An MRI-based deep transfer learning radiomics nomogram for predicting meningioma grade

Area of Science:

• Neurosurgery
• Radiology
• Artificial Intelligence in Medicine

Background:

• Accurate grading of meningiomas is crucial for treatment planning and prognosis.
• Traditional methods for grading meningiomas often rely on histopathology, which can be invasive.
• Developing non-invasive predictive models using imaging features is highly desirable.

Purpose of the Study:

• To develop and validate a nomogram for predicting meningioma grade.
• To integrate clinical, radiomics, and deep transfer learning (DTL) features for enhanced prediction.
• To compare the predictive performance of the integrated model against models using individual feature types.

Main Methods:

• A retrospective study utilizing enhanced T1-weighted imaging (WI) from 340 meningiomas (training set) and 102 (test set).
• Clinical features, radiomics features, and DTL features were extracted and analyzed.
• A deep transfer learning radiomics (DTLR) nomogram was constructed using selected features.
• Model performance was evaluated using receiver operating characteristic (ROC) and decision curve analysis (DCA).

Main Results:

• The DTLR nomogram achieved the highest Area Under the Curve (AUC) of 0.866 in the test set, outperforming the clinical model (AUC=0.788).
• The DTLR nomogram demonstrated superior net benefit in decision curve analysis.
• While the DTLR nomogram showed strong predictive value, further improvements are suggested for future studies.

Conclusions:

• The DTLR nomogram offers a valuable, non-invasive tool for predicting meningioma grade.
• Integrating DTL with radiomics and clinical data enhances predictive accuracy.
• This approach holds potential for improving clinical decision-making in meningioma management.