Deep learning enhances reliability of dynamic contrast-enhanced MRI in diffuse gliomas: bypassing post-processing and...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• Dynamic contrast-enhanced MRI (DCE-MRI) is crucial for assessing diffuse gliomas.
• Accurate pharmacokinetic (PK) parameter mapping is essential for tumor characterization and treatment planning.
• Current methods for PK parameter estimation can be limited by reliability and complexity.

Purpose of the Study:

• To develop and validate a novel deep learning model for direct estimation of PK parameter maps from DCE-MRI.
• To assess the reliability and diagnostic performance of the proposed model for glioma grading and mutation prediction.
• To evaluate the model's ability to provide uncertainty estimation for PK parameters.

Main Methods:

• A spatiotemporal probabilistic deep learning model was developed to generate synthetic PK maps (Ktrans, Vp, Ve) from DCE-MRI.
• The study included 329 patients with diffuse gliomas.
• Model performance was evaluated using Structural Similarity Index Measure (SSIM) against ground truth, intraclass correlation coefficient (ICC) for reliability, and Area Under the Receiver Operating Characteristic Curve (AUROC) for clinical validation.

Main Results:

• Synthetic PK maps demonstrated high similarity to ground truth (SSIM > 0.89).
• The deep learning model achieved significantly higher reliability (ICC = 1.00) compared to conventional methods (ICC < 0.68) for all PK parameters (p < 0.001).
• Diagnostic performance for glioma grading and IDH mutation prediction was preserved, with comparable AUROC values between synthetic and ground truth maps.

Conclusions:

• The proposed deep learning model reliably estimates PK parameter maps from DCE-MRI in diffuse gliomas.
• The model enhances imaging consistency and diagnostic performance without compromising clinical utility.
• This approach offers a promising tool for improved tumor characterization and personalized treatment strategies in neuro-oncology.