GlioMODA: Robust glioma segmentation in clinical routine

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Precise glioma segmentation in MRI is crucial for diagnosis, treatment planning, and research.
• Current deep learning methods often require complete MRI protocols, which are not always available in clinical practice.
• This study introduces GlioMODA, a deep learning framework for automated glioma segmentation that performs consistently across varied and incomplete MRI protocols.

Purpose of the Study:

• To develop and evaluate GlioMODA, a robust deep learning framework for automated glioma segmentation.
• To assess GlioMODA's performance consistency across diverse and incomplete MRI protocols.
• To enable reliable quantitative volumetry compatible with clinical standards.

Main Methods:

• GlioMODA was trained and validated on the BraTS 2021 dataset (1251 training, 219 testing cases).
• Performance was systematically assessed across 11 MRI protocol combinations.
• Segmentation accuracy was evaluated using Dice similarity coefficients (DSC) and panoptic quality metrics, with volumetric accuracy benchmarked against manual ground truth.

Main Results:

• GlioMODA demonstrated state-of-the-art segmentation accuracy across tumor subregions, even with incomplete or heterogeneous MRI protocols.
• Protocols including T1-weighted contrast-enhanced and T2-FLAIR sequences showed no statistically significant volumetric differences compared to manual ground truth for enhancing tumor and whole tumor.
• Omitting either T1-contrast or T2-FLAIR sequences resulted in substantial and significant volumetric errors.

Conclusions:

• GlioMODA enables reliable, automated glioma segmentation using a simplified 2-sequence protocol (T1-contrast + T2-FLAIR).
• This facilitates clinical workflow optimization and broader implementation of quantitative volumetry.
• GlioMODA is available as an open-source Python package, promoting wider adoption and research.