Overlaid arrow detection for labeling biomedical image regions.
KC S, Wendling L, Antani SK, Thoma GR
IEEE Intelligent Systems, Special Issue on Pattern Recognition, 31(3):66-75, May-June 2016.DOI: 10.1109/MIS.2016.24.
Abstract:
Applications that use biomedical images are valuable not only in medical research and education but also in clinical decision support systems. In applying content-based image retrieval (CBIR) technology, the term content refers to encoding meaningful regions of interest (ROIs) with visual feature representations of colors, shapes, and textures. A key step, therefore, is to automatically identify such ROIs and annotate them according to concepts from biomedical text. Because medical images tend to be complex, researchers often use pointers (that is, arrows and symbols) to highlight meaningful ROIs (see Figure 1) while minimizing distractions from other, less relevant regions. Additionally, ROIs are often referred to in figure captions and mentioned in the text of biomedical articles. Detecting arrows—a core theme of this article—could help identify meaningful ROIs and improve CBIR performance (see the “Related Work in Arrow Detection” sidebar).
KC S, Wendling L, Antani SK, Thoma GR. Overlaid arrow detection for labeling biomedical image regions.
IEEE Intelligent Systems, Special Issue on Pattern Recognition, 31(3):66-75, May-June 2016.DOI: 10.1109/MIS.2016.24.