Spatial-temporal recurrence patterns of grade 4 glioma using deep learning integrated multiparametric MRI and...

Area of Science:

• Neuro-oncology
• Radiotherapy
• Medical Imaging Analysis

Background:

• Grade 4 glioma recurrence is a significant challenge, often leading to poor prognosis.
• Current radiotherapy guidelines use uniform margins, potentially suboptimal for individual patient needs.
• Predicting tumor recurrence and progression is crucial for effective treatment planning.

Purpose of the Study:

• To investigate if molecular and phenotypic features at diagnosis can predict spatiotemporal recurrence in grade 4 gliomas.
• To develop a risk stratification model for personalized radiotherapy margin determination.

Main Methods:

• Deep learning segmentation of 390 grade 4 gliomas MRIs to identify contrast-enhancing (CEcore) and non-enhancing (NE) tumor compartments.
• Quantification of primary CEcore/NE volume ratio, tumor trajectories, and progression using Hausdorff-95 distance.
• Machine learning and Cox regression analysis correlating imaging features with survival, MGMT/IDH status, age, and tumor location.

Main Results:

• A primary CEcore/NE volume ratio ≤0.324 was independently prognostic for improved overall survival (adjusted HR=0.56, p=0.006).
• This ratio provided a survival benefit of 4.3 months in IDH-wildtype, MGMT-unmethylated patients (median 17.6 vs 13.3 months).
• Tumors originating in the occipital lobe showed a higher propensity for migration (57.1%) and shorter time to progression (adjusted HR=1.90, p=0.026).

Conclusions:

• Primary tumor composition, specifically the CEcore/NE volume ratio, is a significant independent prognostic factor in grade 4 glioma.
• Molecular-demographic-anatomical risk stratification can inform personalized radiotherapy margin determination.
• Identifying high-risk tumor locations like the occipital lobe aids in predicting aggressive progression patterns.