AI tool for predicting MGMT methylation in glioblastoma for clinical decision support in resource limited settings

Abstract

Glioblastoma is an aggressive brain cancer with a poor prognosis. The O6-methylguanine-DNA methyltransferase (MGMT) gene methylation status is crucial for treatment stratification, yet economic constraints often limit access. This study aims to develop an artificial intelligence (AI) framework for predicting MGMT methylation. Diagnostic magnetic resonance (MR) images in public repositories were used for training. The algorithm created was validated in data from a single institution. All images were segmented according to widely used guidelines for radiotherapy planning and combined with clinical evaluations from neuroradiology experts. Radiomic features and clinical impressions were extracted, tabulated, and used for modeling. Feature selection methods were used to identify relevant phenotypes. A total of 100 patients were used for training and 46 for validation. A total of 343 features were extracted. Eight feature selection methods produced seven independent predictive frameworks. The top-performing ML model was a model post-Least Absolute Shrinkage and Selection Operator (LASSO) feature selection reaching accuracy (ACC) of 0.82, an area under the curve (AUC) of 0.81, a recall of 0.75, and a precision of 0.75. This study demonstrates that integrating clinical and radiotherapy-derived AI-driven phenotypes can predict MGMT methylation. The framework addresses constraints that limit molecular diagnosis access.

Fig. 1

Workflow summary demonstrating the VOI’s delineation by experts from the radiation oncology field using T1GD and T2-Flair weighted sequences, the neuro-radiology assessment was carried together in this phase. Subsequently, these images together with the created VOI’s underwent RadF extraction. The data was then generated. Different ML experiments were carried out and compared with DL experiments.

Fig. 2

Architecture of Model 1 (top) and Model 2 (bottom) for predicting MGMT methylation status from T1GD and T2-Flair. Both models take separate T1 and FLAIR images as input, with Model 1 consisting of three convolutional layers and three pooling layers, and Model 2 adding two extra convolutional and pooling layers. Both models end with a flattening layer and two dense layers to output the methylation prediction. The dimensions for each layer can be found in Table S5 in the supplementary appendix.

Fig. 3

ROC curve and performance summary of the best classifier when applied to test_set.

Fig. 4

Heatmap with the correlation between variables in the third quartile of strength regarding both positive and negative correlations.

Fig. 5

(A and B) these graphs represent the accuracy and loss graphs for the first CNN model. (C and D) are the accuracy and loss graphs for the second CNN model.