Predicting stage B2 myxomatous mitral valve disease in dogs using machine learning and routine clinical data

Figures

• 

  Figure 1. Flow chart of the study population. Electronic medical records from Chungbuk National University Veterinary Teaching Hospital between January 2018 and December 2025 were reviewed. Dogs were assigned to the B2 or non-B2 groups based on the 2019 American College of Veterinary Internal Medicine (ACVIM) staging guidelines. The non-B2 group included dogs with ACVIM Stage B1 disease and healthy controls.

• 

  Figure 2. Confusion matrix of the machine learning model for detecting American College of Veterinary Internal Medicine Stage B2 myxomatous mitral valve disease. This figure shows the distribution of the true and predicted labels in the test dataset. The model correctly classified 44 of the 53 non-B2 dogs and 22 of the 27 B2 dogs. The color bar on the right indicates the number of dogs corresponding to the color intensity of the cells, with darker blue representing a higher number of dogs.

• 

  Figure 3. Receiver operating characteristic curve of the final machine learning model for detecting American College of Veterinary Internal Medicine Stage B2 myxomatous mitral valve disease. The curve illustrates the diagnostic performance of the model on the test dataset. The area under the receiver operating characteristic curve was 0.922.

• 

  Figure 4. Calibration plot of the gradient boosting model for detecting American College of Veterinary Internal Medicine Stage B2 myxomatous mitral valve disease. The calibration plot compares the mean predicted probability with the observed proportion of Stage B2 dogs in the test dataset (n = 80). Predicted probabilities were grouped into five bins, with each data point representing 16 dogs (n = 16). The dashed diagonal line represents perfect calibration. The Brier score was 0.120.

• 

  Figure 5. Feature importance ranking of the gradient boosting model. The horizontal bars represent the relative importance scores of the top 15 variables. Importance scores were normalized so that their sum across all evaluated features was 1.0. Vertebral heart scale had the highest importance score (0.263), followed by vertebral left atrial size (0.078), C-reactive protein (0.046), and monocyte count (0.039). VHS, vertebral heart scale; VLAS, vertebral left atrial size; CRP, C-reactive protein; MCH, mean corpuscular hemoglobin; PCV, packed cell volume; MCHC, mean corpuscular hemoglobin concentration; SDMA, symmetric dimethylarginine; NLR, neutrophil-to-lymphocyte ratio.