Hybrid Lexical–Semantic AI Architecture for Automated Cancer Registry Coding for the Vet-ICD-O-Canine-1 System from Free-Text Veterinary Pathology Reports

Abstract

Simple Summary Veterinary pathology reports are usually written as free text, which makes it difficult to organize cancer diagnoses into standardized registry codes. This study evaluated a hybrid artificial intelligence approach designed to automatically assign standardized tumour morphology codes from veterinary pathology diagnoses. The method first reduced the number of possible codes using lexical similarity and then used a language model to select the most appropriate code from this smaller set. The approach achieved high agreement with expert-reviewed reference codes while substantially reducing computational requirements compared to providing the complete coding system directly to the language model. However, the system was less reliable when diagnostic descriptions lacked enough information for a specific classification. These findings support the use of artificial intelligence as a decision-support tool for veterinary cancer registries, while highlighting the continued need for expert review and validation in larger, independent datasets.