Supervised line attention for tumor attribute classification from pathology reports: Higher performance with less...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Computational Pathology

Background:

• Pathology report annotation is time-consuming and expensive.
• Accurate classification of tumor attributes is crucial for cancer diagnosis and treatment.
• Existing machine learning methods struggle with limited annotated data.

Purpose of the Study:

• To develop an accurate machine learning system for classifying tumor attributes from cancer pathology reports.
• To address the challenge of limited annotated data in pathology report analysis.
• To leverage enriched labeling schemes for improved classification performance.

Main Methods:

• Utilized a dataset of 500 colon and kidney cancer pathology reports.
• Developed a hierarchical machine learning model using enriched annotations (label + location).
• Compared the model's performance against state-of-the-art methods for attribute classification.

Main Results:

• The hierarchical method consistently outperformed state-of-the-art approaches across different cancer types and training set sizes.
• Achieved comparable performance to existing methods using approximately 50% less labeled data.
• Demonstrated the effectiveness of enriched annotations in improving sample efficiency.

Conclusions:

• Enriched document annotations significantly enhance the sample efficiency of machine learning models for pathology report analysis.
• The proposed hierarchical method offers a more efficient and accurate approach to classifying tumor attributes.
• This approach has the potential to reduce the burden of data annotation in computational pathology.