MRI-based deep transfer learning models for predicting progesterone receptor expression in meningioma

Area of Science:

• Neuro-oncology
• Biomarker Discovery
• Artificial Intelligence in Medicine

Background:

• Progesterone receptor (PR) expression is a critical biomarker in meningiomas, impacting tumor behavior and treatment strategies.
• Accurate prediction of PR status is essential for personalized patient management.

Purpose of the Study:

• To develop and validate a deep transfer learning (DTL) model for predicting PR expression in meningiomas.
• To assess the performance of different machine learning models in classifying PR status.

Main Methods:

• Utilized a dataset of 307 meningioma patients (173 PR-positive, 134 PR-negative).
• Extracted DTL features using a fine-tuned ResNet 50 model, followed by feature selection using ICC, Spearman correlation, and LASSO.
• Developed predictive models using logistic regression (LR), support vector machine (SVM), and naive Bayes, evaluating performance with ROC analysis, AUC, accuracy, sensitivity, and specificity.

Main Results:

• Extracted 2048 DTL features, selecting 35 for model construction.
• Achieved AUC values of 0.819 (LR), 0.83 (Naive Bayes), and 0.842 (SVM) in the test set.
• The SVM model demonstrated superior predictive performance and clinical utility, as indicated by decision curve analysis.

Conclusions:

• The developed SVM model, based on DTL features, effectively predicts PR expression in meningiomas.
• This AI-driven approach offers a promising and non-invasive tool for meningioma evaluation.