A recently published retrospective study has shown that a deep learning model based on non-contrast computed tomography (CT) can detect endoleak after endovascular aneurysm repair (EVAR) “with high sensitivity”.
Writing in CardioVascular and Interventional Radiology (CVIR), Qingqi Yang, Jinglang Hu (both Sun Yat-Sen University, Guangzhou, China) and colleagues share that their study involved 245 patients who underwent EVAR for abdominal aortic aneurysm (AAA) between September 2016 and December 2022, with all patients undergoing both non-enhanced and enhanced followup CT. The researchers note that the presence of endoleak was evaluated based on CT angiography (CTA) and radiology reports.
Yang et al detail in their methods section that, first, the aneurysm sac was segmented, and radiomic features were extracted on noncontrast CT, after which statistical analysis was conducted to investigate differences in shape and density characteristics between aneurysm sacs with and without endoleak. Subsequently, a deep learning model was trained to generate predicted segmentation of the endoleak, and a binary decision was made based on whether the model produced a segmentation to detect the presence of endoleak. “The absence of a predicted segmentation indicated no endoleak, while the presence of a predicted segmentation indicated endoleak,” the authors write.
Finally, Yang and colleagues add that the performance of the model was evaluated by comparing the predicted segmentation with the reference segmentation obtained from CTA, with model performance assessed using metrics such as dice similarity coefficient, sensitivity, specificity, and the area under the curve (AUC).
In CVIR, the authors relay that their study finally included 85 patients with endoleak and 82 patients without endoleak. “Compared to patients without endoleak, patients with endoleak had higher CT values and greater dispersion,” they report, noting that the AUC in the validation group was 0.951, dice similarity coefficient was 0.814, sensitivity was 0.877, and specificity was 0.884.
Yang et al highlight several limitations to their study. These include its retrospective nature and the fact that the team used imaging data from the past seven years, which the researchers acknowledge “may introduce variations due to differences in imaging equipment,” among others.
The authors conclude that the deep learning model constructed using non-contrast CT “can accurately detect endoleaks”. However, they also stress that “multicentre studies should be conducted to evaluate the diagnostic robustness of the algorithm, and the feasibility of using non-contrast CT for endoleak satisfaction should be explored to further refine the model”.
In an associated commentary, Robert Morgan (St George’s University of London, London, UK) writes that, despite the limitations of the research, the data presented “confirm that the use of AI [artificial intelligence] in the detection of endoleaks is feasible”.
Morgan goes on to say that the data also “confirm that endoleaks can be detected using deep learning techniques even in the absence of intravenous contrast medium,” which he highlights as a “very interesting piece of news”.
Whilst acknowledging that further work is required to assess the applicability of these techniques to a wider population, Morgan states that “if the results are replicable, the implications for patients and health systems post-EVAR are potentially significant, removing the need for contrast medium in CT follow-up in many cases”.
