RapidAI Shows High Sensitivity, Specificity for Detecting Intracranial Hemorrhage, Large Vessel Occlusion

Russell Cerejo, MD

RapidAI, a telehealth software designed to improve clinical decision making and patient workflow, showed relatively high sensitivity and very high specificity in detecting intracranial hemorrhage (ICH), as well as demonstrated high accuracy in detecting high grade stenosis (HGS) and large vessel occlusion (LVO). These findings were presented at the 2022 International Stroke Conference (ISC), held February 9-11, in New Orleans, Louisiana.^1,2

Of the 2 abstracts presented, the first included data from a study of 1388 patients, 1251 from the emergency department (ED) and 137 inpatients (IP) who received head CTs on the ED scanner at a level 1 trauma center. Rapid ICH, an imaging tool derived from RapidAI, was accessed by initial interpreting emergency radiologists or neuroradiologists. Each case was reviewed by a board certified neuroradiologist who had access to their initial report Rapid ICH output, and prior exams. If there was a disagreement between readers, a third reader adjundicated the result and their decision was considered final.¹

Among those in the ED, 139 had ICH and 1112 did not, whereas in the IP cohort, 100 had ICH and 37 did not. At the conclusion of the analysis, Rapid ICH showed an overall sensitivity of 79% (ED: 73%; IP: 88%), overall specificity of 95% (ED: 96%; IP: 84%), overall positive predictive value of 76% (ED: 65%; IP: 94%) and overall negative predictive value of 96% (ED: 97%; IP: 72%).

READ MORE: Understanding the Precursors of Neurological Brain Diseases and Catching up With Heart Disease

"RapidAI is basically an artificial intelligence software that can detect whether there’s any bleeding or blood on noncontrast CT head scans. Secondly, it can also detect if there’s any large vessel occlusion in the brain,” Russell Cerejo, MD, vascular neurologist, Allegheny Health Network and study author, told NeurologyLive^®. "It sort of bypasses the radiologist and spits out the results to a phone or the back system. Obviously, the radiologist reads the scan over to confirm that’s what we’re seeing, but it’s so quick and takes somewhere less than a minute to get the scan to your phone to tell you if there’s bleeding in the brain or if there’s large vessel occlusion. That can help expedite stroke therapy."

The second study evaluated Rapid LVO, another subtool of RapidAI, on its ability to detect LVO and its impact treatment time and clinical outcomes. Led by the same authors, 1203 patients presenting with stroke symptoms who received CTAs with Rapid LVO were included. Radiology exams were considered positive if LVO (ICA/M1 or M2 occlusion) or HGS was found.2

Of the cohort, 126 patients had M1/ICA occlusion, 71 had M2 occlusion, and 134 had HGS of the ICA or MCA. In the detection of M1 occlusion, LVO, or HGS/LVO, Rapid LVO software showed sensitivities of 90%, 82%, and 82%, and specificities of 82%, 85%, and 95%, respectively. Furthermore, this approach had positive predictive values of 37%, 52%, and 87%, and NPVs of 99%, 96%, and 93%, respectively, in the detection of these outcomes.

Those eligible for mechanical thrombectomy (MT) were transferred to a comprehensive stroke center to receive treatment. CTA to groin puncture time was calculated for 1-year time intervals before and after Rapid LVO installation for patients undergoing MT. After deployment of Rapid LVO, the CTA to groin puncture time was significantly lower (93 minutes vs 68 minutes; P <.05). Relative to those who didn’t receive Rapid LVO, a higher proportion of patients achieving functional independence, indicated by modified Rankin Scale (mRS) scores of 2 or below (P <.04). Additionally, these patients also demonstrated lower average 90-day mRS scores (P <.02).

"It’s a useful software. Validating this software to see that it can help is important because when something is new, everybody is excited about it, but it has to show that there is benefit to that,” Cerejo added. “In our limited experience, we were able to show that there is benefit to using the software."

For more coverage of ISC 2022, click here.

REFERENCES
1. Chang W, Eisenmenger L, Cerejo R, Li C, Goldberg MF. Artificial intelligence for automated detection of intracranial hemorrhage (Rapid ICH): initial clinical experience. Presented at: ISC 2022; February 9-11; New Orleans, LA. Poster WP129

2. Chang W, Eisenmenger L, Cerejo R, Li C, Goldberg MF. Artificial intelligence for automated detection of large vessel occlusion using relative vessel density (Rapid LVO): one year experience in a multihospital integrated delivery network. Presented at: ISC 2022; February 9-11; New Orleans, LA. Poster WMP68