Efficient Evolving Deep Ensemble Medical Image Captioning Network.

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Image Analysis
• Deep Learning for Diagnostics

Background:

• Current AI diagnostic models offer binary outcomes (disease present/absent).
• There's a need for AI models that provide explanatory information, including infection rates.
• Existing deep ensemble models can be sensitive to parameter control.

Purpose of the Study:

• To propose an efficient deep ensemble medical image captioning network (DCNet) for enhanced disease diagnosis.
• To develop an evolving DCNet (EDC-Net) that optimizes control parameters for improved performance.
• To provide doctors and patients with more detailed diagnostic information beyond binary classification.

Main Methods:

• Ensemble learning by combining VGG16, ResNet152V2, and DenseNet201 models in DCNet to prevent overfitting.
• Utilizing self-adaptive parameter control-based differential evolution (SAPCDE) to tune EDC-Net's control parameters.
• Evaluating EDC-Net's feature extraction capabilities on biomedical images.

Main Results:

• EDC-Net demonstrates efficient extraction of potential features from biomedical images.
• Comparative analysis on the Open-i dataset shows EDC-Net outperforms existing models.
• Significant improvements were observed in BLUE scores (1.258% to 1.098%) and kappa statistics (1.548%).

Conclusions:

• EDC-Net offers a robust solution for detailed medical image analysis and diagnosis.
• The proposed method enhances diagnostic accuracy by providing infection rates alongside disease identification.
• EDC-Net represents a significant advancement in AI-driven E-healthcare applications.