A deep learning model for CT-based kidney volume determination in dogs and normal reference definition.

Area of Science:

• Veterinary Radiology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Accurate kidney volume assessment is crucial for evaluating renal function and disease severity in dogs.
• Traditional methods like voxel counting are accurate but time-consuming and laborious.
• There is a need for automated, efficient, and accurate kidney volume estimation techniques.

Purpose of the Study:

• To develop the first deep learning model for automatic kidney detection and volume estimation in canine computed tomography (CT) images.
• To create a fast and accurate alternative to manual methods for canine kidney volume measurement.
• To establish a reference range for CT-based normal kidney volume in dogs.

Main Methods:

• Development of a deep learning model using a U-Net Transformer architecture on 182,974 CT image slices from 211 dogs.
• Application of a combined loss function and data augmentation to enhance model performance.
• Validation using Dice Similarity Coefficient (DSC), Lin's Concordance Correlation Coefficient (CCC), and Intraclass Correlation Coefficient (ICC).

Main Results:

• The deep learning model achieved a DSC of 0.915 ± 0.054 for kidney segmentation.
• High agreement was observed between the model's estimated kidney volume and manual voxel count (r = 0.960, CCC = 0.95, ICC = 0.975).
• Kidney volume showed a strong positive correlation with body weight (BW) and a moderate correlation with the body condition score (BCS) index.

Conclusions:

• The developed deep learning model provides a useful tool for automatic kidney volume estimation in dogs.
• The model demonstrates high accuracy and efficiency compared to manual methods.
• Established reference ranges for kidney volume, considering BW and BCS, can aid in canine kidney assessments.