Deep Learning Framework for Vertebral Heart Size and Cardiothoracic Ratio Estimation in Dogs and Cats Using Thoracic Radiographs

Habtamu Tilahun Mekonnen^1*Robert Martí^1*Núria Puig¹Alejandro Elson²Elena López³Paula Martínez⁴Selene Campos⁴Francisco Hernandez⁴Daniel Mayor⁵Jorge Quilis⁵Xavier Llado¹Arnau Oliver¹Jordi Freixenet¹Xavier Cufí¹

• ¹University of Girona, Girona, Spain
• ²Integral Clinica Veterinaria Cullera, Valencia, Spain
• ³Hospital Veterinario Bluecare, Malaga, Spain
• ⁴Substrate AI, Valencia, Spain
• ⁵4D Medica, Valencia, Spain

Heart disease is a major cause of mortality in aging dogs and cats, with cardiomegaly being the most common radiographic finding. Recent advances in deep learning have demonstrated significant potential in improving the classification and assessment of cardiomegaly. However, clinical adoption of these methods remains limited due to challenges in interpretability and alignment with clinical practice. This study introduces a deep learning framework for automatic estimation of the Vertebral Heart Size (VHS) and Cardiothoracic Ratio (CTR) from thoracic radiographs of dogs and cats, aiding in the early detection and monitoring of cardiac diseases. Leveraging a diverse dataset from two veterinary institutions, the proposed method integrates segmentation models based on Mask R-CNN to extract anatomical regions of interest, followed by precise measurements of cardiac and thoracic dimensions. The method demonstrates strong agreement with manual radiologist evaluations, achieving Pearson Correlation Coefficients of up to 0.922 for VHS and 0.933 for CTR, with regression slope close to 1 and minimal intersection points. Supporting analysis of both lateral and ventrodorsal radiographs, the framework offers a versatile and practical solution for veterinary applications. This work represents a significant advancement toward developing automated, robust diagnostic tools to assist veterinarians, especially in settings with limited access to specialized expertise.