Differentiation between Germinoma and Craniopharyngioma Using Radiomics-Based Machine Learning

Area of Science:

• Neuro-oncology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Germinoma and craniopharyngioma (CP) are rare anterior skull base tumors with overlapping clinical and imaging features.
• Accurate preoperative diagnosis is critical due to differing treatment strategies and patient outcomes.

Purpose of the Study:

• To develop and evaluate radiomics-based machine learning models for the differential diagnosis of germinoma and CP.
• To compare the diagnostic performance of models using contrast-enhanced T1WI and T2WI MRI sequences.

Main Methods:

• Retrospective analysis of 107 patients (44 germinoma, 63 CP).
• Extraction of radiomic features from contrast-enhanced T1WI and T2WI MRI sequences.
• Development of diagnostic models using combinations of feature selection methods and classifiers, evaluated by Area Under the Curve (AUC).

Main Results:

• Optimal models for contrast-enhanced T1WI achieved an AUC of 0.91 (RF_S + RF_C and LASSO + LDA).
• Optimal models for T2WI achieved an AUC of 0.88 (DC + LDA and LASSO + LDA).
• Radiomics-based models demonstrated reliable diagnostic performance in the validation cohort.

Conclusions:

• Radiomics combined with machine learning offers a potentially valuable radiological tool for presurgical differential diagnosis of germinoma and CP.
• The findings support the use of radiomics as a non-invasive method to improve diagnostic accuracy for these challenging tumors.