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

EfficientNet-based deep learning models show strong potential for automating VHS measurement, but require broader clinical validation.

This systematic review evaluates recent advances in deep learning approaches for automated Vertebral Heart Score (VHS) estimation in dogs, with particular relevance to the diagnosis and monitoring of Myxomatous Mitral Valve Disease (MMVD), the most prevalent acquired cardiac condition in canine patients. VHS is a standard radiographic tool for assessing cardiac enlargement, but traditional manual measurement introduces inter-observer variability that may affect diagnostic consistency. Following a structured search of three major scientific databases and backward citation screening, 3,729 records were identified, with 94 studies meeting inclusion criteria and 17 selected for detailed analysis. The review found that convolutional neural networks, EfficientNet architectures (particularly B3 and B7 variants), and transformer-hybrid models showed strong performance in automated landmark localization and VHS prediction. Transfer learning was identified as a critical enabler of model performance, compensating for the limited availability of large, annotated veterinary imaging datasets. Despite promising accuracy metrics across reviewed models, the clinical translation of these tools remains constrained by heterogeneous dataset characteristics, inconsistent annotation protocols, and a lack of multi-centre external validation. The authors emphasize that future research should prioritize dataset diversity, model interpretability, and rigorous external validation studies to facilitate the safe and reliable integration of AI-assisted VHS estimation into routine veterinary radiographic practice. This technology holds significant potential to standardize cardiac assessments and reduce diagnostic variability across clinical settings.

This summary was distilled by AI and may occasionally misinterpret data. Confirm critical details with the primary literature before clinical application.