A Text-Image Network for Isocitrate Dehydrogenase(IDH) Mutation Status Prediction in Glioma Diagnosis Using...

Area of Science:

• Neuro-oncology
• Medical imaging
• Artificial intelligence in medicine

Background:

• Isocitrate dehydrogenase (IDH) mutation status is critical for brain glioma diagnosis and prognosis.
• Current IDH mutation detection methods are expensive and impractical for routine clinical use.
• Magnetic resonance imaging (MRI) shows a correlation with IDH mutation status, but existing machine learning models often overlook radiology report text.

Purpose of the Study:

• To develop and evaluate a multimodal deep learning model for predicting IDH mutation status in brain glioma.
• To integrate 3D MRI images and text from radiology reports for enhanced diagnostic accuracy.
• To overcome the limitations of single-modality approaches in clinical brain glioma diagnosis.

Main Methods:

• A multimodal deep learning framework was proposed, incorporating both 3D MRI scans and associated text radiology reports.
• The model was trained and validated on the BraTS20 challenge dataset.
• Text data was annotated by clinicians from the First Affiliated Hospital of Zhengzhou University, China.

Main Results:

• The proposed multimodal model achieved a 4% improvement in prediction accuracy for IDH mutation status compared to state-of-the-art methods.
• The study demonstrated the effectiveness of combining imaging and textual data for improved diagnostic performance.
• The model showed superior overall performance in predicting IDH mutation status.

Conclusions:

• Multimodal deep learning, integrating MRI and radiology reports, offers a more accurate and practical approach for predicting brain glioma IDH mutation status.
• This approach enhances the benefits of machine learning in clinical diagnosis by leveraging diverse data sources.
• The findings suggest a promising direction for improving glioma patient management and treatment planning.