SA-LuT-Nets: Learning Sample-Adaptive Intensity Lookup Tables for Brain Tumor Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate brain tumor segmentation from MRI is crucial for clinical diagnosis and treatment planning.
• Variability in MRI signal intensity poses challenges for automated segmentation networks.
• Existing methods struggle with the non-quantitative nature of MR images, impacting generalizability.

Purpose of the Study:

• To develop a novel deep learning framework, SA-LuT-Net, for robust and accurate automatic brain tumor segmentation.
• To address the challenge of MRI signal variability by introducing a sample-adaptive intensity lookup table (LuT).
• To improve the performance and generalizability of brain tumor segmentation models.

Main Methods:

• Proposed a deep SA-LuT-Net framework integrating a LuT module with segmentation modules (DMFNet, 3D Unet).
• Learned sample-specific nonlinear intensity mapping functions (piece-wise linear, power functions) for adaptive contrast transformation.
• Trained the framework end-to-end, optimizing LuT parameters for improved segmentation performance.
• Validated the approach on BRATS2018 and BRATS2019 datasets using single and multi-modal MRI data.

Main Results:

• SA-LuT-Nets significantly improved the performance of baseline segmentation models (DMFNet, 3D Unet).
• The proposed method achieved state-of-the-art results on the BRATS2018 and BRATS2019 brain tumor segmentation datasets.
• Learned LuTs demonstrated generalizability, enhancing segmentation even when applied to different models.
• Superior performance was observed across single and multiple MRI modalities.

Conclusions:

• The SA-LuT-Net framework effectively addresses MRI intensity variations for enhanced brain tumor segmentation.
• The sample-adaptive LuT approach offers a significant advancement in automated medical image analysis.
• The learned LuTs possess transferable knowledge, benefiting diverse segmentation architectures and improving clinical decision-making.