Reproducible Evaluation of Data Augmentation and Loss Functions for Brain Tumor Segmentation
Saumya B
TL;DR
This paper addresses reproducible evaluation for brain tumor segmentation by applying a U-Net with focal loss to MRI data and analyzing simple data augmentation strategies. The two-phase approach first tunes focal loss parameters and then assesses augmentation effects, establishing a transparent baseline for future research. Key findings show that focal loss parameter choices significantly affect segmentation quality, with horizontal flip and rotation augmentations improving robustness, while scaling offers limited benefit; the results are competitive with state-of-the-art methods. The work provides a public, reproducible framework and codebase to guide subsequent exploration of loss functions and augmentation strategies in brain tumor segmentation.
Abstract
Brain tumor segmentation is crucial for diagnosis and treatment planning, yet challenges such as class imbalance and limited model generalization continue to hinder progress. This work presents a reproducible evaluation of U-Net segmentation performance on brain tumor MRI using focal loss and basic data augmentation strategies. Experiments were conducted on a publicly available MRI dataset, focusing on focal loss parameter tuning and assessing the impact of three data augmentation techniques: horizontal flip, rotation, and scaling. The U-Net with focal loss achieved a precision of 90%, comparable to state-of-the-art results. By making all code and results publicly available, this study establishes a transparent, reproducible baseline to guide future research on augmentation strategies and loss function design in brain tumor segmentation.