A predictive model incorporating 3D shape features such as aneurysmal flow lumen and outer wall structure extracted from computed tomography (CT) imaging can improve the ability to predict the growth of abdominal aortic aneurysms (AAAs).
This is according to research presented during yesterday morning’s William J. von Liebig Forum by Anirudh Chandrashekar, DPhil, of Stony Brook Medicine, Stony Brook, New York, who tells VS@VAM that the research has the potential to add to existing metrics for predicting aneurysmal growth, used to inform surveillance intervals and timing for surgical intervention.
Alongside senior author Regent Lee, PhD, from University of Oxford, England, Chandrashekar and colleagues set out to develop a means of predicting aneurysm growth beyond the current gold standard—namely assessment of the diameter of the aneurysmal sac. “What we wanted to do was to develop a prediction paradigm to individualize the follow-up and risk factor management for these patients,” Chandrashekar comments. “This may better inform the timing of surgery down the line.”
The first arm of the project saw the researchers develop a deep learning process to extract and categorize geometric features of the aneurysmal sac that may be associated with AAA-growth, including curvature of the sac and surface irregularity.
The latest step of their research, which will be presented here at VAM, looks at both the lumen and outer aneurysmal wall, and how these correlate with AAA growth. “The methodology is extremely novel, and this is the first time we are incorporating the lumen in this whole decision model,” says Chandrashekar.
To test their model, the researchers have conducted a retrospective analysis of serial CT images taken during surveillance in 192 patients with infrarenal AAAs, categorising patients into either “slow” or “fast” cohorts based upon the progression of their aneurysm.
Chandrashekar will report that integrating the lumen and outer wall structure as unique components within the 3D statistical shape model captured the lumen-thrombus interface, marking this as superior to max diameter, undulation index and radius of curvature in the prediction of AAA growth phenotype, with a p-value of <0.001.
“We can take a CT image, isolate the aneurysm section, extract out the aneurysm shape and extract measurements from the defined shape to a level that is clinically acceptable, and we can do this automatically,” says Chandrashekar of the top line messages of the research.
“Secondly, we are able to improve on already published metrics to predict aneursymal growth, by incorporating not only the aneurysmal sac, but also how the flow lumen interacts with the surrounding thrombus and the surrounding aneurysmal wall,” he adds, commenting that this is “extremely novel in itself.”
Further work has been conducted to validate the model in an independent cohort of patients, which “excitingly,” according to Chandrashekar, shows that the growth model still holds to be predictive.
“Going on from there the next step is to try to establish a prospective study, a longitudinal study following aneurysm patients over time, extracting out additional clinical metrics, for example blood pressure, medication regimen and all those metrics that you would obtain in a randomized controlled study, to see whether this model truly is effective.”
Senior author Regent Lee told VS@VAM: “I congratulate Dr. Chandrashekar in spearheading this international collaborative project and for delivering refinement of our AAA growth prediction model as an independent postdoctoral researcher.”
“The results presented here further highlight the concept of CT image-derived indices as ‘standalone’ biomarkers to predict AAA growth. With appropriate regulatory approvals, this can be further field validated at scale by utilizing existing data already stored in the clinical picture archiving and communication systems (PACS) archives. We look forward to hearing from colleagues who are interested to participate in such collaborative studies.”