CTA Scans and Clinical Data for Abdominal Aortic Aneurysm Rupture Risk

A retrospective cohort of 263 symptomatic abdominal aortic aneurysm (AAA) patients, with data split into a 230-patient development cohort and a 33-patient independent temporal test set. This multimodal dataset combines sequential computed tomography angiography (CTA) slices with six key clinical biomarkers, created by Jiaxin Cheng and published in April 2026 to develop a deep learning model for predicting impending rupture.

Use Cases

• Train a multimodal deep learning model to predict AAA rupture risk by fusing sequential CTA slices with six clinical biomarkers via a bidirectional cross-attention mechanism.
• Validate model robustness using the independent temporal test set of 33 patients to assess performance against domain shifts and treatment-related data leakage.
• Generate Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations from the CTA data to interpret model predictions and assess anatomical plausibility.
• Benchmark predictive performance against clinical baseline models using metrics like area under the curve (AUC), sensitivity, and negative predictive value (NPV).
• Conduct propensity score matching on features like age, sex, and maximum aortic diameter to create balanced case-control cohorts for model development.

Strengths

• Includes 263 patient records with a clearly defined independent temporal validation cohort of 33 recent cases.
• Model achieved an AUC of 0.880 on the independent test set, demonstrating validated predictive performance.
• Dataset supports interpretability, with Grad-CAM visualizations reported as anatomically plausible in 78.8% of cases.
• Uses propensity score matching to create a balanced development cohort of 75 impending rupture cases and 75 stable controls.

Limitations

• Small sample size of 263 total patients and only 33 in the independent test set limits statistical power and generalizability.
• Retrospective, single-study design may introduce selection bias and limit external validity to other populations or clinical settings.
• The dataset is very small in scale at 157.1 KB, containing a research paper rather than the raw CTA images or structured clinical data tables.

Provenance

Source

Jiaxin Cheng, published on figshare.

Collection Method

Retrospective cohort study of symptomatic AAA patients.

Time Range

Not specified, but uses a temporal test set from the most recent year of the study period.

Freshness

Last updated April 2026, representing recent research.