Multi-View Data Integration Methods for Radiotherapy Structure Name Standardization.

Area of Science:

• Medical Physics
• Oncology
• Data Science

Background:

• Standardizing radiotherapy structure names is crucial for data-driven personalized treatment.
• Existing methods often rely on a single data type, limiting model performance.
• Radiotherapy structures involve diverse data: text labels, geometric information, and Dose-Volume Histograms (DVH).

Purpose of the Study:

• To develop and evaluate novel machine learning approaches for standardizing radiotherapy structure names.
• To investigate the effectiveness of integrating complementary data views (textual and geometric) for improved standardization.
• To enhance the foundation for big data analytics in radiation oncology.

Main Methods:

• Two integration methods were developed: intermediate and late integration.
• These methods combined physician-given textual structure names with geometric structure information.
• The study utilized a large dataset from prostate and lung cancer patients across multiple radiotherapy centers.

Main Results:

• The intermediate integration approach demonstrated superior performance compared to single-view models.
• Late integration showed performance comparable to single-view methods.
• The findings highlight the benefit of multi-view data integration for structure name standardization.

Conclusions:

• Combining different data views significantly improves machine learning models for radiotherapy structure name standardization.
• This approach facilitates more robust big data analytics in radiation oncology.
• The developed methods pave the way for more accurate and personalized radiotherapy treatments.