Related Experiment Video
Updated: Jun 21, 2026
17:06
Co-analysis of Brain Structure and Function using fMRI and Diffusion-weighted Imaging
Published on: November 8, 2012
26.3K
Combining Multi-Shell Diffusion with Conventional MRI Improves Molecular Diagnosis of Diffuse Gliomas with Deep
Golestan Karami1,2, Riccardo Pascuzzo3, Matteo Figini4
1Department of Neuroscience, Imaging and Clinical Sciences, Gabriele D'Annunzio University, 66100 Chieti, Italy.
Cancers
|January 21, 2023
Summary
Combining conventional MRI (cMRI) and multi-shell diffusion MRI (dMRI) improves deep learning models for predicting glioma molecular subtypes. This integrated approach enhances diagnostic accuracy for IDH-mutation and 1p/19q-codeletion status, crucial for treatment decisions.
Area of Science:
- Neuro-oncology
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- The World Health Organization (WHO) classification emphasizes molecular diagnosis for adult diffuse gliomas, impacting prognosis and treatment.
- Non-invasive methods like MRI are crucial for pre-surgical molecular subtyping of gliomas.
- Deep learning (DL) models primarily use conventional MRI (cMRI), but multi-shell diffusion MRI (dMRI) may offer complementary information.
Purpose of the Study:
- To evaluate the added value of combining cMRI and multi-shell dMRI in DL-based models for glioma molecular subtyping.
- To assess the performance of integrated imaging modalities in predicting IDH-mutation, 1p/19q-codeletion, and WHO 2021 molecular subtypes.
Main Methods:
- A deep residual neural network model was employed for classification tasks.
- The model was trained and validated using nested cross-validation on a dataset of 146 patients with gliomas (WHO grades 2-4).
- Pre-operative cMRI, multi-shell dMRI, and their combination were used as input features.
Main Results:
- Combining cMRI and multi-shell dMRI achieved the highest accuracy in predicting IDH-mutation (75 ± 9% in lower grades, 81 ± 5% overall) and 1p/19q-codeletion (72 ± 4% in lower grades).
- The integrated approach outperformed individual modalities for IDH-mutation and 1p/19q-codeletion prediction.
- For WHO 2021 molecular subtype diagnosis, the combined approach yielded 60 ± 5% accuracy, surpassing cMRI (57 ± 8%) and dMRI (56 ± 7%) alone.
Conclusions:
- Integrating cMRI and multi-shell dMRI in DL models significantly enhances the accuracy of predicting IDH and 1p/19q status in gliomas.
- The combined imaging approach offers superior performance for molecular subtyping compared to using either modality independently.
- This multimodal strategy holds promise for improving non-invasive pre-surgical glioma diagnosis and guiding treatment decisions.
