Gradient-Based Radiomics for Outcome Prediction and Decision-Making in PULSAR: A Preliminary Study

Area of Science:

• Neuro-oncology
• Radiotherapy
• Medical Imaging

Background:

• Personalized ultrafractionated stereotactic adaptive radiation therapy (PULSAR) offers high-dose radiation with adaptive treatment based on patient response.
• Current PULSAR adaptation relies on physician experience and tumor size, lacking a data-driven approach for improved outcome prediction.

Purpose of the Study:

• To develop and validate a data-driven radiomics approach for predicting treatment response in PULSAR therapy.
• To enhance decision-making and personalize treatment planning in neuro-oncology.

Main Methods:

• Analysis of 69 lesions from 39 patients undergoing PULSAR treatment.
• Extraction of gradient-based features (magnitude, radial gradient, radial deviation) from intratumoral and peritumoral regions.
• Development of an ensemble feature selection (EFS) model using support vector machines and validated on a separate cohort.

Main Results:

• The EFS model accurately predicted tumor volume reduction exceeding 20% at 3 months post-radiation.
• Features from octant subregions showed superior prediction compared to core or margin features.
• The gradient-based approach outperformed conventional radiomics and demonstrated generalizability across different cohorts.

Conclusions:

• The gradient-based radiomics approach significantly enhances treatment response prediction in PULSAR therapy.
• This method shows potential as a robust tool for personalized treatment planning in neuro-oncology.
• The approach is applicable to both photon and particle therapies.