An Optimization Framework for Processing and Transfer Learning for the Brain Tumor Segmentation
Tianyi Ren, Ethan Honey, Harshitha Rebala, Abhishek Sharma, Agamdeep Chopra, Mehmet Kurt
TL;DR
To address BraTS 2023 lesion-wise evaluation challenges, the paper proposes an optimization framework built on a 3D U-Net with carefully designed data preprocessing (Z-score normalization, intensity rescaling, histogram matching), loss functions, post-processing, and transfer learning. The framework demonstrates that combining preprocessing with a multi-channel training scheme on overlapping regions and subsequent post-processing can improve lesion-wise Dice to $0.79$, $0.72$, and $0.74$ across Challenges 1–3, indicating improved generalization and robustness for clinical deployment. The main contributions include a threshold-based inference pipeline, dust-removal strategies, and decoder-focused transfer learning that boosts cross-dataset performance. The work highlights practical steps toward clinically usable automated brain tumor segmentation and suggests avenues for further gains via data augmentation and more advanced loss designs.
Abstract
Tumor segmentation from multi-modal brain MRI images is a challenging task due to the limited samples, high variance in shapes and uneven distribution of tumor morphology. The performance of automated medical image segmentation has been significant improvement by the recent advances in deep learning. However, the model predictions have not yet reached the desired level for clinical use in terms of accuracy and generalizability. In order to address the distinct problems presented in Challenges 1, 2, and 3 of BraTS 2023, we have constructed an optimization framework based on a 3D U-Net model for brain tumor segmentation. This framework incorporates a range of techniques, including various pre-processing and post-processing techniques, and transfer learning. On the validation datasets, this multi-modality brain tumor segmentation framework achieves an average lesion-wise Dice score of 0.79, 0.72, 0.74 on Challenges 1, 2, 3 respectively.