MRI-Based Meningioma Firmness Classification Using an Adversarial Feature Learning Approach

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Meningioma firmness is crucial for surgical planning, but current MRI assessment methods are subjective and time-consuming.
• Machine learning offers potential for objective classification, but often relies on manual feature engineering.

Purpose of the Study:

• To develop and evaluate a novel adversarial feature learning approach for accurate meningioma firmness classification using MRI.
• To improve upon existing machine learning methods for meningioma consistency classification.

Main Methods:

• Utilized Bidirectional Generative Adversarial Network (BiGAN) for unsupervised feature extraction from MRI scans.
• Developed a depth-wise separable deep learning model to map extracted MRI features to meningioma firmness classes (firm or soft).

Main Results:

• The combined BiGAN encoder and depth-wise separable model significantly enhanced classification performance.
• The proposed model achieved a high accuracy of 94.7% and a weighted F1-score of 95.0% in meningioma firmness classification.
• Outperformed existing state-of-the-art methods in classifying meningioma consistency.

Conclusions:

• The novel adversarial feature learning approach effectively extracts discriminative MRI features for meningioma firmness classification.
• This method offers a more accurate and objective alternative to conventional subjective assessments, aiding surgical decision-making.