What is the clinical evidence from this study?

Study design: Observational. Population: Type II endoleak after endovascular aortic aneurysm repair (n=277). Intervention: 3D convolutional neural network (CNN) using preoperative volumetric CTA data vs. Follow-up CTA data. Primary outcome: Classification into no T2EL, benign T2EL, or malignant T2EL (Accuracy 76.7%, 95% CI 0.63-0.90).

What question did this study set out to answer?

This study aims to develop a deep learning framework for predicting the occurrence and severity of type II endoleaks from preoperative imaging.

June 5, 2026

NornirNet: A Deep Learning Framework to Distinguish Benign From Malignant Type II Endoleaks After Endovascular Aortic Aneurysm Repair Using Preoperative Imaging

Key Result

A 3D convolutional neural network predicted the presence and severity of Type II endoleak from preoperative CTA with an overall accuracy of 76.7% (95% CI 0.63-0.90) and an AUC of 0.93.

Key Points

This study aims to develop a deep learning framework for predicting the occurrence and severity of type II endoleaks from preoperative imaging.
Retrospective analysis of 277 patients undergoing EVAR from 2010 to 2024.
Preoperative CTA scans were processed and analyzed using a 3D CNN trained on 175 cases and validated on 72 cases, testing on 30 cases.
Evaluated CNN performance metrics included accuracy, precision, recall, F1-score, and AUC based on follow-up CTA data.
78 patients (29.6%) experienced type II endoleaks during follow-up, with 38 (46.3%) showing significant sac enlargement.
The CNN demonstrated overall accuracy of 76.7% (95% CI: 0.63–0.90) and macro-averaged F1-score of 0.77, with class-specific AUCs of 0.93 for no T2EL, 0.91 for benign, and 0.96 for malignant cases.
Misclassifications primarily occurred between adjacent categories, indicating the model's discriminative capacity.

Study Design

Type

Observational (n=277)

Structured PICO

Can a 3D convolutional neural network accurately predict the occurrence and severity of Type II endoleak using preoperative CTA data in patients undergoing EVAR?

Population

277 patients who underwent standard EVAR between 2010 and 2024, followed for a median of 55.5 months to evaluate a deep learning framework for predicting Type II endoleak.

Exposure

Preoperative volumetric computed tomography angiography (CTA) data processed by a 3D convolutional neural network (CNN)

Outcome

Prediction of Type II endoleak (T2EL) occurrence and severity (categorized as no T2EL, benign T2EL, or malignant T2EL)surrogate

A novel 3D deep learning framework can accurately predict the presence and severity of Type II endoleaks directly from preoperative CTA scans, potentially guiding personalized pre-emptive embolization strategies.

Main Result

Effect estimate: Accuracy 76.7% (95% CI 0.63-0.90)

Abstract

Abstract Background Type II endoleak (T2EL) is the most common complication after endovascular aortic aneurysm repair (EVAR). While pre-emptive embolization of side branches may reduce T2EL and reintervention rates, its clinical benefit remains unconfirmed. Current guidelines recommend considering pre-emptive embolization only in selected cases. Aims This study proposes a deep learning framework for preoperative prediction of T2EL occurrence and severity using volumetric computed tomography angiography (CTA) data. Methods A retrospective analysis was conducted on 277 patients who underwent standard EVAR (2010–2024). Preoperative CTA scans were processed for volumetric normalization and fed into a 3D convolutional neural network (CNN), which was trained to classify patients into three categories: no T2EL, benign T2EL, or malignant T2EL. The model was trained on 175 cases, validated on 72, and tested on an independent cohort of 30 patients. The CNN’s performance was evaluated by comparing its predictions with follow-up CTA data. Performance metrics included accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). Results Median follow-up time was 55.5 months 28.03-91.6. During this time a total of 82 (29.6%) T2EL were recorded: 38 (46.3%) displayed significant sac enlargement. The CNN achieved an overall accuracy of 76.7% (95% CI: 0.63–0.90), macro-averaged F1-score of 0.77, and AUC of 0.93. Class-specific AUCs were 0.93 for no T2EL, 0.91 for "benign", and 0.96 for "malignant" cases, confirming high discriminative capacity across outcomes. Most misclassifications occurred between adjacent categories. Conclusion This study introduces the first end-to-end 3D CNN capable of predicting both presence and severity of T2EL directly from preoperative CTA, without manual segmentation or handcrafted features. These findings suggest that preoperative imaging encodes latent structural information predictive of endoleak-driven sac reperfusion, potentially enabling personalized pre-emptive embolization strategies and tailored surveillance after EVAR.

KI fragen

Bookmark