Machine learning classification of complex vasculature structures from in-vivo bone marrow 3D data

• Published in: 2016 IEEE 13th International Symposium on Biomedical Imaging (ISBI) pp. 1217 - 1220
• Main Authors: Khorshed, R. A., Celso, C. Lo
• Format: Conference Proceeding Journal Article
• Language: English
• Published: IEEE 01.04.2016
• Subjects: 3D images; Bifurcation; Biomedical imaging; Blood vessels; Bone marrow; Bones; Classification; Decision Tree Classifier; Decision trees; Machine learning; Stem cells; Three dimensional; Three-dimensional displays; Training; Transgenic; vasculature
• ISSN: 1945-8452
• DOI: 10.1109/ISBI.2016.7493485

Blood vessels inside the bone marrow (BM) play a vital role in the maintenance of hematopoietic stem cell (HSCs). Investigating the interaction of HSCs relative to vasculature has become the main headline for many recent studies. Advances in microscopy and image analysis using mouse models have allowed detection, identification and automated quantification of HSCs alongside their vascular niche. This resulted in new hypotheses concerning the activation state of HSCs adjacent to different blood vessel types (for example sinusoids vs. arterioles). Identifying the different types of BM vasculature has become critically important, however it still requires the use of complex immunostainings ex vivo or transgenic reporter mouse lines in vivo. To eliminate these requirements and increase the throughput of studies focusing on the HSC niche, we present a machine learning classification approach based on the Decision Tree Classifier to classify different regions of bone marrow vasculature into four distinct classes based on their discriminative features.