A Dirichlet Distribution-Based Complex Ensemble Approach for Breast Cancer Classification from Ultrasound Images with...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast ultrasound is crucial for early breast cancer detection.
• AI-powered systems enhance diagnostic accuracy and speed.
• Deep learning is increasingly vital for medical image analysis and disease diagnosis.

Purpose of the Study:

• To propose a deep ensemble learning model for breast cancer classification from ultrasound images.
• To leverage transfer learning and a novel spaced repetition method for improved AI performance.
• To enhance the explainability of AI models in medical diagnostics.

Main Methods:

• Developed a deep ensemble learning model using Dirichlet distribution and pre-trained transfer learning models (DenseNet201, InceptionV3, VGG16, ResNet152).
• Utilized data augmentation for imbalanced datasets and fivefold cross-validation.
• Applied statistical analyses (ANOVA, Tukey HSD) and Grad-CAM for explainability.
• Adapted the spaced repetition method for AI training to reduce learning times and boost success.

Main Results:

• The proposed model achieved a validation accuracy of 99.60% on the Breast Ultrasound Images Dataset (BUSI).
• The spaced repetition method significantly increased model success and decreased training duration.
• Grad-CAM provided visual explanations, enhancing the interpretability of the AI model's decisions.

Conclusions:

• The developed deep ensemble learning model demonstrates high effectiveness for breast cancer classification using ultrasound images.
• Integrating transfer learning, spaced repetition, and ensemble methods offers a promising approach for AI in medical diagnostics.
• The study highlights the potential of explainable AI (XAI) to build trust and reliability in AI-driven healthcare solutions.