Enhancing bone radiology images classification through appropriate preprocessing: a deep learning and explainable...

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer Science

Background:

• Deep learning models are crucial for medical image classification.
• Explainable Artificial Intelligence (XAI) is increasingly used to validate deep learning models.
• Beyond accuracy, result authenticity and model accountability are vital in medical AI.

Purpose of the Study:

• To highlight the importance of result authenticity and model accountability in medical deep learning.
• To propose a targeted preprocessing method for medical datasets used in deep learning.

Main Methods:

• Comparison experiments on bone radiology image datasets using various deep learning neural networks.
• Evaluation of preprocessing methods' impact on model prediction performance.
• Quantitative and visual assessment using XAI to determine prediction reasonability and reliability.

Main Results:

• DenseNet201 achieved the highest validation accuracy (78%) on the bone radiology dataset.
• Appropriate preprocessing increased model performance by an average of 0.06.
• XAI confirmed that preprocessing helps models focus on abnormality areas.

Conclusions:

• Novel application of targeted preprocessing (background/irrelevant part removal) to medical images.
• Enhances deep learning model performance and reliability in classification tasks.
• Improves medical diagnostics accuracy and reliability by removing redundant features.