A multimodal deep learning model for predicting impending rupture in symptomatic abdominal aortic aneurysms using CTA...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Timely diagnosis of impending rupture in symptomatic abdominal aortic aneurysms (AAA) is crucial for hemodynamically stable patients.
• Current diagnostic methods face challenges in accurately assessing rupture risk, impacting emergency decision-making.

Purpose of the Study:

• To develop and validate an interpretable multimodal deep learning model for assessing AAA rupture risk.
• To support emergency decision-making by providing rapid and accurate risk stratification.

Main Methods:

• A retrospective cohort study of 263 symptomatic AAA patients.
• Development of a multimodal deep learning model combining sequential CTA slices with six clinical biomarkers using a ResNet-50 image encoder and a bidirectional cross-attention (BCA) mechanism.
• Interpretability was assessed using Gradient-weighted Class Activation Mapping (Grad-CAM).

Main Results:

• The multimodal model achieved an AUC of 0.898 with 93.3% sensitivity and NPV in the development set, outperforming clinical baselines.
• In an independent temporal validation cohort, the model attained an AUC of 0.880, sensitivity of 92.9%, and NPV of 87.5%.
• Grad-CAM visualizations were anatomically plausible in 78.8% of cases, supporting model interpretability.

Conclusions:

• An interpretable multimodal model integrating CTA and clinical biomarkers was developed and validated.
• The model enables rapid AAA rupture risk stratification, offering improved safety and efficiency in emergency triage.
• Prospective validation is pending to further confirm clinical utility.