Glioblastoma Overall Survival Prediction With Vision Transformers.

Area of Science:

• Neuro-oncology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Glioblastoma is an aggressive brain tumor with poor prognosis.
• Accurate Overall Survival (OS) prediction is crucial for personalized glioblastoma treatment.
• Traditional OS prediction methods often require complex tumor segmentation.

Purpose of the Study:

• To develop a novel Artificial Intelligence (AI) approach for predicting Glioblastoma Overall Survival (OS).
• To utilize Vision Transformers (ViTs) for direct feature extraction from Magnetic Resonance Imaging (MRI) data, bypassing the need for tumor segmentation.
• To simplify the OS prediction workflow and reduce computational demands.

Main Methods:

• Employing Vision Transformers (ViTs) to analyze MRI images for feature extraction.
• Developing an AI model for direct OS prediction from raw MRI data.
• Evaluating the model on the BRATS dataset.

Main Results:

• The AI model achieved 62.5% accuracy on the test set, comparable to leading methods.
• The model demonstrated superior performance in precision, recall, and F1 score compared to existing benchmarks.
• Dataset size limitations impacted the generalization capabilities of the ViT model, a common observation in similar studies.

Conclusions:

• Vision Transformers (ViTs) are applicable for downsampled medical imaging tasks like OS prediction in glioblastoma.
• The proposed AI method offers a computationally efficient alternative to traditional segmentation-based approaches.
• This study lays the groundwork for developing advanced, segmentation-free AI models for glioblastoma survival prediction.