Performance of image-based deep learning models for aortic dissection segmentation and diagnosis: a systematic review...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Diseases

Background:

• Aortic dissection is a life-threatening condition requiring accurate and timely diagnosis.
• Deep learning (DL) models are emerging as powerful tools for medical image analysis.

Purpose of the Study:

• To systematically evaluate the accuracy of image-based DL models for aortic dissection segmentation and diagnosis.
• To provide an evidence base for developing intelligent detection tools for aortic dissection.

Main Methods:

• A comprehensive literature search was conducted across major databases (Cochrane Library, PubMed, Embase, Web of Science) up to November 3, 2024.
• Included studies were assessed for risk of bias using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool.
• Meta-analysis was performed on data from 48 included studies (28 segmentation, 20 diagnostic tasks).

Main Results:

• DL models achieved high segmentation accuracy (Dice coefficients >89% for lumen and aorta).
• Computed tomography (CT)-based DL models demonstrated pooled sensitivity of 0.94 and specificity of 0.92 for diagnosis.
• Computed tomography angiography (CTA)-based DL models showed pooled sensitivity of 0.94 and specificity of 0.95.
• Electrocardiogram (ECG)-based DL models had pooled sensitivity of 0.85 and specificity of 0.90.
• DL models performed comparably to or better than clinicians in diagnostic tasks (pooled sensitivity 0.79, specificity 0.95).

Conclusions:

• Image-based DL models exhibit high accuracy in segmenting and diagnosing aortic dissections.
• These AI models show potential as clinical assistive tools, performing on par with or exceeding clinician performance.
• Future research should focus on multicenter validation, clinical workflow integration, and enhancing model generalizability for broader adoption.