Leveraging descriptor learning and functional map-based shape matching for automated anatomical Landmarking in mouse...

Area of Science:

• * Biological Sciences
• * Geometric Morphometrics
• * Computational Geometry

Background:

• * Manual landmark placement in geometric morphometrics is time-consuming, labor-intensive, and prone to human error, limiting scalability.
• * Traditional methods require significant effort to adapt landmarks to new hypotheses or large datasets.
• * A need exists for efficient, flexible, and automated landmarking methods in biological studies.

Purpose of the Study:

• * To investigate the precision and accuracy of landmarks generated via functional correspondences using a deep functional map network.
• * To automate anatomical landmark placement by interrogating learned functional maps.
• * To compare the performance of the automated method against MALPACA, a standard tool for automatic landmark placement.

Main Methods:

• * Utilized a deep functional map network to learn shape descriptors for establishing point-to-point correspondences between specimens.
• * Developed a methodology to automate landmark identification by querying functional maps.
• * Applied the method to a dataset of rodent mandibles and benchmarked against MALPACA.

Main Results:

• * The developed model demonstrated a speed improvement over MALPACA while achieving competitive accuracy.
• * Performance was comparable to MALPACA, with strong generalizability, especially on smaller training datasets.
• * Visual assessments confirmed the precision of automated landmark placements, with deviations within acceptable ranges.

Conclusions:

• * Unsupervised learning models show significant potential for automating anatomical landmark placement.
• * The proposed approach offers a practical, efficient, and flexible alternative to traditional manual methods.
• * This advancement can enhance the scalability and applicability of geometric morphometrics in large-scale biological research.