Related Experiment Video
Updated: Jun 12, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.8K
Explainable artificial intelligence in deep learning-based detection of aortic elongation on chest X-ray images
Estela Ribeiro1,2, Diego A C Cardenas1, Felipe M Dias1,3
1Heart Institute (InCor), Clinics Hospital University of Sao Paulo Medical School (HCFMUSP), Av. Dr. Enéas Carvalho de Aguiar, 44 - Cerqueira César, São Paulo, SP 05403-900, Brazil.
European Heart Journal. Digital Health
|September 25, 2024
Summary
Deep learning models accurately detect aortic elongation on chest X-rays. Explainable AI methods reveal model decision-making, aiding clinical diagnosis of this cardiovascular condition.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Cardiovascular Disease Diagnostics
- Radiology and Image Analysis
Background:
- Aortic elongation, a sign of cardiovascular disease, can stem from aging, congenital issues, or vascular elasticity problems.
- It is linked to serious complications such as aortic aneurysms and dissection, necessitating accurate detection methods.
- Current diagnostic approaches may benefit from advanced computational tools for improved accuracy and efficiency.
Purpose of the Study:
- To evaluate deep learning models (DenseNet, EfficientNet) for detecting aortic elongation using chest X-ray (CXR) images.
- To apply explainable artificial intelligence (XAI) techniques for understanding model decision-making processes.
- To quantitatively assess model interpretations for clinical utility.
Main Methods:
- Chest X-ray images were used to train and fine-tune DenseNet and EfficientNet models via transfer learning.
- Explainable AI methods, including Gradient-weighted Class Activation Mapping (Grad-CAM) and Local Interpretable Model-agnostic Explanations (LIME), were employed.
- The pixel-flipping method was utilized for quantitative assessment of model interpretability.
Main Results:
- EfficientNet demonstrated superior performance over DenseNet in detecting aortic elongation, achieving higher accuracy (86.7%) and AUC (92.7%).
- XAI techniques successfully identified the regions in CXR images indicative of aortic elongation, aligning with clinical expectations.
- Quantitative analysis using pixel-flipping provided insights into the reliability and behavior of the deep learning models.
Conclusions:
- Integrating deep learning with XAI offers a robust strategy for analyzing CXR images to detect aortic elongation.
- Enhanced model interpretability can support clinicians in making timely and accurate diagnoses.
- This approach has the potential to improve patient outcomes through earlier and more precise identification of aortic elongation.
