Enhancing AI-based decision support system with automatic brain tumor segmentation for EGFR mutation classification.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Glioblastoma (GBM) has a poor prognosis, worsened by epidermal growth factor receptor (EGFR) mutations.
• Invasive biopsies are currently required for GBM characterization and EGFR mutation status.
• There is a need for non-invasive methods to support clinical decision-making in GBM management.

Purpose of the Study:

• To develop and validate a fully automated MRI-based decision-support system (DSS) for GBM segmentation and EGFR status classification.
• To reduce reliance on invasive biopsy procedures for GBM diagnosis and molecular subtyping.
• To provide a tool for faster and more accurate EGFR prediction in GBM patients.

Main Methods:

• A novel segmentation module (UNet SI) was developed, fusing multiresolution shearlet and CNN features for detailed GBM segmentation.
• An Inception ResNet-v2 classifier was employed for EGFR status classification using segmented tumor masks.
• The system was validated on a cohort of 98 contrast-enhanced T1-weighted MRI scans and externally on the BraTS 2019 dataset.

Main Results:

• The UNet SI segmentation module achieved high performance metrics (Dice 0.873, Jaccard 0.853) on the internal cohort.
• The EGFR classification component demonstrated excellent accuracy (0.960), precision (1.000), recall (0.871), and AUC (0.94).
• The system achieved rapid inference times (≤0.18 s/slice) and surpassed state-of-the-art results.

Conclusions:

• The developed MRI-based DSS effectively segments GBM and classifies EGFR status with high accuracy.
• This automated system offers a non-invasive alternative to biopsy, potentially improving clinical workflow and patient management.
• The DSS shows promise for integration into routine clinical practice for enhanced glioblastoma care.