Radiomics-based causal machine learning for exploratory treatment-effect estimation of neoadjuvant chemotherapy cycle...

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Radiomics shows potential for characterizing tumor heterogeneity.
• Integrating radiomics with causal machine learning for treatment effect estimation in osteosarcoma is underexplored.

Purpose of the Study:

• To develop a proof-of-concept framework combining radiomics and causal machine learning.
• To estimate average and individual treatment effects of neoadjuvant chemotherapy intensity in osteosarcoma.

Main Methods:

• Retrospective study of 34 osteosarcoma patients.
• Extracted radiomic features from MRI and combined with clinical data.
• Applied S-Learner, T-Learner, and X-Learner for causal inference.

Main Results:

• The framework enabled estimation of population-level and individual treatment effects.
• Estimated average treatment effects varied across meta-learners, indicating instability.
• Model performance metrics indicated technical feasibility, not generalizable validity.

Conclusions:

• Demonstrated methodological feasibility of radiomics-based causal machine learning for osteosarcoma treatment effect estimation.
• Findings are hypothesis-generating due to limitations (small size, retrospective design, imbalance).
• Larger, prospective multicenter studies are needed for clinical relevance.