Using Artificial Intelligence for Aneurysm Rupture Risk Monitoring

Clinicians are leveraging artificial intelligence to measure cerebral aneurysms and stratify rupture risk to enhance clinical decision-making and improve outcomes.

Cerebral aneurysms, also known as intracranial or brain aneurysms, present a unique challenge for clinicians, as most are small dilations that occur at weak points along the arteries of the brain and have no symptoms. However, as the aneurysm grows, the risk of rupture increases.

Once the aneurysm expands and the blood vessel wall becomes too thin, the aneurysm will rupture, causing bleeding into the space around the brain known as subarachnoid hemorrhage (SAH), a life-threatening event that requires emergency medical care, according to insights from Johns Hopkins Medicine.

The risks associated with ruptures make detecting and monitoring aneurysms and their growth key to improving patient outcomes, but tracking rupture risk can be challenging due to limitations in current aneurysm measurement approaches.

In a study published in the Journal of NeuroInterventional Surgery last year, researchers sought to evaluate whether artificial intelligence (AI) could help address these issues and improve aneurysm measurement. Using an AI-based volumetric measurement tool known as Rapid Aneurysm to study aneurysms that ruptured during conservative management, the team found that the tool accurately identified aneurysm enlargement not detected from standard measurement practices.

The researchers concluded that the increased sensitivity of the tool to detect aneurysm growth has significant implications for clinical practice. David Fiorella, MD, PhD, director of the Stony Brook Cerebrovascular Center and co-director of the Stony Brook Cerebrovascular and Comprehensive Stroke Center, and David Dodick, MD, professor emeritus at Mayo Clinic and chief science officer and chair for Atria Academy of Science and Medicine, have experience leveraging the tool in the clinical setting. They discussed the tool’s potential to augment clinical decision-making in email interviews with HealthITAnalytics.

THE CHALLENGES OF STRATIFYING ANEURYSM RUPTURE RISK

The difficulties associated with tracking aneurysms are varied, and traditional tools may not be up to the challenge.

“The greatest challenge associated with tracking aneurysms lies in the conventional tools and approaches used to assess their precise shape and size,” Fiorella explained.

Aneurysms can occur in all shapes, sizes, and positions, making them difficult to identify, measure, and track.

“It’s a manual and, oftentimes, time-consuming process for over-burdened care teams,” Dodick said. “Variability between physicians is also a problem when measuring aneurysms, making it difficult to evaluate aneurysm growth accurately and consistently. Not properly identifying growth can have a significant influence on treatment decisions, with potentially serious consequences.”

The clinical strategy to address these challenges and stratify rupture risk often relies on capturing aneurysm images over time to determine what changes may be taking place and if they necessitate further intervention.

“Currently, we assess rupture risk by manually measuring the largest diameter of the aneurysm on [computed tomography angiography] or [magnetic resonance angiography] 2D source images, visualizing aneurysm shape, and comparing it over time in consecutive scans,” Fiorella stated.

Dodick added that aneurysms are typically monitored over specific time intervals — which are adjusted based on aneurysm stability — using these methods until there is a strong indicator of rupture risk. Then, the care team will consider possible interventions and their associated risks.

However, Fiorella noted that with this approach, it could be difficult for clinicians to get accurate and reproducible assessments of an aneurysm’s maximum diameter and shape. This, in turn, can affect their ability to detect growth and predict rupture risk.

“Usually, the clinician looks at the scan to the best of their ability and compares to [prior scans],” Fiorella explained. “Oftentimes, this is limited, as the scans have been performed at different facilities, and sometimes the provided reconstructed imaging planes are different. Direct measurements are performed, but depending on these reconstructions, it can be difficult to measure in the exact same area on each comparison scan, and subtle differences in shape can be difficult or even impossible to detect. In addition, these comparisons are time and labor-intensive for the busy clinician.”

These limitations have led some health systems to look for digital tools that can bolster rupture risk stratification in-clinic.

HOW AI CAN ENHANCE ANEURYSM MANAGEMENT

One of the most significant problems with the current approach to rupture risk monitoring is that it isn’t always accurate, and clinicians can’t always see every detail, particularly if an aneurysm has morphological changes without major changes in size.

“Again, it can be difficult to visually and manually detect small but relevant changes in size or morphology that may influence clinical decision-making and patient outcomes. We have been searching for solutions to help us fill the gaps,” Dodick said.

He also indicated that inter- and intra-observer variability, which are defined as the difference in the measurements between observers and the difference in repeated measurements by the same observer, respectively, are adding to these issues. The process clinicians must use to manually compare multiple scans is also labor- and time-intensive.

