An automated deep learning method and novel cardiac index to detect canine cardiomegaly from simple radiography

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Medical Imaging

Background:

• Cardiomegaly is a common complication of degenerative heart disease in dogs.
• Early detection of cardiac enlargement is crucial for canine health management.
• Current methods for assessing heart size may not be sufficiently accurate for early diagnosis.

Purpose of the Study:

• To develop and validate a novel deep learning-based radiographic index for quantifying canine heart size.
• To introduce the adjusted heart volume index (aHVI) as an objective measure of cardiac enlargement.
• To compare the efficacy of aHVI with existing clinical standards in predicting cardiac enlargement.

Main Methods:

• A deep learning model was developed using 1000 retrospective canine radiographic images.
• The model employed semantic segmentation with tversky loss functions to identify heart and fourth thoracic vertebral body (T4) regions.
• The adjusted heart volume index (aHVI) was calculated using heart area, height, and T4 length from lateral X-rays.

Main Results:

• The developed deep learning algorithms accurately segmented cardiac and T4 regions.
• The adjusted heart volume index (aHVI) was successfully calculated from standard lateral radiographs.
• aHVI demonstrated superior performance in predicting cardiac enlargement compared to the current clinical standard.

Conclusions:

• The adjusted heart volume index (aHVI) is a reliable and accurate deep learning-based tool for assessing canine heart size.
• aHVI offers a promising advancement in the early detection of cardiac enlargement in dogs.
• This AI-driven approach can improve diagnostic accuracy and patient outcomes for canine heart conditions.