The RSNA-MICCAI brain tumor radiogenomic classification challenge focused on predicting the MGMT biomarker status in glioblastoma using binary classification on Multi-parameter mpMRI scans. This task is crucial for personalized medicine, as MGMT status can guide treatment decisions for patients with brain tumors. The dataset used in this challenge was divided into three cohorts: a training set, a validation set, and a testing set.
The training set and validation set were used during the model training phase, while the testing set was only used for final evaluation. The images in the dataset were provided either in the DICOM format or PNG format, which allowed participants to leverage different pre-processing techniques according to their preferences.
To solve the classification problem, participants explored various deep learning architectures. Notably, the 3D version of Vision Transformer (ViT3D), ResNet50, Xception, and EfficientNet-B3 were among the architectures investigated. These models have been previously successful in computer vision tasks and were adapted to handle the brain tumor radiogenomic classification challenge.
The performance of the models was evaluated using the area under the receiver operating characteristic curve (AUC), a widely used metric for binary classification tasks. The results indicated that both the ViT3D and Xception models achieved promising AUC scores of 0.6015 and 0.61745, respectively, on the testing set.
Comparing these results with previous studies and benchmarks, it is evident that the achieved AUC scores are competitive and valid, considering the complexity of the task. However, there is still room for improvement. To enhance the performance of the models, future research could explore different strategies, experiment with alternative architectures, and incorporate more diverse datasets.
Overall, this brain tumor radiogenomic classification challenge has shed light on the potential of deep learning models, such as ViT3D and Xception, for predicting MGMT biomarker status in glioblastoma. Exciting advancements can be expected in this field as researchers continue to refine their approaches and leverage new technologies.