Can Radiomics Enhance Differentiation of Intracranial Aneurysms on Computed Tomography Angiography?

A new meta-analysis suggests that radiomics models, based on computed tomography angiography (CTA) assessment, may facilitate improved differentiation of intracranial aneurysms.

For the meta-analysis, recently published in the European Journal of Radiology, researchers reviewed data from six studies that included a total of 4,408 patients with mean age ranging between 55 to 66. The researchers noted that three studies focused solely on radiomics features, one study only focused on combined radiomics and image-based features and two studies evaluated combined features and radiomic-only features models.

The meta-analysis authors found that radiomics models provided a pooled area under the curve (AUC) of 86 percent for diagnosing ruptured and unruptured brain aneurysms.

“Prior research indicates that morphological features alone cannot reliably pinpoint the rupture site in multi-aneurysm contexts, raising a need for all lesions to be treated to confirm the obliteration of the ruptured aneurysm. However, radiomics offers a solution by evaluating each aneurysm independently, potentially identifying the ruptured ones with high accuracy and aiding early decision-making to reduce mortality rates,” wrote lead study author Ahmadreza Sohrabi-Ashlaghi, M.D., who is affiliated with the Advanced Diagnostic and Interventional Radiology Research Center at the Tehran University of Medical Science in Tehran, Iran, and colleagues.

In a subgroup analysis, the researchers noted an 87 percent AUC when original filters were utilized for image processing and an 86 percent AUC for the use of enhanced filters designed to bolster spectral dimensions and reduce noise.

“This result is supported by research conducted on six public datasets, which proposes that complex and potentially destabilizing high-dimensional preprocessing techniques might not be essential for the effective classification of rupture status in intracranial aneurysms,” added Sohrabi-Ashlaghi and colleagues.

While the meta-analysis authors suggested that deep learning and artificial intelligence may further enhance the characterization of aneurysms, the current utilization of radiomic features in machine learning models has provided a “suitable tool” in evaluating intracranial aneurysms.

“The success of machine learning models incorporating radiomics over traditional morphological analyses suggests a significant step forward in the non-invasive assessment of aneurysm rupture risk,” emphasized Sohrabi-Ashlaghi and colleagues.

(Editor’s note: For related content, see “What New Research Reveals About Deep Learning and CT Angiography Detection of Cerebral Aneurysms,” “What a Prospective CT Study Reveals About Adjunctive AI for Triage of Intracranial Hemorrhages” and “What a Meta-Analysis Reveals About Cone-Beam CT for Diagnosing Acute Intracranial Hemorrhage.”)

In regard to study limitations, the authors noted the reviewed studies did not evaluate the impact of aneurysm site and size on the performance of the radiomics models. While digital subtraction angiography (DSA) is considered the gold standard for diagnosing intracranial aneurysms, the researchers acknowledged the focus of the meta-analysis on studies of radiomics models modeled on CTA images. The study authors also conceded a lack of external validation across independent data sets. They cautioned about limited extrapolation of the meta-analysis findings to broader populations given that all reviewed studies were conducted on cohorts in China.