Can radiomics outperform radiologists in predicting Ewing sarcoma response to neoadjuvant chemotherapy from longitudinal MRI?

Objective This retrospective, multicenter study evaluated whether MRI-based radiomics can predict the histological response to neoadjuvant chemotherapy in Ewing sarcoma patients and compared their performance to that of radiological findings. Definitive histological assessment is only possible during surgery, potentially delaying therapy changes and increasing metastatic risk; therefore, a positive outcome would enable early adaptations. Methods Radiomics features were extracted from pre- and post-neoadjuvant (after first chemotherapy cycle) chemotherapy scans of patients treated from 2004 to 2020. Four machine learning models were trained using nested cross-validation to predict histological response, and their performance was compared to models trained on semantic features derived from expert radiological readings. Diagnostic performance was measured by the area under the receiver operating characteristic curve (AUROC). Results 96 patients (mean age, 19.9 ± 11.7 years; 41 women) with Ewing sarcoma were evaluated. Radiomics features related to low gray level intensity changes in T1 fat-saturated contrast-enhanced sequences showed the highest responder identification performance (AUROC, 0.62; 95% CI: 0.59, 0.65), surpassing the best radiological model, which relied on the relative change in maximum diameter of extratumoral soft tissue components (AUROC, 0.58; 95% CI: 0.56, 0.60). Radiomics models relying on pre-treatment T2-weighted data performed similarly (AUROC, 0.61; 95% CI: 0.58, 0.64), outperforming the clinical baseline (AUROC, 0.53; 95% CI: 0.51, 0.55). Conclusion MRI-based radiomics moderately identified responders to neoadjuvant chemotherapy, suggesting a role for earlier, tailored therapeutic strategies in Ewing sarcoma. Larger studies to validate these findings within a multimodal framework are warranted. Clinical relevance statement Radiomics extracted from MRI scans may aid with non-size tissue signals in early prediction of neoadjuvant chemotherapy response in Ewing sarcoma patients, potentially guiding personalized treatment adaptations, reducing delays for non-responders, and improving patient outcomes. These exploratory findings are complementary to standard size-based assessment from clinical routine, and warrant larger, prospective trials to be established.