Evaluating sexual dimorphism in Romanov sheep: A comparative 3D shape analysis of manual and automated landmarking

3D geometric morphometric techniques are increasingly employed to assess shape variation with high precision. A critical step is anatomical landmark placement, traditionally done manually, an accurate but time-consuming and inconsistent process for large datasets. To overcome these limitations, automated landmarking tools using artificial intelligence have emerged. This study compared manual and automated landmarking methods to evaluate cranial sexual dimorphism in Romanov sheep.

Thirty sheep cranium (15 males, 15 females) were scanned using high-resolution computed tomography with 0.6 mm slice thickness. Manual and ALPACA-based landmarking were applied to reconstructed 3D models, and shape analyses were performed using GPA and PCA in 3D slicer, followed by statistical testing in PAST.

Manual landmarking revealed sex-specific shape differences, particularly in the foramen magnum, occipital condyles, processus paracondylaris, protuberantia occipitalis externa, linea nuchae, prosthion, and palatal regions. ALPACA successfully identified biologically meaningful variation, mainly in the nasal, dental, and caudal skull regions. Both approaches confirmed significant sexual dimorphism, with ALPACA offering faster processing and reduced observer bias. PCA results indicated that manual landmark placement was more successful in distinguishing male and female cranial morphologies.

Automated landmarking via ALPACA demonstrated robust performance in capturing cranial sexual dimorphism, offering a reproducible and efficient alternative to manual methods. These findings highlight the utility of AI-supported morphometric workflows in veterinary anatomy, zooarchaeology, and forensic applications. The fact that manual landmarking is more successful in distinguishing females from males in the PCA scatter plot also highlights the need for further development of automated landmarking.