Veterinary Clinical Pathologist - Imagyst Development, Medical Affairs Zoetis Parsippany, NJ, United States
Abstract:
Background: Urine sediment exam is vital to a complete urinalysis. A deep learning artificial intelligence (AI) tool for identification of less common crystals and spermatozoa at the point of care is currently lacking. Hypothesis/
Objectives: Vetscan Imagyst® AI Urine Sediment v2.0 will accurately identify cystine, bilirubin, and ammonium biurate crystals, and spermatozoa in agreement with digital review by ACVP-boarded clinical pathologists (ACVP-CPs). Animals: Urine samples included those from client-owned dogs and cats undergoing urinalysis for any reason submitted to Zoetis Reference Laboratories. A total of 30 samples (canine,77%; feline, 23%) were evaluated.
Methods: Ten repeats of these 30 samples were prepped by the Vetscan Imagyst® preparation method, scanned, and converted to WSIs for review by three, blinded ACVP-CPs and Vetscan Imagyst® AI Urine Sediment v2.0 algorithm for a total of 319 scans. Some samples contained more than one element of interest and only 9 repeats were possible for one sample. Algorithm performance and agreement was calculated as compared to ACVP-CP results recorded as an average of 10, 40X fields.
Results: Accuracy, sensitivity, and specificity for urine sediment object classes ranged from 86-99%, 76-100%, and 81-99%, respectively (Table 1). Agreement between Vetscan Imagyst® AI Urine Sediment v2.0 and ACVP-CPs was good to excellent (Table 2). Conclusions and Clinical Importance: Vetscan Imagyst® AI Urine Sediment v2.0 was comparable to ACVP-CPs in the identification and semi-quantitative agreement of less common crystals and spermatozoa in urine sediment.
