Tools and Resources for Large-Scale Morphometrics
Date
2023-08
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Recent advances in imaging and machine learning have revolutionized both the acquisition and analysis of high-dimensional data. Unfortunately, these techniques have not been effectively exported to morphometrics, or the quantitative analysis of shape and form, which is central to biology and related disciplines for understanding phenotypic variation. Lack of implementations, different model systems and experimental designs, small sample sizes, and limited expertise continue to yield self-contained morphometric studies with unstandardized datasets that cannot be aggregated for increasingly powerful analyses. This dissertation presents heuristic and learning-based tools alongside big data resources for large-scale morphometrics. I begin by summarizing key morphometric paradigms, focusing on data acquisition, aggregation, and classification. Next, I develop an automated landmarking and shape optimization pipeline based on image registration and artificial neural networks. After demonstrating its efficacy, I use this phenotyping framework to build MusMorph, a database of mouse morphology data (N=10,056) containing anatomical atlases, aligned micro-computed tomography images, landmark configurations, and segmentations spanning numerous strain/genotype combinations and developmental stages. Finally, I introduce the R package pheble to perform a meta-analysis of learning-based classification algorithms across high-dimensional phenotypic datasets ranging in organismal family and anatomy. My analysis shows how ensemble learning, or combining predictions from individual learners, can maximize classification performance whilst being data agnostic. Altogether, these studies represent tools and resources that can accelerate, standardize, and augment morphometric analyses for novel phenomic inquiries.
Description
Keywords
classification, image registration, landmark, machine learning, morphometrics, phenomics
Citation
Devine, J. (2023). Tools and resources for large-scale morphometrics (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.