Combined molecular subtyping, grading, and segmentation of glioma using multi-task deep learning
Sebastian R van der Voort1, Fatih Incekara2,3, Maarten M J Wijnenga1,4
1Biomedical Imaging Group Rotterdam, Department of Radiology and Nuclear Medicine, Erasmus MC University Medical Centre Rotterdam, Rotterdam, the Netherlands.
Neuro-Oncology
|July 5, 2022
Summary
This study presents a novel AI method that simultaneously predicts glioma molecular subtype and grade while also delineating tumors from MRI scans. This approach offers a more efficient and comprehensive non-invasive tool for glioma characterization.
Area of Science:
- Neuroimaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Accurate glioma characterization is vital for clinical decisions but tumor delineation is time-consuming.
- Existing deep learning methods can predict glioma features or delineate tumors, but not both simultaneously.
- This study addresses the need for a unified, non-invasive method for comprehensive glioma assessment.
Purpose of the Study:
- To develop a single AI model capable of predicting molecular subtype and grade, and simultaneously segmenting glioma.
- To provide a non-invasive tool that reduces the time and complexity of glioma characterization.
Main Methods:
- A multi-task convolutional neural network (CNN) was developed using 3D structural preoperative MRI scans.
- The CNN was trained on 1508 glioma patients from 16 institutes and validated on an independent dataset of 240 patients from 13 institutes.
- The model predicts IDH mutation status, 1p/19q co-deletion status, and tumor grade, alongside tumor segmentation.
Main Results:
- The model achieved high performance on the independent test set with an IDH-AUC of 0.90, 1p/19q co-deletion AUC of 0.85, and grade AUC of 0.81.
- Tumor delineation achieved a mean whole tumor Dice score of 0.84.
- The method demonstrated strong generalization capabilities across diverse patient data.
Conclusions:
- A novel AI method non-invasively predicts multiple clinically relevant glioma features and provides tumor delineation.
- The high performance and generalizability in an independent dataset confirm its clinical utility.
- This unified approach paves the way for more generalized AI applications in neuro-oncology, moving beyond hyper-specialized models.
