Distinguishing lumpy skin disease from coat patterns using morphological priors in deep learning.

Area of Science:

• Veterinary Medicine
• Computer Vision
• Animal Health

Background:

• Lumpy skin disease (LSD) is a significant transboundary cattle disease causing economic losses.
• Early detection of LSD on farms is difficult due to subtle or obscured lesions.
• Mobile technology offers potential for image-based disease screening.

Purpose of the Study:

• To develop a morphology-driven image analysis model for lumpy skin disease detection in cattle (LSDD).
• To create a tool that assists in early, on-farm screening of cattle for LSD-compatible lesions.

Main Methods:

• Developed a morphology-driven image analysis model (LSDD) incorporating texture refinement and morphology consistency modules.
• Trained and tested the model using datasets reflecting diverse farm conditions, including varying illumination and coat patterns.
• The system provides an image-level classification of 'healthy' or 'lesion'.

Main Results:

• The LSDD model effectively highlights subtle nodules and suppresses background noise.
• It distinguishes between circular/clustered lesions and elongated coat markings.
• Extensive experiments demonstrated the model's practicality under varied farm conditions.

Conclusions:

• The LSDD model serves as a practical, morphology-driven screening aid for herd-level monitoring.
• It supports earlier identification of animals with LSD-compatible lesions.
• This tool enhances on-farm surveillance and facilitates timely control measure implementation.