Multimodal medical case retrieval using the Dezert-Smarandache theory

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Data Science

Background:

• Digitized medical images and semantic information create extensive medical case databases for diagnostic aid.
• Challenges in utilizing these databases arise from incomplete, uncertain, and conflicting information.
• Existing retrieval systems struggle with the inherent complexities of medical data.

Purpose of the Study:

• To develop a Case Based Reasoning (CBR) system for enhanced medical case retrieval.
• To address limitations of incomplete and uncertain data in medical databases.
• To integrate heterogeneous information sources, including images and symbolic data.

Main Methods:

• A CBR system is developed based on the Dezert-Smarandache Theory (DSmT).
• A specific frame of discernment (theta) is introduced to associate database cases with elements.
• Hybrid DSmT models are utilized for flexible and detailed database modeling.
• Integration of heterogeneous data sources, including image content and symbolic information.

Main Results:

• The developed CBR system demonstrates promising retrieval performance.
• High retrieval precision was achieved: 81.8% on the diabetic retinopathy follow-up (DRD) database.
• Excellent retrieval precision was observed: 84.8% on the screening mammography (DDSM) database.

Conclusions:

• The DSmT-based CBR system effectively handles complex and uncertain medical data for improved case retrieval.
• The system facilitates the integration of diverse data types, enhancing its utility in medical diagnosis.
• The promising results indicate significant potential for this approach in clinical decision support systems.