Radiomics Can Distinguish Pediatric Supratentorial Embryonal Tumors, High-Grade Gliomas, and Ependymomas.

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Machine learning in medicine

Background:

• Pediatric supratentorial tumors like embryonal tumors, high-grade gliomas, and ependymomas often present with overlapping histopathological and imaging features.
• Accurate differentiation of these tumor types is crucial for effective diagnosis, risk stratification, and treatment planning.

Purpose of the Study:

• To apply machine learning to MR imaging-based radiomics phenotypes for differentiating pediatric supratentorial tumor types.
• To develop accurate classifiers for distinguishing between embryonal tumors, high-grade gliomas, and ependymomas using imaging data.

Main Methods:

• A retrospective cohort of 231 pediatric patients with supratentorial tumors was analyzed.
• 900 radiomic features were extracted from T2-weighted and gadolinium-enhanced T1-weighted MR images using PyRadiomics.
• Sparse regression analysis reduced the feature set, which was then used to train and test six candidate classifier models.

Main Results:

• A classifier for embryonal tumor versus high-grade glioma achieved an AUC of 0.98 with 89% accuracy.
• A classifier for embryonal tumor versus ependymoma achieved an AUC of 0.82 with 81% accuracy.
• A classifier for high-grade glioma versus ependymoma achieved an AUC of 0.96 with 91% accuracy.

Conclusions:

• Distinct radiomic phenotypes can accurately differentiate between pediatric supratentorial embryonal tumors, high-grade gliomas, and ependymomas.
• Integrating radiomics into diagnostic algorithms can enhance the diagnosis, risk stratification, and treatment planning for these pediatric brain tumors.