Glioblastoma and radiotherapy: A multicenter AI study for Survival Predictions from MRI (GRASP study)

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Glioblastoma survival prediction post-radiotherapy is crucial for treatment planning.
• Assessing outcomes after radiotherapy and adjuvant temozolomide is essential.
• Current prognostic methods may not fully leverage post-treatment imaging data.

Purpose of the Study:

• To develop and validate a deep learning model for predicting glioblastoma survival at 8 months post-radiotherapy.
• To evaluate the efficacy of using early post-treatment brain MRI scans for survival prediction.
• To establish a prognostic biomarker for glioblastoma patients.

Main Methods:

• Collected retrospective and prospective data from 206 glioblastoma patients across 11 UK centers.
• Trained deep learning models using T2-weighted and contrast-enhanced T1-weighted MRI sequences.
• Incorporated nonimaging data (demographics, MGMT status, treatment) and investigated pretrained models.

Main Results:

• The deep learning imaging model significantly outperformed nonimaging models in predicting survival.
• Models achieved high areas under the receiver-operating characteristic curve (AUC) on internal and external validation sets.
• Pretraining the imaging model with large datasets of brain MRIs substantially improved performance.

Conclusions:

• A deep learning model utilizing post-radiotherapy MRI scans reliably predicts glioblastoma survival.
• This model acts as a prognostic biomarker, identifying patients needing intensified or alternative treatment strategies.
• The findings support the use of AI-driven imaging analysis for personalized glioblastoma management.