Principal component analysis-based pattern analysis of dose-volume histograms and influence on rectal toxicity

Area of Science:

• Radiation Oncology
• Medical Physics
• Biostatistics

Background:

• Dose-volume histograms (DVHs) are crucial for assessing radiotherapy toxicity.
• Quantifying DVH shape variability in patient populations is challenging.
• Understanding DVH-toxicity relationships can improve treatment planning.

Purpose of the Study:

• To apply principal component analysis (PCA) to rectal DVHs in prostate cancer patients.
• To investigate the correlation between PCA parameters and late rectal bleeding.
• To explore PCA's utility in normal tissue complication probability modeling.

Main Methods:

• PCA applied to rectal wall DVHs of 262 prostate cancer patients treated with conformal adaptive radiotherapy.
• Eigenmodes identified to represent DVH pattern variability.
• Correlation of the first three principal components (PCs) with Grade 2+ rectal bleeding analyzed using logistic regression.

Main Results:

• The first three PCs explained 94-96% of rectal DVH shape variability.
• PC1 correlated with mean dose, PC2 with rectal volume overlap, and PC3 with maximal dose.
• Multivariate analysis showed increased bleeding probability with multiple large PCs, indicating complex DVH-toxicity relationships.

Conclusions:

• PCA effectively quantifies DVH shape variability and its correlation with toxicity.
• PCA provides insights into treatment technique-imposed DVH patterns.
• PCA can enhance normal tissue complication probability models for radiotherapy.