This is where AI comes in, according to Dodick and Fiorella. In rupture risk stratification, RapidAI can assist clinicians by creating 3D aneurysm models and providing aneurysm measurement tools that extend beyond traditional linear measurements.

“Tools like RapidAI allow us to quickly turn 2D images into 3D vasculature models, acquire accurate and reproducible measurements (including novel metrics like surface area and volume), find and track the slightest morphological changes, and increase physician and patient confidence in management decisions,” Dodick stated.

Clinicians can then use these features to get a more accurate, complete picture of what’s happening with a patient’s aneurysm rather than relying on a simple visual comparison alone.

“While definitely subject to differences in technique or source data quality, an indication that the volume or surface area has changed directs the clinician to place greater scrutiny on his subsequent evaluation of the scans,” Fiorella said.

The tool can also be applied to aneurysms at various stages of monitoring, whether they are newly detected or have been under surveillance for an extended period. This can increase the efficacy of risk monitoring early in an aneurysm’s growth, which may positively impact patient outcomes.

Most patients with unruptured aneurysms will not have noticeable symptoms. They typically undergo scans or imaging for unrelated reasons, such as headaches or dizziness. Upon detection, however, clinicians can leverage AI to help investigate an aneurysm further.

“After a patient undergoes imaging, we will use RapidAI’s software to automatically process those scans to help us locate and assess the 3D architecture of aneurysms faster and make more informed decisions,” Dodick explained. “This will allow us to easily compare scans side-by-side to examine changes, educate patients on any changes and risks, and, in the case that intervention is required, share imaging analysis across care teams which may be located in different hospital systems.”

The software helps track changes in aneurysms over time using serial imaging taken approximately every year. Each time clinicians perform a new scan, RapidAI provides clinical decision support to address any changes in the aneurysm, Fiorella said. The tool also serves as an educational resource to support patient engagement, helping patients and caregivers visualize how the aneurysm is growing and changing over time.

As with any new, promising tool implemented in the clinical setting, there is significant buzz around the potential to help patients and reduce clinician burden, but evaluating the tool’s real-world performance is critical.

“There are a number of measures that we use to assess the efficacy of the technology we implement in our hospital,” Fiorella said. “Of course, patient outcomes are the most important indicator — but we also measure things like accuracy in measurement, speed, and the number of patients treated. Since implementing the tool, we have certainly been able to identify more patients with aneurysms, and we feel that we will be able to more accurately track aneurysm growth and improve outcomes.”

AVOIDING PITFALLS AND PREVENTING DISPARITIES

While measuring the success of a tool’s deployment, health systems must be vigilant to avoid potential problems, such as bias, the perpetuation of health disparities, and the limitations of the AI itself.

“With any technology, there are benefits, and there are limitations,” Fiorella stated. “The technology helps us to detect, measure, and evaluate growth of the aneurysm, but it does not know the patient’s full medical history and therefore cannot provide treatment recommendations for the patient. There are always things that a physician can do better than AI — the real power is in how we use AI to enhance our capabilities and efficiency.”

Dodick expanded on this point by highlighting that any clinical AI tool may not be 100 percent accurate at all times. Further, there are many unexplored areas of study related to aneurysms and rupture risk that require investigation.

“Whether subtle changes can be seen depending on the cardiac cycle at the time of the image, or whether other systemic factors, such as blood gas balance or medications with vasomotor influence, can influence the 3D morphology of the aneurysm is not yet known,” he noted. “The significance of subtle changes in morphology without a change in overall size is also unclear. These are all fertile areas for prospective research studies.”

Further, to prevent health disparities, those working with clinical AI tools must use methods like extensive clinical validation to ensure that the tools aren’t providing biased results.

“Ensuring new clinical AI tools don’t perpetuate disparities among patients must always be top of mind, and we all need to be intentional to eliminate the disparities in care for both access to diagnostic technology as well as standard and state-of-the-art care,” Dodick said. “Evaluating diverse patient populations used for clinical validation, as well as understanding the data set used to train the algorithms and ensuring regulatory requirements have been met are all essential criteria.”

Despite potential pitfalls and current challenges, Fiorella is confident that AI technology will have a positive impact on rupture risk monitoring and treatment.

“Around 6.5 million people (or 2 percent of the adult population) in the United States alone have an unruptured brain aneurysm,” he stated. “Most often, there aren’t symptoms associated with unruptured aneurysms. When they rupture, they are often deadly. AI has the potential to significantly improve our ability to detect, measure, and track aneurysms to prevent these ruptures from occurring. As research, education, and awareness across the general population progress, we can only expect to see an even more immense impact on how we think about and treat this condition.”

Source: Healthit Analytics