Caption-based topical descriptors for microscopic images as published in academic papers

Area of Science:

• Biomedical Informatics
• Medical Imaging
• Digital Pathology

Background:

• Visual findings in academic papers are crucial for research.
• Figure and table captions are often overlooked for image retrieval.
• Effective retrieval of biomedical images is essential for scientific advancement.

Purpose of the Study:

• To evaluate caption-based topical descriptors for microscopic breast neoplasm images.
• To compare human-generated descriptors with those from an automated tool (TAPoR).
• To develop a core set of descriptors and map them to the Unified Medical Language System (UMLS).

Main Methods:

• Collected captions from PubMed Central for breast neoplasm microscopic images.
• Generated topical descriptors using human indexers and the TAPoR tool.
• Compared human and machine-generated descriptors.
• Mapped a core descriptor set to the UMLS Metathesaurus via MetaMap Transfer.

Main Results:

• Key descriptors included histologic names, lab procedures, and genetic components.
• Human indexers yielded more relevant descriptors than TAPoR.
• UMLS semantic types identified: Indicator/Reagent/Diagnostic Aid, Organic Chemical, Laboratory Procedure, Spatial/Qualitative/Quantitative Concept.

Conclusions:

• Caption-based descriptors can augment title/abstract indexing for image retrieval.
• Identified UMLS semantic types can form a core metadata set for image description.
• This approach can standardize microscopic image description protocols for medical training.