DeepTag: inferring diagnoses from veterinary clinical notes

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Biomedical Informatics

Background:

• Veterinary clinical records are valuable but underutilized due to the lack of standardized diagnostic coding.
• Free-text notes dominate veterinary records, posing challenges for data analysis and research.
• Accurate coding is crucial for improving patient care and enabling cross-species translational research.

Purpose of the Study:

• To develop an automated system for inferring diagnostic codes from veterinary free-text notes.
• To reduce the manual coding burden on veterinary clinicians.
• To facilitate large-scale data analysis and translational research in veterinary medicine.

Main Methods:

• Developed DeepTag, a deep learning algorithm utilizing a multitask LSTM architecture.
• Trained DeepTag on a curated dataset of 112,558 expert-annotated veterinary notes.
• Incorporated a hierarchical objective to capture semantic disease structures and a human-machine collaboration mechanism for uncertain cases.

Main Results:

• DeepTag accurately infers disease codes from veterinary free-text notes.
• The algorithm demonstrates strong performance even in cross-hospital settings.
• Automated annotation is achieved across a broad range of diagnoses with minimal preprocessing.

Conclusions:

• DeepTag effectively automates the diagnostic coding process in veterinary medicine.
• The system enhances the utility of clinical records for patient care and research.
• The underlying framework has potential applications in other medical domains lacking coding resources.