Landmark detection in 2D bioimages for geometric morphometrics: a multi-resolution tree-based approach.

Area of Science:

• Bioimage analysis
• Computational biology
• Geometric morphometrics

Background:

• Accurate detection of anatomical landmarks in bioimages is crucial for geometric morphometrics.
• Current landmark detection methods can be time-consuming and labor-intensive.

Purpose of the Study:

• To propose and evaluate variants of a multi-resolution tree-based approach for accelerated landmark detection in bioimages.
• To identify key parameters influencing method performance and compare against existing techniques and human ground truths.

Main Methods:

• Development of multi-resolution tree-based algorithms for automated landmark detection.
• Extensive evaluation on diverse datasets: cephalometric, zebrafish, and drosophila images.
• Comparative analysis against human annotations and established methods.

Main Results:

• The proposed method achieves competitive recognition performance.
• Key parameters, particularly multi-resolution settings, were identified as critical for performance.
• The approach demonstrates speed and generality across different bioimage types.

Conclusions:

• The developed multi-resolution tree-based method offers an efficient and accurate solution for anatomical landmark detection in bioimages.
• The algorithms are integrated into the open-source Cytomine software, with guidelines provided for user accessibility.
• Availability of datasets via a Cytomine server encourages further research in the field.