Related Experiment Video
Updated: Jun 19, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.1K
Four Transformer-Based Deep Learning Classifiers Embedded with an Attention U-Net-Based Lung Segmenter and Layer-Wise
Siddharth Gupta1, Arun K Dubey2, Rajesh Singh3
1Department of Computer Science and Engineering, Bharati Vidyapeeth's College of Engineering, New Delhi 110063, India.
Diagnostics (Basel, Switzerland)
|July 27, 2024
Summary
This study enhances lung disease diagnosis using Attention U-Net for segmentation and Vision Transformers (ViTs) for classification on chest X-rays. Explainable AI methods improve model reliability and clinical acceptance.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Healthcare
- Deep Learning for Diagnostics
Background:
- Accurate lung disease diagnosis is critical for effective treatment.
- Deep learning models like Convolutional Neural Networks (CNNs) show promise but face challenges in explainability and reliability.
- Integrating advanced architectures can overcome these limitations.
Purpose of the Study:
- To enhance lung disease segmentation and classification accuracy using Attention U-Net and Vision Transformers (ViTs).
- To improve the explainability and reliability of deep learning models in medical diagnostics.
- To assess the clinical utility of these advanced AI techniques.
Main Methods:
- Comparative evaluation of deep learning models, including Attention U-Net for segmentation and various CNNs and ViTs for classification, using chest X-ray data.
- Implementation of explainability techniques such as Gradient-weighted Class Activation Mapping plus plus (Grad-CAM++) and Layer-wise Relevance Propagation (LRP).
- Assessment of model performance using metrics like Dice Coefficient, Jaccard Index, and classification accuracy.
Main Results:
- Attention U-Net achieved high segmentation performance with a Dice Coefficient of 98.54% and Jaccard Index of 97.12%.
- Vision Transformers (ViTs) demonstrated superior classification performance over CNNs, with MobileViT reaching 98.52% accuracy, a 9.26% improvement.
- Classification accuracy increased by 8.3% when using segmented images compared to raw data.
- Explainability methods provided insights into model decision-making processes.
Conclusions:
- The integration of Attention U-Net and ViTs offers significant advantages for lung disease analysis.
- Enhanced explainability of deep learning models fosters trust and aids clinical acceptance.
- These advanced AI approaches hold considerable potential for improving diagnostic accuracy and patient outcomes in clinical settings.
