83 - Using Regularized Deep Network for troublesome bones: a novel segmentation method for exploring trabecular morphology of the femoral head in a 55 Ma stem primate

Poster Board Number: 83
There are separate poster presentation times for odd and even posters.
Odd poster #s – first hour
Even poster #s – second hour

Co-authors:

Sharon Kuo - Department of Biomedical Sciences - University of Minnesota Duluth; Timothy Ryan - Department of Anthropology - Pennsylvania State University; Mary Silcox - Department of Anthropology - University of Toronto, Scarborough

Abstract Body :Intro and Objective:The use of microCT scanning in the field of biological anthropology allows users to view structures previously impossible to access because of small size, obscuring matrix, or the need to use destructive techniques. However, there have traditionally been challenges associated with segmenting fossil material including poor quality due to taphonomic processes, sediment infill, and bias in manual segmentation. Here we present the first trabecular observations for a stem primate from a well preserved femoral head of a 55 Ma fossil (Microsyops latidens), part of the first dentally associated partial skeleton for Microsyopidae. We also make preliminary qualitative comparisons to extant taxa to assess whether the fossil segmentation is reasonable given the range of variation seen in modern primates. From this we conclude that the RDN segmentation method is capable of processing challenging scans and is retrainable for a variety of uses, including very old fossil material. Materials and Methods:Using the new machine learning based domain enriched regularized deep network (RDN) model, we segmented high resolution microCT scans (resolution of 15.5 - 20 µm) of the femoral heads of M. latidens, Galago senegalensis and Callithrix argentata. Using the unique relearning capabilities of RDN and training data from 5 slices of manually segmented data from the Microsyops fossil per plane, we fine tuned the segmentation to overcome the challenges of segmenting trabecular bone as old, small, and complex as that of a plesiadapiform.Results:Here we show that RDN is capable of accurately segmenting even extremely challenging bone while minimizing the time and bias associated with a fully manual segmentation. Moreover, we present preliminary qualitative comparisons of the M. latidens femoral head internal morphology to that of two diverse comparative extant taxa (G. senegalensis and C. argentata), providing the first glimpse into the functional information available within plesiadapiform trabecular bone. Significance and Conclusions:We conclude that RDN is a promising method for segmenting complex trabecular bone such as that of stem primates. This method of segmentation will allow for the most accurate quantitative comparisons of trabecular variables when applied to both ancient fossil and extant materials. This study is the first to use microCT segmentation methods to explore plesiadapiform trabecular anatomy and therefore is key to the continuing studies of the earliest members of the primate clade. Using RDN’s advanced segmentation capabilities opens the doors to a wealth of trabecular data to be explored within ancient fossil materials in the future. Key Words: microCT, segmentation, plesiadapiformsFunding: NSERC Discovery Grant to MTS