MRI-Based Deep Learning Method for Classification of IDH Mutation Status.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Isocitrate dehydrogenase (IDH) mutation status is a key prognostic factor in gliomas.
• Accurate non-invasive IDH classification is crucial for patient management.

Purpose of the Study:

• To develop and compare deep learning networks for non-invasive IDH mutation status classification in gliomas.
• To evaluate a T2w-image-only network against a multi-contrast network.

Main Methods:

• Two 2D deep learning networks (T2-net and MC-net) were developed using nnU-Net.
• Networks were trained on multi-contrast brain tumor MRI and genomic data from TCIA and EGD.
• Simultaneous tumor segmentation and IDH classification were performed.

Main Results:

• Multi-contrast network (MC-net) achieved higher accuracy (up to 92.8%) and AUC (up to 0.96) compared to T2w-image-only network (T2-net).
• Both networks were validated on over 1100 diverse held-out datasets.
• The study represents the largest to date for image-based IDH classification.

Conclusions:

• Deep learning algorithms reliably classify IDH mutation status non-invasively.
• Multi-contrast MRI analysis offers superior performance for IDH classification in gliomas.
• These findings support the clinical utility of AI-driven imaging analysis for glioma prognostication.