Related Experiment Video
Updated: Jun 20, 2025
06:32
Evaluation of Biomarkers in Glioma by Immunohistochemistry on Paraffin-Embedded 3D Glioma Neurosphere Cultures
Published on: January 9, 2019
7.8K
Deep learning-based IDH1 gene mutation prediction using histopathological imaging and clinical data.
Riku Nakagaki1, Shyam Sundar Debsarkar2, Hiroharu Kawanaka1
1Graduate School of Engineering, Mie University, 1577 Kurima-machiya, Tsu, Mie 514-8507, Japan.
Computers in Biology and Medicine
|July 22, 2024
Summary
This study developed an AI model to predict IDH1 mutations in glioma patients using whole slide images and clinical data. Combining both data types improved prediction accuracy, aiding in disease progression assessment.
Area of Science:
- Histopathology
- Oncology
- Artificial Intelligence in Medicine
Background:
- Gliomas are brain tumors classified into astrocytoma, oligodendroglioma (collectively low-grade glioma, LGG), and glioblastoma (GBM).
- Isocitrate dehydrogenase (IDH) mutations are crucial prognostic indicators in glioma, with IDH-mutated patients generally exhibiting better outcomes.
- Accurate IDH mutation status is essential for glioma classification and treatment planning.
Purpose of the Study:
- To develop and evaluate an AI model for classifying the presence or absence of IDH1 mutations in glioma patients.
- To investigate the combined utility of whole slide images (WSIs) and clinical data for IDH1 mutation prediction.
- To compare the performance of different deep learning and machine learning models for this classification task.
Main Methods:
- Utilized a dataset of 546 glioma patients, incorporating hematoxylin and eosin (H&E) stained WSIs and associated clinical data.
- Employed deep learning models, including attention-based multiple instance learning (ABMIL) for image analysis, and LightGBM for clinical data.
- Implemented ensemble learning by combining WSI and clinical data models, with hyperparameter optimization to enhance classification accuracy.
Main Results:
- The WSI model achieved an Area Under the Curve (AUC) of 0.823, while the clinical data model reached an AUC of 0.782.
- The ensemble model, particularly the MaxViT and LightGBM combination, demonstrated the highest performance with an AUC of 0.852.
- Integrating both imaging and clinical data significantly improved the overall accuracy of AI-driven IDH1 mutation prediction.
Conclusions:
- AI models can effectively predict IDH1 mutation status in glioma using WSIs and clinical data.
- Ensemble learning approaches that combine multi-modal data (imaging and clinical) offer superior predictive performance.
- This integrated AI approach holds promise for enhancing glioma classification and disease progression assessment.
