Unveiling Incomplete Modality Brain Tumor Segmentation: Leveraging Masked Predicted Auto-Encoder and Divergence Learning

TL;DR

This paper introduces a novel pre-training and knowledge distillation approach for brain tumor segmentation in multi-modal MRI, effectively handling incomplete modality data and improving segmentation accuracy.

Contribution

It proposes masked predicted auto-encoder pre-training and divergence-based knowledge distillation to enhance robustness against missing modalities.

Findings

Significant performance improvements on BRATS datasets.

Effective handling of incomplete modality data.

Enhanced model robustness and interpretability.

Abstract

Brain tumor segmentation remains a significant challenge, particularly in the context of multi-modal magnetic resonance imaging (MRI) where missing modality images are common in clinical settings, leading to reduced segmentation accuracy. To address this issue, we propose a novel strategy, which is called masked predicted pre-training, enabling robust feature learning from incomplete modality data. Additionally, in the fine-tuning phase, we utilize a knowledge distillation technique to align features between complete and missing modality data, simultaneously enhancing model robustness. Notably, we leverage the Holder pseudo-divergence instead of the KLD for distillation loss, offering improve mathematical interpretability and properties. Extensive experiments on the BRATS2018 and BRATS2020 datasets demonstrate significant performance enhancements compared to existing state-of-the-art methods.