Treatment-Aware Diffusion Probabilistic Model for Longitudinal MRI Generation and Diffuse Glioma Growth Prediction

Area of Science:

• Neuro-oncology
• Medical imaging
• Artificial intelligence in medicine

Background:

• Diffuse gliomas are challenging to model due to complex tumor-host interactions and treatment effects.
• Accurate prediction of glioma progression and treatment response is crucial for effective clinical management.

Purpose of the Study:

• To develop a novel end-to-end network for predicting future tumor masks and multi-parametric MRI in diffuse gliomas.
• To enable personalized treatment planning by simulating tumor evolution under different therapeutic strategies.

Main Methods:

• Utilized cutting-edge diffusion probabilistic models and deep-segmentation neural networks.
• Incorporated sequential multi-parametric MRI and treatment data as conditioning inputs for a generative diffusion process.
• Trained the model on real-world, longitudinal MRI data of glioma patients with time-series tumor segmentation maps.

Main Results:

• The network demonstrated high-quality generation of multi-parametric MRI with tumor masks.
• Achieved accurate time-series tumor segmentation and provided reliable uncertainty estimates for predictions.
• Generated treatment-aware MRI, facilitating realistic tumor growth simulations.

Conclusions:

• The developed AI model shows promising performance in predicting diffuse glioma growth and appearance.
• Treatment-aware MRI generation and uncertainty quantification can significantly aid clinical decision-making in neuro-oncology.
• This approach offers a powerful tool for personalized medicine in brain tumor management.