[Paper Review] Deep Learning Based Brain Tumor Segmentation: A Survey

Zhihua Liu, Tong Lei+7 more|arXiv (Cornell University)|Jul 18, 2020

Brain Tumor Detection and Classification17 citations

TL;DR

This survey provides a comprehensive analysis of deep learning-based brain tumor segmentation methods, covering network architectures, handling of data imbalance, and multi-modality fusion. It synthesizes insights from over 100 studies, highlighting key trends, technical challenges, and future research directions in automated brain tumor segmentation using deep learning.

ABSTRACT

Brain tumor segmentation is one of the most challenging problems in medical image analysis. The goal of brain tumor segmentation is to generate accurate delineation of brain tumor regions. In recent years, deep learning methods have shown promising performance in solving various computer vision problems, such as image classification, object detection and semantic segmentation. A number of deep learning based methods have been applied to brain tumor segmentation and achieved promising results. Considering the remarkable breakthroughs made by state-of-the-art technologies, we use this survey to provide a comprehensive study of recently developed deep learning based brain tumor segmentation techniques. More than 100 scientific papers are selected and discussed in this survey, extensively covering technical aspects such as network architecture design, segmentation under imbalanced conditions, and multi-modality processes. We also provide insightful discussions for future development directions.

Motivation & Objective

• To provide a systematic review of deep learning-based brain tumor segmentation techniques published between 2012 and 2020.
• To analyze technical advancements in network architecture, data imbalance handling, and multi-modal MRI fusion.
• To identify methodological gaps and emerging trends in brain tumor segmentation using deep learning.
• To offer insights into future research directions based on critical evaluation of existing methods and challenges.
• To fill the gap in existing surveys by focusing specifically on deep learning methods with technical depth and methodological categorization.

Proposed method

• The survey conducts a structured literature review of over 100 scientific papers on deep learning-based brain tumor segmentation.
• Methods are categorized into network architecture design, handling of class imbalance, and multi-modality learning strategies.
• The authors analyze modality learning techniques, including modality ranking, modality pairing, and feature fusion mechanisms such as concatenation and attention modules.
• The survey evaluates methods for missing modality scenarios, including generative adversarial networks (GANs) for synthetic modality generation.
• Technical comparisons are made across methods in terms of performance, parameter efficiency, and generalization capability.
• The analysis includes evaluation metrics such as Dice score, particularly from the BraTS challenge, to benchmark method performance.

Experimental results

Research questions

• RQ1 How have deep learning architectures evolved to improve brain tumor segmentation accuracy and robustness?
• RQ2 What are the most effective strategies for addressing class imbalance in brain tumor segmentation datasets?
• RQ3 How do multi-modal MRI fusion techniques enhance segmentation performance compared to single-modality approaches?
• RQ4 What are the key challenges and solutions in handling missing MRI modalities in clinical settings?
• RQ5 What are the most promising future research directions in deep learning-based brain tumor segmentation?

Key findings

• Multi-modal MRI fusion significantly improves segmentation performance, with state-of-the-art models on BraTS achieving whole tumor Dice scores above 0.85.
• Attention mechanisms and learnable fusion modules enhance feature representation and improve segmentation accuracy, though at the cost of increased computational complexity.
• Methods that generate missing modalities using GANs show promise, but performance heavily depends on the quality of the input modality.
• Networks with hierarchical or decoupled convolutional designs (e.g., HDC-Net) demonstrate improved feature learning and generalization on complex tumor morphologies.
• Self-supervised and contrastive learning-based approaches reduce reliance on large-scale annotated data, improving label efficiency.
• The integration of knowledge distillation and ensemble techniques leads to improved robustness and performance in the BraTS 2020 challenge.