ML-morph

ML-morph automates detection and landmarking of biological structures in images to enable high-throughput morphometric data collection and analysis.

Key Features:

• High-Throughput Automation: Performs automated landmarking at high throughput, exemplified by placing 23 landmarks on 13,686 objects in 3.12 minutes on a personal computer.
• Accuracy and Precision: Produces automated landmarking with errors ranging from 0.5% to 2% of structure length, comparable to manual annotation on standardized datasets.
• General Applicability: Applicable across diverse study systems and suited for semi-rigid biological structures.
• Minimal Specimen Impact: Enables dense phenotyping while minimizing impact on specimens.
• Efficient Data Processing: Includes a file conversion algorithm that leverages existing morphometric datasets to enable end-to-end processing from model training to prediction in a matter of hours.

Scientific Applications:

• Phenomics: Facilitates large-scale phenotypic data collection for studies of size and shape variation.
• Morphometric studies: Increases scale and reproducibility of morphometric analyses across various fields of biology.
• Dense phenotyping: Supports comprehensive, high-dimensional characterization of biological structures.

Methodology:

Employs machine learning techniques to detect and place landmarks on images of semi-rigid biological structures; includes a file conversion algorithm and handles multiple image file formats, with filenames containing special characters (e.g., "&") noted as potentially problematic.