Point-of-care platform integrated with deep-learning, convolutional neural network algorithms effectively evaluates...

Area of Science:

• Veterinary diagnostics
• Artificial intelligence in medicine
• Hematology

Background:

• Point-of-care diagnostics are crucial for timely veterinary care.
• Automated hematology analyzers provide CBC results but may lack detailed morphological assessment.
• Deep learning algorithms offer potential for automated analysis of blood smears.

Purpose of the Study:

• To evaluate a novel veterinary multiuse platform with AI algorithms for analyzing canine and feline peripheral blood smears.
• To compare the diagnostic performance of the AI platform against board-certified clinical pathologists.

Main Methods:

• A blinded, randomized study design was employed.
• The platform utilized deep-learning, convolutional neural network algorithms for cell identification and enumeration.
• Performance was assessed through sensitivity, specificity, and agreement metrics compared to clinical pathologists.

Main Results:

• High agreement was observed for leukocyte differential counts (96.6% canine, 91.7% feline).
• Agreement for cell counts ranged from 70% to 95% across species.
• The algorithm achieved 90% sensitivity and 88% specificity for platelet clump identification and 100% agreement for polychromatophils.

Conclusions:

• The AI-powered point-of-care platform provides diagnostic results comparable to clinical pathologists.
• This technology can serve as a valuable adjunct to automated CBC results in veterinary practice.
• Veterinarians can integrate this assessment into routine in-clinic hematology analysis.