Magnetic Resonance Imaging Radiomics Predicts Histological Response to Neoadjuvant Chemotherapy in Localized High-grade Osteosarcoma of the Extremities.

RATIONALE AND OBJECTIVES

Research involving radiomics models based on magnetic resonance imaging (MRI) has mainly used radiomics features derived from a single MRI sequence at a single time point to develop predictive models. This study aimed to construct radiomics models based on before and after neoadjuvant chemotherapy (NAC) MRI for predicting the histological response to NAC in patients with high-grade osteosarcoma.

MATERIALS AND METHODS

We included 109 patients with localized high-grade osteosarcomas of the extremities, who underwent pre- and post-NAC MRI examinations, from which radiomics features were extracted. According to the tumor necrosis rate, all patients were classified as good responders (GRs) or poor responders (PRs) and were randomly allocated into training and test sets at a 7:3 ratio. Radiomics features were extracted from T2-weighted (T2WI) and contrast-enhanced T1-weighted imaging (T1CE) of the two MRI scans to construct three models: pre-NAC, post-NAC, and combined pre-NAC and post-NAC (combined model).

RESULTS

In total, 1175 radiomics features were extracted from each sequence. Following feature selection, nine radiomics features were selected for each model to construct radiomics signatures. The radiomics signatures of the pre-NAC, post-NAC, and combined models demonstrated good predictive performance for chemotherapy response in osteosarcoma. The combined model achieved the highest areas under the receiver operating curve (AUC) values of 0.999 and 0.915 in the training and test sets, respectively. The AUCs of the post-NAC model were higher than those of the pre-NAC model.

CONCLUSION

MRI-based radiomics models demonstrate excellent performance in predicting the histological response to NAC in patients with high-grade osteosarcoma.