(OT07) Deep Learning Artificial-intelligence (AI) Based Approach for Evaluation of Canine and Feline Dermatologic Cytology Samples

Abstract:

Background:
Dermatology is a top reason for veterinary visits for dogs/cats, and diagnostic testing is a vital part of case management. A digital tool for rapid, consistent evaluation of dermatologic cytology samples is currently lacking.

Hypothesis/

Objectives:
The VETSCAN IMAGYST AI Dermatology application will accurately identify inflammatory leukocytes, bacteria, and yeast on digital slides in agreement with ACVP-boarded (ACVP) clinical pathologists (CPs).

Animals:
No animals were utilized in this study. Slide specimens from general practice veterinarians and boarded dermatologists represented a mix of client-owned canine and feline samples.

Methods:
Slides (ear swabs, skin swabs, skin impressions; n=218) were prepared in-clinic, quick Romanowsky stained, coverslipped, and scanned by Grundium Ocus 40. Three ACVP CPs independently scored slides as positive or negative for presence or absence of object classes: inflammatory leukocytes (neutrophils, macrophages, lymphocytes, eosinophils), bacteria (cocci, bacilli), and yeast (Malassezia spp.). The VETSCAN IMAGYST AI Dermatology application analyzed these same 218 slides for presence/absence of each object class.


Results:
The VETSCAN IMAGYST AI Dermatology application reliably identified bacteria (cocci and rods), yeast, and inflammatory leukocytes. Sensitivity for object classes ranged from 77% - 92%; specificity ranged from 73% - 92%.

Conclusions and clinical importance:
Performance of the VETSCAN IMAGYST AI Dermatology application was comparable to expert ACVP CPs. In-house utilization of the VETSCAN IMAGYST AI Dermatology application can provide a diagnostic tool for consistently identifying inflammatory cells, bacteria, and yeast.