Automated Organ-Level Classification of Free-Text Pathology Reports to Support a Radiology Follow-up Tracking Engine.

Area of Science:

• Medical informatics
• Computational pathology
• Machine learning in healthcare

Background:

• Semistructured pathology reports contain valuable data for patient care.
• Automated analysis of these reports can enhance clinical workflows.
• Integrating pathology findings with imaging follow-up is crucial for patient management.

Purpose of the Study:

• To assess machine learning algorithms for classifying pathology reports at the organ level.
• To develop an automated system for surgical pathology monitoring.
• To integrate this into an imaging recommendation follow-up engine.

Main Methods:

• Retrospective analysis of 2013 abdominal pathology reports.
• Organ-level classification (liver, kidneys, pancreas/adrenal glands, none) by human annotators.
• Comparison of machine learning models: string matching, random forests, XGBoost, SVM, CNN, and LSTM.

Main Results:

• Neural networks (CNN, LSTM) achieved high F1 scores (96.3%, 96.7%) for organ classification.
• These models demonstrated generalizable feature encoding, classifying unseen report formats accurately.
• Network decision-making processes were visualized, showing human-like heuristics.

Conclusions:

• Neural network approaches are highly effective for organ-level pathology report classification.
• Feasibility of using these methods in automated tracking systems is demonstrated.
• This facilitates enhanced surgical pathology monitoring and automated imaging follow-up.