Automated meningioma detection using skull X ray images with deep learning and machine learning classifiers

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Meningioma detection often relies on advanced imaging, which can be costly and inaccessible.
• Developing cost-effective and accessible diagnostic tools for meningioma is crucial.

Purpose of the Study:

• To develop and evaluate a novel automated diagnostic tool for meningioma detection using skull X-ray images.
• To combine deep learning (EfficientNetB0) with traditional machine learning classifiers (Random Forest, XGBoost) for enhanced diagnostic performance.
• To assess the feasibility of using skull X-rays as a primary imaging modality for meningioma screening.

Main Methods:

• Retrospective collection of skull X-ray images from meningioma patients and control subjects.
• Utilized EfficientNetB0 with transfer learning and attention mechanisms as a deep learning backbone.
• Integrated extracted deep learning features with Random Forest and XGBoost classifiers for hybrid model development.
• Evaluated model performance using accuracy, sensitivity, specificity, F1-score, and AUROC, with internal and external validation.

Main Results:

• The hybrid EfficientNetB0-Random Forest model achieved high internal validation performance: 0.97 accuracy and 0.999 AUROC.
• External validation demonstrated Random Forest as the best classifier with 0.74 accuracy and 0.76 AUROC.
• Grad-CAM visualizations confirmed the model's focus on relevant cranial regions for meningioma identification.

Conclusions:

• Skull X-rays are a feasible imaging modality for automated meningioma detection.
• The proposed hybrid deep learning and machine learning approach offers a promising, cost-effective diagnostic tool.
• This method holds significant potential for improving meningioma diagnosis in resource-limited healthcare settings.