Explore the scholarly record.

Abstract

Artificial intelligence (AI) is growing in veterinary medicine, but many tools work as ’black boxes’ with limited clinical validation. This study aimed to design, develop and validate, under veterinary leadership, a knowledge-based expert system to support differential diagnosis in dogs and cats: the Veterinary Differential Diagnosis System (VDDS, SDDV by its Spanish acronym). Knowledge from a reference textbook on small animal internal medicine was extracted, formalised and stored in XML. The final knowledge base included 191 lists with 2,409 normalised diagnoses. This system combines the lists of differential diagnoses of several clinical signs and clinicopathological abnormalities and groups the diagnoses by priority, using terminology normalisation and a single-inheritance. The VDDS was evaluated in a retrospective methodological validation with 130 real cases (104 dogs, 26 cats) from a university veterinary hospital, in two paired scenarios: without and with the results of complementary tests. The validated diagnosis was retrieved in 70.8% of cases without tests and in 98.5% with tests (difference, +28.2 percentage points; P < 0.001). The tie-aware expected Top-3 increased from 29.6% to 53.3% and the expected mean reciprocal rank from 0.255 to 0.474. With tests, coverage was ≥ 83.3% in all subspecialties. The VDDS shows that a transparent and traceable expert system for small animal internal medicine is feasible. It is designed to support, not replace, clinical reasoning. Prospective, multicentre studies are needed to evaluate its clinical utility.