Robust and explainable framework to address data scarcity in diagnostic imaging.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Deep Learning

Background:

• Deep learning advances medical diagnostics but is limited by data scarcity.
• Developing robust diagnostic tools for areas with limited training data is crucial.

Purpose of the Study:

• Introduce the Efficient Transfer and Self-supervised Learning based Ensemble Framework (ETSEF) to address data scarcity in medical diagnostics.
• Evaluate ETSEF's effectiveness and robustness across diverse medical imaging tasks.

Main Methods:

• ETSEF combines Transfer Learning and Self-supervised Learning with ensemble methods.
• Employs data augmentation, feature fusion, feature selection, and decision fusion.
• Validated on five medical imaging tasks: endoscopy, breast cancer, monkeypox, brain tumor, and glaucoma detection.

Main Results:

• ETSEF improved diagnostic accuracy by up to 13.3% over baseline ensemble models.
• Achieved up to 14.4% improvement compared to state-of-the-art methods.
• Demonstrated robustness and trustworthiness using explainable AI techniques (Grad-CAM, SHAP, t-SNE).

Conclusions:

• ETSEF offers a flexible and high-performing solution for medical imaging tasks with limited data.
• The framework shows potential for real-world application in data-scarce environments.
• ETSEF outperforms large-scale models in challenging medical imaging scenarios.