Related Experiment Videos
AI-assisted post-mortem classification of vehicle trauma in free-roaming cats using transformer-based large language
Hyunji Jo1, Ji-Su Baek2, Donghee Lee1
1Department of Small Animal Clinical Sciences, College of Veterinary Medicine, University of Florida, Gainesville, Florida, USA.
The Veterinary Quarterly
|August 3, 2026
Summary
This study developed an AI framework using large language models (LLMs) to classify veterinary forensic autopsy reports for feline deaths. The AI accurately identified vehicular trauma (VT) and other causes of death in cats.
Area of Science:
- Veterinary Forensic Pathology
- Artificial Intelligence in Forensics
- Natural Language Processing
Background:
- Determining causes of death in free-roaming cats is challenging due to complex factors and limited pattern recognition.
- Natural language processing (NLP) offers potential for automating document analysis in forensic investigations.
Purpose of the Study:
- To develop an AI framework using a transformer-based large language model (LLM).
- To classify veterinary forensic autopsy reports from feline cases.
- To identify causes of death, specifically differentiating vehicular trauma (VT).
Main Methods:
- Utilized Term Frequency-Inverse Document Frequency (TF-IDF) analysis to identify lexical patterns.
- Developed and evaluated an AI model based on Clinical-Longformer for text classification.
- Employed Integrated Gradients for model interpretability.
Main Results:
- The Clinical-Longformer model achieved high performance: 0.945 accuracy, 0.885 precision, 1.000 recall, 0.939 F1 score, and 0.950 AUC-ROC.
- VT cases were characterized by terms related to traumatic injuries.
- Non-VT cases showed a wider variety of diagnostic descriptions.
Conclusions:
- Large language models (LLMs) are feasible for analyzing veterinary forensic pathological texts.
- Domain-specific interpretation of tokens is crucial for AI model explainability in veterinary forensics.
- This AI framework supports decision-making in veterinary forensic investigations.