Precision Segmentation for Brain Tumors with R2U-Net

Accurate segmentation of gliomas, primary brain tumors, presents a complex challenge due to their heterogeneity in terms of aggressiveness, prognosis, and histology. A recent study presented a Triplanar (2.5D) model based on Attention-Gated Recurrent Residual U-Net (R2U-Net) to improve the segmentation of these tumors.

Model Architecture and Performance

The proposed model integrates residual, recurrent, and triplanar architectures to enhance feature representation and segmentation accuracy, while maintaining computational efficiency. This approach could support better treatment planning. The model achieved a Dice Similarity Score (DSC) of 0.900 for Whole Tumor (WT) segmentation on the BraTS2021 validation set, demonstrating performance comparable to leading models.

Survival Prediction

The triplanar network extracts 64 features per planar model for survival days prediction, reduced to 28 using an Artificial Neural Network (ANN). This approach achieved an accuracy of 45.71%, a Mean Squared Error (MSE) of 108,318.128, and a Spearman Rank Correlation Coefficient (SRC) of 0.338 on the test dataset.