Detection of Left Atrial Enlargement in Dogs with Deep Learning in Thoracic Radiographs
Computer-aided diagnosis; Heart disease; Image classifier
The study aimed to develop a tool to assist the veterinarian in diagnosing left atrial enlargement on chest X-rays in dogs. The model contains a total of 652 images all used in training and testing divided into two categories “positive” and “negative” three algorithms were used, obtaining the following results: The accuracy of the Neural Network was 89.7%, sensitivity of 90%, specificity of 89.5%, and Area Under the Curve (AUC) 95.8%. The accuracy of Logistic Regression was 88.2%, sensitivity 88.7%, specificity 87.8%, and AUC 94.1%. The Decision Tree accuracy was 69.6%, sensitivity 68.0%, specificity 71.0%, and AUC 69.6%. The classifier model with different algorithms can help radiologists improve the analysis of medical images with error reduction, starting a selective double reading.