DeepTag

DeepTag infers standard medical diagnostic codes from unstructured veterinary clinical notes to enable standardized phenotyping and support cross-species translational research.

Key Features:

• Automatic Inference of Diagnostic Codes: Uses a deep learning algorithm to infer diagnostic codes directly from free-text clinical notes.
• Training on Curated Dataset: Trained on 112,558 manually annotated veterinary notes labeled by domain experts.
• Advanced Multitask LSTM Architecture: Employs a multitask Long Short-Term Memory (LSTM) network with a hierarchical objective function to capture semantic relationships between diseases.
• Abstention Mechanism: Incorporates an abstention mechanism to defer uncertain cases to human experts.
• Cross-Hospital Generalization: Demonstrates robustness for inferring disease codes from notes originating in different clinical environments.
• Minimal Preprocessing Requirement: Enables annotation of a broad spectrum of clinical diagnoses with minimal preprocessing.

Scientific Applications:

• Data Standardization and Codification: Converts unstructured veterinary clinical notes into standardized diagnostic codes for structured analyses.
• Patient Cohort Identification: Facilitates accurate identification of patient cohorts based on specific diagnostic criteria.
• Cross-Species Translational Research: Supports comparative studies that draw parallels between human and animal health conditions.
• Improved Clinical Data Analysis and Resource Allocation: Enables more efficient analysis of clinical data and supports decision-making for resource allocation within veterinary practices.

Methodology:

Trained a deep learning model on a corpus of 112,558 expert-annotated clinical notes using a multitask LSTM architecture with a hierarchical objective function and an abstention mechanism for uncertain predictions.