LF-SynthSeg: Label-Free Brain Tissue-Assisted Tumor Synthesis and Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Unsupervised brain tumor segmentation is crucial for clinical applications but often neglects prior brain tissue knowledge.
• Existing methods typically treat segmentation as anomaly detection, limiting their effectiveness.

Purpose of the Study:

• To develop an innovative unsupervised brain tumor segmentation strategy by integrating prior brain tissue knowledge.
• To improve the accuracy and robustness of brain tumor segmentation from T2-weighted MRI scans.

Main Methods:

• A novel tumor synthesis mechanism using ellipsoids and brain tissue intensity profiles to create varied synthetic T2-weighted MRI scans.
• A training protocol encompassing both tumor and brain tissue segmentation to enhance boundary relationship learning.
• Evaluation on five public datasets (BRATS 2019-2021, PED, SSA).

Main Results:

• The proposed method significantly outperforms state-of-the-art unsupervised brain tumor segmentation techniques.
• Achieved over 92% of fully supervised performance on the same testing datasets.
• Demonstrated robust performance across diverse datasets.

Conclusions:

• The integration of prior brain tissue knowledge into tumor synthesis and segmentation enhances unsupervised brain tumor segmentation.
• This approach offers a powerful, data-efficient alternative to fully supervised methods for clinical applications.