Object Recognition Methodology for Detecting Multiple Diseases in Musculoskeletal and Chest Images of Companion Animals

This study proposes a disease detection method that utilizes object recognition to assist veterinarians in X-ray image diagnosis.The proposed method can identify diseases and their locations as well as distinguish between the background and foreground of the veterinary images by improving their fuzzy images.This was achieved by educating YOLOv5 and YOLOv7 to identify five types of musculoskeletal and thoracic diseases in companion animals.The results revealed that fuzzy YOLOv5-l6 had the highest mean average precision (mAP) of 69.8%, whereas fuzzy YOLOv7-d6 and fuzzy YOLOv7-e6e had the highest detection accuracy of 78.35%.The use of YOLOv5 with fuzzy image enhancement shows significant performance improvements, with a detection accuracy increase of approximately 10%-15% compared to previous studies using ResNet50 or YOLOv3.Although the fuzzy image enhancement technique significantly improved the YOLOv5 performance across different test scenarios, it also significantly degraded the YOLOv7 performance.By utilizing this methodology, medical images can simultaneously detect multiple diseases, leading to enhanced diagnostic capabilities compared with single-disease detection strategies.