Multi-scale segmentation in GBM treatment using diffusion tensor imaging.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Glioblastoma (GBM) remains a challenging brain tumor with limited treatment outcome improvements.
• Local recurrence is a primary cause of treatment failure, necessitating enhanced surgical and radiotherapy planning.
• Current GBM segmentation for treatment planning is labor-intensive, especially with advanced MRI techniques.

Purpose of the Study:

• To develop an automated segmentation technique for glioblastoma using diffusion tensor imaging (DTI) and conventional MRI sequences.
• To improve the accuracy and efficiency of brain tumor segmentation for surgical and radiotherapy planning.
• To explore the clinical utility of DTI-derived maps (p and q) in conjunction with CNNs for precise tumor definition.

Main Methods:

• Modified a convolutional neural network (CNN) architecture for automated multi-sequence GBM segmentation.
• Utilized DTI-derived isotropic (p) and anisotropic (q) maps alongside conventional MRI sequences (T2-FLAIR, T1c).
• Employed individually defined ground truths for each MRI sequence to assess segmentation performance.

Main Results:

• Achieved high accuracy and efficiency in automated GBM segmentation using the proposed CNN model.
• Demonstrated the effectiveness of combining DTI-derived maps with conventional MRI sequences.
• Validated the model's potential for precise tumor delineation in a proof-of-concept study.

Conclusions:

• Automated segmentation of DTI-derived maps holds significant clinical utility for neurosurgery and radiotherapy.
• The developed CNN model shows promise for integrating diffusion MR images into precision radiation treatment planning.
• This approach can aid in improving patient outcomes by enhancing the accuracy of surgical and radiotherapy targets.