Early Prediction of Planning Adaptation Requirement Indication Due to Volumetric Alterations in Head and Neck Cancer...

Area of Science:

• Radiotherapy
• Medical Imaging
• Machine Learning

Background:

• Radiation therapy (RT) tumor response can alter coverage and overdose organs at risk.
• Early prediction of volumetric changes can reduce RT-related adverse effects.
• Machine learning on CBCT radiomics for RT volume prediction is underdeveloped.

Purpose of the Study:

• To develop and validate a machine learning framework for early prediction of significant volumetric changes during RT.
• To utilize radiomic features from CBCT images for predicting tumor and organ response.

Main Methods:

• Collected weekly CBCT images from 40 head and neck cancer patients undergoing RT.
• Extracted 104 delta-radiomics features from Clinical Target Volume (CTV) and Parotid Glands (PG) regions.
• Employed feature selection and classification for predicting volumetric alterations.

Main Results:

• Achieved 0.90 classification accuracy in predicting volumetric changes.
• Identified a small subset of discriminative features from the first week of RT.
• Analyzed selected features for their impact on anatomical and tumor response dynamics.

Conclusions:

• Machine learning offers a promising approach for rapid and reliable prediction of volumetric deviations during RT.
• Exploiting hidden patterns in anatomical characteristics via ML can enhance treatment safety and efficacy.
• Early identification of volume changes can optimize radiation targeting and reduce toxicity.