Detection of canine external ear canal lesions using artificial intelligence

Area of Science:

• Veterinary Dermatology
• Artificial Intelligence
• Machine Learning

Background:

• Accurate diagnosis of otitis externa in dogs is challenging but critical for effective treatment.
• Artificial intelligence (AI) offers a promising diagnostic aid, yet no such tools currently exist for veterinary otology.
• This study introduces a novel AI application for canine ear canal assessment.

Purpose of the Study:

• To develop and evaluate a YOLOv5 object detection model for identifying healthy ear canals, otitis, or masses in dogs.
• To serve as a proof-of-concept for AI-driven veterinary dermatology tools.
• To assess the impact of dataset composition on model performance.

Main Methods:

• Four YOLOv5 model variants were trained using distinct datasets of canine ear canal images (healthy, otitis, masses).
• Performance was evaluated using metrics like F1/confidence-curves, mean average precision (mAP50), precision (P), recall (R), and average precision (AP).
• The influence of data duplication and dataset size on model accuracy was analyzed.

Main Results:

• All trained YOLOv5 variants successfully detected and classified canine ear canal conditions.
• Datasets with significant duplication led to inflated performance metrics due to data leakage.
• Larger datasets without duplicates demonstrated superior performance compared to smaller, duplicate-free datasets.

Conclusions:

• The developed AI object detection model shows potential for veterinary dermatology applications.
• The study highlights the importance of high-quality, diverse training data for AI model efficacy.
• Further external validation is required before clinical implementation of this AI tool.