Voxel-Based Dose-Toxicity Modeling for Predicting Post-Radiotherapy Toxicity: A Critical Perspective.

Area of Science:

• Radiotherapy
• Medical Imaging Analysis
• Biostatistics

Background:

• Voxel-based analysis (VBA), or image-based data mining (IBDM), is emerging for radiotherapy dose-toxicity modeling.
• It offers insights into localized organ subregions linked to toxicity.

Purpose of the Study:

• Critically examine the role and limitations of VBA/IBDM in post-radiotherapy toxicity assessment.
• Identify key challenges hindering clinical validation and application.

Main Methods:

• Perspective paper based on prior studies and practical experience.
• Analysis of seven key limitations in current VBA/IBDM methodologies.

Main Results:

• Identified limitations include lack of clinical validation, statistical method dependence, insensitivity to dose scaling, and issues with statistical testing.
• Other limitations involve misapplication of permutation testing, reliance on dose-only predictors, and misinterpretation of associations as causal.

Conclusions:

• Current univariable VBA/IBDM are hypothesis-generating, not clinical decision tools, due to validation and methodological issues.
• Advancing personalized radiotherapy requires rigorous validation, multivariable/causal modeling, and integration of clinical/genomic data for improved precision and utility.