Evaluation of radiosensitivity for high grade gliomas patients using a multi-temporal graph convolutional networks

Area of Science:

• Neuro-oncology
• Radiotherapy
• Medical Imaging

Background:

• Assessing radiotherapy efficacy in high-grade gliomas (HGGs) is difficult due to pseudo-progression and radionecrosis.
• Accurate prediction of treatment response is crucial for optimizing patient outcomes.

Purpose of the Study:

• To develop and validate a novel deep learning model for predicting radiotherapy sensitivity in HGG patients.
• To leverage multi-temporal MR image features for improved prediction accuracy.

Main Methods:

• A directed-graph multi-temporal graph convolution network (D-MTGCN) was developed using MR image features from multiple time points.
• The study included 120 HGG patients for training/validation and 29 from multicenter data for external testing.
• Radiosensitivity was defined by recurrence within one year post-radiation.

Main Results:

• The D-MTGCN achieved an area under the curve (AUC) of 0.98 and an accuracy (ACC) of 0.95.
• The model demonstrated significantly higher predictive efficacy compared to clinical models and RECIST 2.0 criteria (p<0.01).
• D-MTGCN outperformed support vector machine (ACC=0.87) and STGCN (ACC=0.93) using initial time point data.

Conclusions:

• The D-MTGCN model accurately predicts radiotherapy sensitivity and outcomes in HGG patients using short-term MRI sequences.
• This AI-driven tool can assist clinicians in making timely and precise treatment decisions for HGG.
• The findings highlight the potential of advanced graph neural networks in neuro-oncology treatment planning.