RAPID Aneurysm, a semi-automated artificial intelligence (AI) software programme (RapidAI), is highly accurate in detecting cerebral aneurysms on computed tomography angiography (CTA) imaging, according to a retrospective study published recently in the Journal of Stroke and Cerebrovascular Diseases by investigators Jeremy Heit (Stanford University School of Medicine, Stanford, USA) and colleagues.
Detailing their objectives in this first-of-its-kind evaluation, Heit and colleagues initially note that cerebral aneurysms may result in significant morbidity and mortality, and, as such, identifying them on CTA studies is “critical” to guide patient treatment. They also state that AI platforms to assist with automated cerebral aneurysm detection are “of high interest”, which led them to assess the performance of the RAPID Aneurysm programme in the detection of these aneurysms.
In the study, RAPID Aneurysm was used to retrospectively detect the presence of cerebral aneurysms on CTA studies performed between January 2019 and December 2020. The ‘gold standard’ here was aneurysm presence and location—as determined by the consensus of three expert neuroradiologists. Aneurysm detection accuracy; sensitivity and specificity; positive and negative predictive values; and positive and negative likelihood ratios; were all determined by the programme.
Heit and colleagues report that 51 patients (mean age=56 years, 24 women [47.1%]) with a single CTA were included, and a total of 60 aneurysms were identified for the analysis.
RAPID Aneurysm had a sensitivity of 0.95 (95% confidence interval [CI], 0.863–0.983), a specificity of 1 (95% CI, 0.996–1), a positive predictive value of 1 (95% CI, 0.937–1) and a negative predictive value of 0.997 (95% CI, 0.991–0.999). The authors further state that the programme demonstrated an accuracy of 0.997 (95% CI, 0.991–0.999) for cerebral aneurysm detection, with their findings suggesting that the adoption of aneurysm detection platforms like RAPID Aneurysm “may be a valuable tool to assist radiologists in their interpretation of CTA studies”.
“Several other studies have used automated, computer-assisted or AI techniques to detect the presence of cerebral aneurysms on CTA or MRA [magnetic resonance angiography] examinations,” Heit et al continue. “Few studies have used these techniques to detect cerebral aneurysms on CTA studies—and these studies reported sensitivities of 82–98% and specificity of 0–19% for cerebral aneurysm detection. The sensitivity (95%) and specificity (100%) of RAPID Aneurysm in our study compares favourably with these prior studies.
“A well-validated AI programme that detects cerebral aneurysms is expected to be an asset to radiologists. AI programmes that detect cerebral aneurysms on CTA have been shown to improve radiologists’ sensitivity for the detection of aneurysms. It will be of interest to conduct additional studies to see how RAPID Aneurysm affects radiologists’ performance in interpreting CTA studies in future studies.”
While their concluding message is that “RAPID Aneurysm is accurate in the detection of cerebral aneurysms”, however, Heit et al go on to highlight several limitations of their study, including its retrospective design and relatively small case numbers—which may introduce bias—as well as the fact it was comprised of a limited number of CT vendors, potentially limiting the generalisability of their findings, and the absence of <3mm aneurysms and ruptured cerebral aneurysms in their dataset.
“Whether RAPID Aneurysm improves the performance of human radiologists was not assessed in this study and requires further investigation,” they add.
Speaking to NeuroNews, Heit said: “I believe we are on the cusp of seeing a sea change in our healthcare system through the adoption and implementation of AI. We have an ageing population that will severely strain the ability of our healthcare system to provide for them. We have fewer people going into medical school and [the US] Congress has not increased funding for residency training programmes to increase the number of physicians to care for this ageing population. I think it is very likely that these pressures will help to usher in the rapid adoption of AI in the healthcare system, which will be needed to meet these significant workforce challenges.
“More evidence that AI solutions lead to improvements in care and cost efficiency is needed to help make this transition, and there will need to be changes in reimbursement to hospital systems and physicians to implement these AI platforms. It will certainly be interesting to watch this space in the coming years, but I am hopeful that healthcare can do more with less by leveraging AI technology.”