A generative model based approach to retrieving ischemic stroke images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Medical image retrieval systems often face challenges with manual annotation, leading to inefficiencies in teaching, research, and diagnosis.
• Existing systems struggle with inherent image uncertainties and limited training data, impacting retrieval accuracy.

Purpose of the Study:

• To propose a generative model for automated medical image annotation.
• To enhance the efficiency and effectiveness of medical image retrieval systems.
• To address challenges posed by imperfect image segmentation and data limitations.

Main Methods:

• A generative model approach is proposed, capturing probabilistic relationships between classification tags, lesion patterns, and input features.
• The probabilistic framework operates on imperfect segmentation results, managing uncertainties and insufficient data.
• The model was assessed using ischemic stroke subtype classification.

Main Results:

• The proposed system successfully generated relevant tags for ischemic stroke brain images.
• Demonstrated capability in handling uncertainties and limited information within medical images.
• Preliminary assessments indicate the system's effectiveness in a specific clinical context.

Conclusions:

• The generative model offers a scalable solution for automatic medical image annotation.
• The approach significantly reduces the need for manual labeling and high-precision segmentation.
• This method has broad applicability for large medical image databases, improving retrieval systems.