Can Ultrasound AI Help Detect Advanced CKD in Dogs?

Chronic kidney disease (CKD) is one of the most prevalent renal conditions in elderly dogs, occurring in approximately 0.5–1.5% of the overall canine population¹. It involves persistent structural or functional kidney impairment for at least 3 months and may develop in association with age-related degeneration, genetic predisposition, infections, toxins, and immune-mediated conditions^2,3. Clinical signs can be nonspecific, including weight loss, vomiting, and lethargy, making accurate diagnosis and staging important. CKD diagnosis and staging follow International Renal Interest Society (IRIS) guidelines³.

Renal ultrasonography is a non-invasive, cost-effective, real-time imaging modality commonly used during evaluation. Typical sonographic findings in canine CKD include irregular renal contours, increased echogenicity, reduced corticomedullary differentiation, and decreased renal volume. Renal cortical thickness is also associated with renal function, and these ultrasonographic parameters can correlate with CKD severity^1,4,5. 

Where Does AI Fit Into Renal Ultrasound? 

Although ultrasound provides useful renal information, estimating the degree of functional impairment from imaging alone can be challenging. Interpretation may be subjective, and diagnostic accuracy can vary according to the radiologist’s experience. 

Deep learning, a branch of artificial intelligence (AI), uses artificial neural networks to learn patterns from data¹. In veterinary medicine, AI has primarily been explored in imaging modalities such as radiography, with applications in other modalities also emerging^1,6,7. Before this work, applications specifically targeting canine kidney ultrasonography had been limited, despite previous AI approaches involving renal CT imaging^8,9. 

The deep-learning framework described here used 883 renal ultrasonograms from 198 dogs. Images were assigned to IRIS stages 0–4, with stage 0 representing dogs without abnormalities in renal function parameters or ultrasound findings. The renal regions were identified and classified using the lightweight YOLO-v8-n convolutional neural network¹. 

A Key Finding: Identifying Stage 3–4 Disease 

The model’s performance differed considerably depending on the classification task. When it attempted to distinguish all five IRIS stages individually, accuracy was only 0.46. Performance improved when the stages were grouped into binary categories. 

The strongest performance occurred when the model distinguished IRIS stages 0–2 from stages 3–4. Accuracy, precision, recall, and F1 score were all 0.85, while the area under the ROC curve was approximately 0.8975. 

This threshold may have clinical relevance. Previous findings have demonstrated differences in renal cortical thickness between IRIS stages 2 and 3, while the number of dogs showing multiple abnormal ultrasound findings increases as CKD progresses through stages 2–4. Prognosis also differs between these stages: dogs with IRIS stage 3 had a 2.62-fold higher risk of CKD-related death compared with dogs in stages 1 and 2, while stage 4 was associated with a 4.71-fold higher risk¹. 

How Did AI Compare With Radiologists?¹ 

For the stage 0–2 versus 3–4 classification, the deep-learning model achieved an accuracy of 0.85, compared with 0.48–0.62 among the four radiologists. The average radiologist accuracy was 0.54. 

Importantly, the findings also showed moderate inter-observer agreement among radiologists, although individual radiologists demonstrated substantial to almost perfect intra-observer reliability. This suggests that AI-assisted interpretation may have potential to provide more consistent criteria when evaluating renal ultrasonograms. 

Practical Clinical Insights 

For practicing veterinarians, the key takeaway is that the deep-learning model performed best when distinguishing IRIS stages 0–2 from stages 3–4, achieving 0.85 accuracy, precision, recall, and F1 score. It also outperformed the four radiologists in this specific classification task. 

This suggests that AI-assisted renal ultrasonography may be useful as a supportive tool for identifying more advanced CKD, while complementing the veterinarian’s overall clinical assessment. 

References 

1. Yu H, Lee IG, Oh JY, Kim J, Jeong JH, Eom K. Deep learning-based ultrasonographic classification of canine chronic kidney disease. Frontiers in veterinary science. 2024 Sep 4;11:1443234. https://www.frontiersin.org/journals/veterinary-science/articles/10.3389/fvets.2024.1443234/pdf 

2. Polzin DJ. Chronic kidney disease in small animals. Veterinary Clinics: Small Animal Practice. 2011 Jan 1;41(1):15-30. https://www.vetsmall.theclinics.com/article/S0195-5616(10)00141-5/pdf 

3. O'neill DG, Elliott J, Church DB, McGreevy PD, Thomson PC, Brodbelt DC. Chronic kidney disease in dogs in UK veterinary practices: prevalence, risk factors, and survival. Journal of veterinary internal medicine. 2013 Jul;27(4):814-21. https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/jvim.12090 

4. Perondi F, Lippi I, Marchetti V, Bruno B, Borrelli A, Citi S. How ultrasound can be useful for staging chronic kidney disease in dogs: ultrasound findings in 855 cases. Veterinary Sciences. 2020 Oct 1;7(4):147. https://www.mdpi.com/2306-7381/7/4/147 

5. Siddappa JK, Singla S, Al Ameen M, Rakshith SC, Kumar N. Correlation of ultrasonographic parameters with serum creatinine in chronic kidney disease. Journal of clinical imaging science. 2013 Jun 30;3:28. https://pmc.ncbi.nlm.nih.gov/articles/PMC3779384/pdf/JCIS-3-28.pdf 

6. Li S, Wang Z, Visser LC, Wisner ER, Cheng H. Pilot study: application of artificial intelligence for detecting left atrial enlargement on canine thoracic radiographs. Veterinary radiology & ultrasound. 2020 Nov;61(6):611-8. https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/vru.12901 

7. Banzato T, Wodzinski M, Tauceri F, Donà C, Scavazza F, Müller H, Zotti A. An AI-based algorithm for the automatic classification of thoracic radiographs in cats. Frontiers in veterinary science. 2021 Oct 15;8:731936. https://www.frontiersin.org/journals/veterinary-science/articles/10.3389/fvets.2021.731936/pdf 

8. Ji Y, Cho H, Seon S, Lee K, Yoon H. A deep learning model for CT-based kidney volume determination in dogs and normal reference definition. Frontiers in Veterinary Science. 2022 Oct 28;9:1011804. https://www.frontiersin.org/journals/veterinary-science/articles/10.3389/fvets.2022.1011804/pdf 

9. Ji Y, Hwang G, Lee SJ, Lee K, Yoon H. A deep learning model for automated kidney calculi detection on non-contrast computed tomography scans in dogs. Frontiers in veterinary science. 2023 Sep 20;10:1236579. https://www.frontiersin.org/journals/veterinary-science/articles/10.3389/fvets.2023.1236579/pdf