(HM03) Diagnostic Assessment of Point-of-care Scanning System Integrated with Deep-learning Algorithms for Canine/feline Blood Film Evaluation.

Abstract:

Background: Comprehensive hematologic assessments in canines and felines involve both quantitative (cell counts) and qualitative (blood smear) analyses. Analyzing blood smears pose challenges due to technique variations, training disparities, workflow complexities, and time constraints impeding routine blood smears review.

Objectives: Assess the performance of the Vetscan Imagyst® (VS-I) Blood Smear application, an artificial intelligence algorithm designed for hematologic analysis and compare it to ACVP board-certified clinical pathologists (CPs). Objectives included: determination of accuracy of monolayer detection, WBC estimate, WBC differential, polychromatophil count, nucleated RBC count, and platelet estimate.

Method: Blood smears (119 total, 59 dogs, 60 cats) were retrospectively collected from Zoetis Reference Labs. The Grundium Ocus 40 was used to scan all slides. A randomized 2 out of 4 CPs and the VS-I blood smear algorithm evaluated the samples. The agreement between VS-I and the CPs was assessed.

Results: The percentage of samples where VS-I was within 99% prediction interval with the CPs for each white blood cell class differential ranged from 93.2%-100% for dogs and 88.3%-100% for cats (Table 1). The number of samples where VS-I was within 95% prediction interval with the CPs for each cell class estimated number ranged from 69.5 – 95.0% for dogs and 76.3 – 95.0% for cats. (Table 2). Conclusion/Significance The VS-I Blood Smear application demonstrated strong performance with results comparable to ACVP-board-certified clinical pathologists hematologic assessments making it a tool for utilization by veterinarians to obtain a comprehensive CBC.