External Validation of a Convolutional Neural Network for IDH Mutation Prediction

Area of Science:

• Radiogenomics and Artificial Intelligence in Neuro-oncology
• Medical Imaging and Diagnostic Tools for Glioma
• Biomarker Discovery and Validation in Oncology

Background:

• Isocitrate dehydrogenase (IDH) status is a critical prognostic factor for gliomas.
• Current IDH status determination requires invasive procedures.
• Radiomics and radiogenomics offer non-invasive prediction potential using AI and MRI.

Purpose of the Study:

• To externally validate an AI-based algorithm for predicting IDH status in gliomas.
• To assess the algorithm's performance using preoperative MRI sequences and patient age.
• To evaluate the generalizability of AI-driven radiogenomic predictions.

Main Methods:

• Application of Yoon Choi et al.'s IDH prediction algorithm on T1c, T2, and FLAIR MRI scans.
• Utilized preoperative MRI data from 21 adult glioma patients (WHO grades II-IV).
• Developed an automated script for processing DICOM MRI sequences and predicting IDH status.

Main Results:

• The external validation achieved a relative accuracy of 76% (95% CI: 53%-92%).
• Area Under the Curve (AUC) was 0.74 (95% CI: 0.53-0.91), with a p-value of 0.021.
• Sensitivity and Specificity were 0.78 (95% CI: 0.45-0.96) and 0.75 (95% CI: 0.47-0.91), respectively.

Conclusions:

• The algorithm's performance on external data, while consistent with prior tests, is not yet sufficient for clinical application.
• Radiogenomic approaches show significant potential for rapid and accurate glioma diagnosis and prognosis.
• Further validation and refinement of AI algorithms are necessary for clinical integration.