Causal ML with MRI radiomics estimates neoadjuvant chemo effect in osteosarcoma: feasibility only

In a proof-of-concept study of 34 osteosarcoma patients, radiomics-based causal ML for estimating neoadjuvant chemotherapy effects showed instability; CIs crossed zero for 2 of 3 learners, limiting clinical applicability. Findings are hypothesis-generating only.

• Retrospective single-center study of 34 patients with osteosarcoma; pre-treatment T1 MRI radiomics combined with clinical variables in S-, T-, and X-learners.
• Estimated average treatment effects varied across models; confidence intervals included zero for two of three meta-learners, indicating no reliable effect could be estimated.
• Major limitations: very small sample, retrospective design, treatment-group imbalance, missing data, unvalidated causal assumptions; no generalizable predictive validity.

RadPigeon summaries are original and for information only. They are not clinical advice.