Radiomic Phenotypes Distinguish Atypical Teratoid/Rhabdoid Tumors from Medulloblastoma

Area of Science:

• Neuroradiology
• Medical Imaging Analysis
• Machine Learning in Oncology

Background:

• Atypical teratoid/rhabdoid tumors (AT/RT) and medulloblastomas share similar imaging and histologic features.
• These tumors have significantly different prognoses, necessitating accurate differentiation.

Purpose of the Study:

• To investigate if MR imaging-based radiomic phenotypes can distinguish between AT/RT and medulloblastomas.
• To evaluate the diagnostic performance of radiomic features in differentiating these pediatric brain tumors.

Main Methods:

• Retrospective analysis of T2-weighted and gadolinium-enhanced T1-weighted MR images from 48 AT/RT and 96 medulloblastomas.
• Application of sparse regression to 1800 Imaging Biomarker Standardization Initiative (IBSI)-based radiomic features, reducing the set to 6 key features (1 from T1WI, 5 from T2WI).
• Evaluation of 6 candidate classifier models, including logistic regression, using a holdout test set.

Main Results:

• Sparse regression identified 6 critical IBSI radiomic features (voxel intensity, inverse difference moment, kurtosis from T2WI were most important).
• Logistic regression achieved the highest performance with an Area Under the Curve (AUC) of 0.86.
• The model demonstrated high sensitivity (0.80), specificity (0.82), accuracy (0.81), and F1 score (0.85).

Conclusions:

• Quantitative radiomic signatures derived from MR imaging can effectively differentiate AT/RT from medulloblastomas.
• This radiomics approach offers high prediction performance across various machine learning models.
• Preoperative differentiation using this technique can significantly guide therapeutic strategies and patient care discussions.