A Visual Analytics Approach Using the Exploration of Multidimensional Feature Spaces for Content-Based Medical Image...

Area of Science:

• Medical Imaging
• Computer Science
• Data Visualization

Background:

• Content-based image retrieval (CBIR) aids medical diagnosis but suffers from the semantic gap, where visual similarity doesn't match user intent, hindering retrieval of relevant images.
• Existing CBIR methods lack intuitive explanations for similarity, making it hard for users to verify results or discover missed relevant images with outlier features.

Purpose of the Study:

• To propose a novel approach for medical CBIR that enhances search space exploration.
• To overcome limitations of current CBIR systems by addressing the semantic gap and improving result verification.

Main Methods:

• Developed Visual Analytics for Medical Image Retrieval (VAMIR), a tool enabling guided visual exploration of medical image datasets.
• Utilized the query image as a reference point for interactive exploration of the entire dataset.
• Conducted user and case studies involving computed tomography (CT) and positron emission tomography-CT (PET-CT) images.

Main Results:

• Demonstrated VAMIR's capability to facilitate interactive exploration and improve the retrieval process for medical images.
• Showcased the tool's effectiveness in retrieving relevant computed tomography and multimodality positron emission tomography and computed tomography images.
• Provided a mechanism for users to better understand and verify retrieved image similarities.

Conclusions:

• VAMIR offers a promising solution to the semantic gap in medical CBIR by enabling guided visual exploration.
• The developed visual analytics approach enhances the discovery and verification of relevant medical images.
• VAMIR has potential applications in medical diagnosis, education, and research by improving image retrieval efficiency and user trust.