Deep Learning for Canine Cardiac Radiography: A Comprehensive Review of Automated Vertebral Heart Score Estimation

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Cardiology

Background:

• Myxomatous mitral valve disease (MMVD) is the most common cardiac condition in dogs.
• Thoracic radiography and the vertebral heart score (VHS) are standard for assessing cardiac enlargement.
• Manual VHS measurement introduces inter-observer variability.

Purpose of the Study:

• To review recent advancements in deep learning for automated VHS estimation in dogs.
• To evaluate the performance and generalizability of different deep learning architectures for VHS prediction.

Main Methods:

• Systematic literature search across three databases, including citation screening.
• Eligibility assessment of 3,729 records, selecting 17 studies for detailed evaluation.
• Comparison of deep learning models based on landmark localization, feature extraction, and predictive performance.

Main Results:

• EfficientNet-B3 and EfficientNet-B7 models exhibited strong performance in landmark regression and VHS prediction.
• Transfer learning significantly improved model performance, especially with limited annotated veterinary data.
• Reviewed studies varied in dataset characteristics and annotation protocols.

Conclusions:

• Deep learning, particularly EfficientNet models, offers promising automated VHS estimation for canine MMVD.
• Further research is needed on dataset diversity, model interpretability, and multi-center validation for clinical integration.
• Automated VHS estimation can enhance diagnostic consistency in veterinary cardiology.