Arrow detection in biomedical images using sequential classifier

Abstract

Biomedical images are often complex, and contain several regions that are annotated using arrows. Annotated arrow detection is a critical precursor to region-of-interest (ROI) labeling, which is useful in content-based image retrieval (CBIR). In this paper, we propose a sequential classifier comprising of bidirectional long short-term memory (BLSTM) classifier followed by convexity defect-based arrowhead detection. Different image layers are first segmented via fuzzy binarization. Candidate regions are then checked whether they are arrows by using BLSTM classifier, where Npen++ features are used. In case of low confidence score (i.e., BLSTM classifier score), we take convexity defect-based arrowhead detection technique into account. Our test results on biomedical images from imageCLEF 2010 collection outperforms the existing state-of-the-art arrow detection techniques, by approximately more than 3% in precision, 12% in recall, and therefore 8% in F 1 score. © 2016, Springer-Verlag Berlin Heidelberg.