as opposed to 47% (p=0.075); baseline GCS in the alert group was 9.5 compared to 10 (p=0.35); baseline NIHSS was 19.1 compared to 16.3 (p=0.13); 59% of patients had a history of HTN compared to 67% (p=0.27); 19% had a history of diabetes mellitus compared to 23% (p=0.49) and 18% of patients in both groups had atrial fibrillation. The only significant baseline difference was 14% of patients in the alert group were on anticoagulation, whereas 27% of patients in the control group were on anticoagulation (p=0.03). These results are summarized in Table 1.
The mean ICH score was slightly higher in the AI-Identified cohort compared to the control cohort(2.5 versus 2.0, p=0.02). There was no significant difference in the volume(39.8 versus 38, p=0.76). There was a higher proportion of patients with IVH in the AI-Identified cohort compared to the control group(56% versus 42%, p=0.04)(Table 2).
The average time from CTH to neurosurgery notification was 23.2 minutes in the AI-Identified cohort compared to 80 minutes in the control group(p<0.001). There was no significant difference between length of stay. There was also no significant difference in the rates of ICH evacuation(12% versus 17%, p=0.43)(Table 3).
Discussion
Our study demonstrates that the use of artificial intelligence to rapidly identify spontaneous intracranial hemorrhage and notify the neurosurgical team can significantly decrease the time from presentation to surgical evaluation. The Viz.ai smartphone application notifies providers about a patient with ICH and allows the non-contrast CT head scan to be viewed rapidly. This led to a more rapid notification and involvement of the neurosurgical team when implemented at our institution. There was no significant difference in the number of patients who underwent surgical evacuation after identification.
Clinical benefits of this technology for ICH are much more difficult to assess. Currently there are no randomized clinical trials which demonstrate a significant benefit to surgical evacuation of ICH. However, this work is ongoing and there is theoretical benefit to earlier intervention. Rapid evacuation of hematomas may prevent delayed hemorrhage expansion and reduce perihematomal edema. Enrollment in clinical trials may also improve as multiple trials require randomization within 24 hours or less of symptom onset19,20. Finally, there are non-surgical interventions which may help reduce hematoma expansion including blood pressure control and reversal of anticoagulation. Earlier notification of the presence of ICH would allow for earlier implementation of these measures. Perhaps the utilization of this program would have an even bigger impact at smaller facilities where off-site radiology is utilized to allow for faster identification of spontaneous ICH and appropriate management by the emergency room.
Our study does have multiple limitations. ICH workflow at our institution is not standardized. Prior to Viz.ai, ICH patients were primarily identified by the emergency department, and the stroke neurology service provided the initial assessment and admission. There remains no protocol in place to consult the neurosurgical service. It is initiated at the discretion of either the emergency department or the stroke service providers, and in some instances by the radiologist. Since the implementation of Viz.ai, the neurosurgical service has played a more active role in early management of these patients as we are notified immediately by the Viz.ai application without the need for intermediaries.
Conclusions
This study is the first to show that artificial intelligence improves the response times for surgical evaluation of spontaneous intracranial hemorrhages. Future research is needed to assess whether there is clinical benefit for these patients.
Data Availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.
Sources of Funding
All funding for this study was provided by the Department of Neurosurgery at Albany Medical Center.
Disclosures
JD is a consultant for MicroVention Terumo.
Acknowledgements
Conception and study design, data collection and analysis, writing and editing: all authors.
References
