DEELE-Rad: exploiting deep radiomics features in deep learning models using COVID-19 chest X-ray images

Area of Science:

• Radiomics and Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Deep Learning for Disease Diagnosis

Background:

• COVID-19 diagnosis relies heavily on medical imaging, with chest X-rays being a primary tool.
• Radiomics and deep learning (DL) offer potential for enhanced diagnostic accuracy and decision support.
• Existing methods may require complex, multi-step processes for image analysis.

Purpose of the Study:

• To develop and evaluate the DEELE-Rad approach, integrating DL and machine learning (ML) for COVID-19 classification using chest X-ray images.
• To leverage DL models with transfer learning for extracting deep radiomics features, bypassing traditional radiomics steps.
• To provide visual explicability for model predictions to support clinical decision-making.

Main Methods:

• The DEELE-Rad approach utilizes DL models (VGG16, ResNet50V2, DenseNet201) pre-trained on ImageNet for feature extraction.
• Deep radiomics features (100 and 500 dimensions) were extracted and fed into ML classifiers with automated parameter tuning and cross-validation.
• A visual explanation method was employed to provide insights into the decision-making process.

Main Results:

• The DenseNet201 end-to-end classifier achieved an 89.97% AUC with 500 deep radiomics features.
• The ensemble DEELE-Rad method improved performance to 96.19% AUC.
• The ML-based DEELE-Rad achieved the highest performance with 98.39% accuracy and 99.19% AUC, demonstrating robustness and confidence in image analysis.

Conclusions:

• The DEELE-Rad approach offers a robust and reliable method for analyzing chest X-ray images in COVID-19 scenarios.
• This technique can significantly benefit healthcare specialists by enhancing diagnostic accuracy and supporting clinical decision-making.
• The study provides reproducible code, facilitating further research and application in clinical settings.