Registration-based workflow for shape study

Quantitative analysis of shape variation across species has been limited by difficulties in aligning anatomical structures. Although computational techniques for registration have become more accessible, challenges remain that hinder their widespread use in morphometrics. We present a two-step process implemented in Python and MATLAB. The first step is Automated 3D Geometric Morphometrics (auto3dgm), a framework for automatically aligning morphologically diverse objects representing the same anatomical structure. We introduce an updated version with a user-friendly installation, improved interface, and enhanced computational efficiency. Auto3dgm outputs "pseudolandmark" files, which can be analyzed using traditional 3D geometric morphometrics, but they present challenges due to naive registration. The second step is Surface Analysis, Mapping, and Segmentation (SAMS), which models shape variation and provides a more informed registration across samples. This continuous registration enables more effective 3D geometric analysis and overcomes the limitations of auto3dgm's pseudolandmarks. We tested the updated auto3dgm on various systems for installation, efficiency, and interface, finding it to be faster and more optimized than previous versions. The new implementation achieved high-quality alignments with fewer pseudolandmarks, improving alignment efficiency. SAMS-based registrations produced biologically homologous feature points and resolved issues with auto3dgm's pseudolandmarks, offering meaningful analysis with fewer pseudolandmarks. In conclusion, this pipeline significantly enhances the practicality of fully automated 3D registration, providing more accurate registrations and enabling advanced machine learning approaches for morphology analysis.