Semisupervised adaptive learning models for IDH1 mutation status prediction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Isocitrate dehydrogenase 1 (IDH1) mutation status is crucial for glioma diagnosis, treatment, and prognosis.
• Accurate prediction of IDH1 status from MRI data presents a significant challenge due to limitations in existing methods.
• Data waste and instability issues hinder current techniques for IDH1 mutation status determination.

Purpose of the Study:

• To develop a semisupervised adaptive deep learning model for predicting IDH1 mutation status in glioma using MRI data.
• To address data insufficiency and feature redundancy issues in medical imaging analysis for glioma.
• To enhance the accuracy and efficiency of intelligent diagnosis for glioma IDH1 mutation status.

Main Methods:

• A semisupervised adaptive deep learning model integrating radiomics and rough sets was developed.
• Radiomics features were refined using rough set algorithms, and pseudo-labels were generated for unlabeled data.
• The model employed U-Net and CRNN architectures, optimized with Sand Cat Swarm Optimization (SCSO), to form the UCNet classifier.

Main Results:

• The proposed model achieved a prediction accuracy of 95.63% for glioma IDH1 mutation status.
• The study demonstrated effective utilization of glioma imaging data through feature selection and pseudo-labeling.
• Experimental results validated the model's capability in accurate intelligent diagnosis of IDH1 mutation status.

Conclusions:

• The developed semisupervised deep learning model significantly improves the prediction accuracy of IDH1 mutation status in glioma.
• This approach enhances the utility of MRI data in clinical decision-making for glioma patients.
• The findings pave the way for more precise and efficient AI-driven diagnostic tools in neuro-oncology.