AI Picks Struvite: Mobile App Accurately Identifies Dog Bladder Stones from X-rays

A Pilot Study on Using an Artificial Intelligence Algorithm to Identify Urolith Composition through Abdominal Radiographs in the Dog

Rute Canejo-Teixeira et al.

Background
Determining urolith composition in dogs is essential for guiding treatment, especially since struvite stones may be managed medically while others require surgery. Traditional prediction methods based on radiographic appearance are complex and subjective. The Minnesota Urolith Center and Hill’s Pet Nutrition developed an AI algorithm, CALCurad, embedded in the MN Urolith mobile app to predict struvite composition from abdominal radiographs. This study evaluated the app’s diagnostic accuracy in clinical cases compared to gold-standard infrared spectroscopy.

Methods
A retrospective multicenter analysis was performed on 139 dogs diagnosed with urolithiasis who had undergone both abdominal radiography and urolith composition analysis. Radiographs were photographed using a smartphone following specific guidelines. The AI model (ResNet-18 based), trained on 8,000 labeled canine radiographs, categorized stones as “struvite” or “non-struvite.” Cases were excluded if radiographs were of poor quality. CALCurad predictions were compared to quantitative mineral analysis, with ≥70% struvite content defining positive agreement.

Results
Among 139 dogs, 58 (41.7%) had struvite and 60 (43.2%) had calcium oxalate stones. The AI achieved 81.3% overall detection accuracy, with sensitivity of 81%, specificity of 81.5%, positive predictive value of 75.8%, and negative predictive value of 85.7%. Misclassification occurred primarily in mixed urolith cases. The app correctly predicted struvite composition in most cases suitable for medical dissolution. Image quality, particularly contrast and absence of superimposition, was critical to application performance.

Limitations
Exclusion of 23 cases due to imaging quality limits generalizability. The model performed less well on mixed-composition uroliths and was not trained to handle varied compositions. Geographic variation in urolith types may have influenced results, and no comparison was made with clinician-based prediction accuracy.

Conclusions
CALCurad offers a fast, accessible, and fairly accurate method to predict struvite uroliths from canine abdominal radiographs. The tool has clinical utility in deciding between medical versus surgical treatment for urolithiasis, though caution is needed in cases with suspected mixed stones. Further development to expand diagnostic categories and handle varied compositions could enhance clinical applicability.

Disclaimer: The summary generated in this email was created by an AI large language model. Therefore errors may occur. Reading the article is the best way to understand the scholarly work. The figure presented here remains the property of the publisher or author and subject to the applicable copyright agreement. It is reproduced here as an educational work. If you have any questions or concerns about the work presented here, reply to this email.