In the high-stakes world of acute medicine, every minute counts. For patients suffering a Large Vessel Occlusion (LVO) stroke, this urgency is absolute, with clinical delays costing millions of neurons. The implementation of Artificial Intelligence (AI) offered a potential breakthrough, but only if the technology could prove its reliability outside of optimized laboratory settings.
At the center of this pivotal transition was Neta Sudry, an informatics specialist affiliated with Tel Aviv University. Her rigorous, independent analysis provided the essential, peer-reviewed evidence that validated the real-world performance of Viz LVO, the groundbreaking AI-powered software developed by Viz.ai.
The Urgent Imperative: Taming the “Time is Brain” Problem
Acute ischemic stroke, particularly LVO, remains a leading cause of long-term disability, affecting hundreds of thousands annually. While treatments like endovascular thrombectomy are highly effective, they are also intensely time-sensitive. Estimates suggest that every delayed minute results in the death of 1.9 million neurons. Before AI, the median time-to-treatment in major stroke centers could range from three to five hours, a timeline far too long given the stakes.
In February 2018, the United States Food and Drug Administration (FDA) cleared Viz ContaCT (commercially known as Viz LVO), establishing a new regulatory category: Computer-Assisted Triage (CADt). This software was designed not to replace human diagnosis, but to analyze computed tomography angiography (CTA) scans rapidly and automatically alert relevant stroke specialists. The objective was simple: cut the workflow time bottleneck.
The Challenge of Validation “In the Wild”
For widespread adoption, the industry needed proof that the AI could perform reliably in the chaotic and diverse clinical environment—a challenge dubbed testing “in the wild.” Standard regulatory data is often generated in controlled settings, which does not reflect the vast heterogeneity of real-world hospital infrastructure.
Neta Sudry’s contribution to the subsequent validation study, “AI-powered stroke triage system performance in the wild,” was foundational. She was explicitly credited with collecting and analyzing the critical data, bringing methodological rigor to the assessment. The study cohort was massive and highly varied, encompassing:
- 2,544 sequential patient records analyzed using the commercial Viz LVO software.
- Data sourced from 139 distinct hospitals across 37 different systems in the US.
- A range of technical variations, including different CT manufacturers, hospital types (Primary and Comprehensive Stroke Centers), and diverse patient demographics.
Sudry’s independent affiliation with Tel Aviv University provided a crucial layer of academic integrity, lending essential credibility to the results of a study involving a commercial product.
Empirical Proof: Quantifying the Clinical Impact
The statistical analysis performed by Sudry and her co-authors yielded compelling metrics that confirmed the AI’s operational value and diagnostic robustness.
1. Workflow Acceleration: Time Saved
The most transformative metric was the efficiency gain. The study demonstrated that the system achieved a median time-to-notification of an LVO alert to the stroke team of just 5 minutes and 45 seconds across all participating hospitals.
This sub-six-minute alert time validated the CADt classification entirely. By automating the detection process and immediately alerting the care team, the system saves critical minutes that directly translate into preserved brain tissue, a verifiable improvement in the traditional stroke pathway.
2. Diagnostic Accuracy: Trust and Reliability
The AI’s diagnostic performance proved resilient to the variations present in the “in the wild” data, providing confidence in its generalizability:
| Metric | Value | 95% Confidence Interval (CI) | Clinical Significance |
| Sensitivity (True Positive Rate) | 96.3% | [92.6% – 98.8%] | Minimizes false negatives, ensuring critical cases are not missed. |
| Specificity (True Negative Rate) | 93.8% | [92.8% – 94.7%] | Minimizes false positives, which is crucial for preventing alert fatigue among stroke teams. |
The high specificity figure is particularly important for institutional adoption; a low false-positive rate maintains the clinical team’s trust in the automated system, ensuring they respond reliably to every alert.
The Lasting Legacy of Rigorous Validation
The definitive conclusion of the validation work was that Viz LVO demonstrates robust performance despite the heterogeneity of setting, equipment, and processes. This finding was the key to unlocking mass deployment. For hospital administrators and clinical leaders, the documented stability across 139 diverse sites effectively de-risked the investment, proving that the benefits observed in specialized centers could be generalized across a broader system.
Neta Sudry’s methodological expertise, rooted in academic rigor, played a foundational role in transitioning AI from a promising technology to a proven clinical necessity. Her work established the empirical standard—high accuracy paired with quantifiable workflow acceleration—that now serves as the benchmark for clinical AI validation in neurovascular care and beyond.
Abdullah Zulfiqar writes about technology in a simple, practical way, helping readers stay updated and make smarter decisions in an ever-evolving digital world.
