MRI-Based Radiomics for Outcome Stratification in Pediatric Osteosarcoma

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Osteosarcoma (OS) is a rare and aggressive bone cancer in children and adolescents with a low survival rate.
• Predicting clinical outcomes in pediatric OS is challenging due to tumor heterogeneity and case complexity.

Purpose of the Study:

• To enhance predictions of progressive disease, therapy response, relapse, and survival in pediatric osteosarcoma (OS).
• To leverage MRI-based radiomics and machine learning for improved outcome prediction in pediatric OS.

Main Methods:

• Utilized pre-treatment contrast-enhanced T1-weighted MR scans from 63 pediatric OS patients and 9 external validation cases.
• Employed MRI-based radiomics with three segmentation strategies (whole-tumor, tumor sampling, bone/soft tissue).
• Integrated radiomic features with clinical data to predict OS clinical outcomes.

Main Results:

• Bone/soft tissue segmentation models generally yielded the best predictive performance.
• Incorporating specific clinical features enhanced predictions for therapy response and mortality.
• Top models achieved high validation ROC AUC (0.94-1.0) and testing ROC AUC (0.63-1.0).

Conclusions:

• MRI-based radiomics, particularly with multi-region segmentation, shows significant predictive power for pediatric OS outcomes.
• This approach offers a promising tool for improving clinical outcome predictions in pediatric osteosarcoma.
• Radiomic features significantly outperformed models relying solely on clinical data.