Two commercially available AI software applications may yield different volume estimates in stroke CT perfusion (CTP) studies, but the differences are not significant enough to impact a patient's eligibility for thrombectomy. No, researchers reported in a paper published May 2. Stroke Cerebrovascular Disease Journal. However, there are still important issues to note.
In a study of 362 stroke cases, researchers led by lead author Benjamin Alwood, MD, of the Department of Vascular Neurology at the University of Florida, Jacksonville, found that core volume and circumference estimated by Viz.ai's AI stroke detection software The volumes were compared. We evaluated eligibility for Rapid AI and DEFUSE-3 thrombectomy. Although there were systematic differences in estimates between software applications, there were no statistical differences in DEFUSE-3 thrombectomy eligibility. However, one eligibility difference was found in his one scanner at one facility. This suggests that the scanner model and local CTP protocol may impact performance and cause discrepancies in thrombectomy eligibility.
“We therefore recommend that centers discuss optimal scanning protocols with software vendors and scanner manufacturers to maximize CTP accuracy,” the authors write.
The study was performed on 10 scanners from four different manufacturers. 4 Aquilion Ones (Canon Medical Systems), 3 Discovery scanners (GE HealthCare), 1 Revolution scanner (GE), 1 Ingenuity scanner (Phillips), and 1 Somatom Definition Flash scanner (Siemens) health-conscious people). The authors note that each scanner utilizes slightly different scanning techniques to optimize performance and radiation dose, resulting in slightly different image quality and temporal resolution. Alwood and his team analyzed CTP data from acute stroke patients with large vessel occlusion (LVO) from four institutions.
- From 2018 to 2021, Center A used Rapid.AI to process CTP images of LVO patients diagnosed with CTA in a 6-24 hour window. For the purpose of this study, the original image data was retrospectively reprocessed in Viz.ai. His three remaining centers used Viz.ai and Rapid.AI simultaneously to handle her CTP volume.
- Center B periodically scans all stroke codes with CTP. Data from January 2021 to July 2022 was screened.
- Center C used CTP for acute stroke patients with NIHSS ≥ 4 or CTA-confirmed LVO. Data from December 2022 to March 2023 was screened.
- Center D used CTP for acute stroke patients with NIHSS ≥ 5 or LVO confirmed on CTA. Data from January 2022 to March 2022 was screened.
Results between software packages were compared and categorized by National Institutes of Health Stroke Scale (NIHSS) score, scanner make/model, and facility.
Alwood et al. found that both the core volume and Viz.ai penumbra volume of the Viz.ai software had significantly higher measurements than Rapid.AI. According to the authors, these differences were primarily caused by the Canon Aquilion One scanners at two separate institutions.
For example, Viz.ai's software provided larger estimates with mean differences of 8cc and 18cc for the core and periphery, respectively (p < 0.001). NIHSS subgroup analysis also showed a systematically higher volume for Viz.ai (p < 0.001).
“Despite the difference in volume, no significant difference in eligibility for thrombectomy was observed in this case,” the authors wrote.
On the other hand, additional subgroup analyzes showed a significant difference in penumbra volume for Phillips Ingenuity scanners and thrombectomy eligibility for Canon Aquilion One scanners at one institution (Viz.ai 7% increase, p = 0.03).
“Despite systematic differences in core and penumbra volume estimates between Viz.ai and RAPID.AI, DEFUSE-3 eligibility was statistically different in primary or NIHSS subgroup analyses. “It wasn't,” Alwood and his team concluded. “However, the difference in DEFUSE-3 eligibility was seen in his one scanner at one facility.”
The researchers noted that their results highlight the challenges and uncertainties inherent in using CTP to determine eligibility for thrombectomy.
Read the full report here.
