Diffusion network with spatial channel attention infusion and frequency spatial attention for brain tumor...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate brain tumor segmentation from MRI is vital for patient diagnosis and treatment.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Current deep learning methods struggle with precise tumor boundary delineation.

Purpose of the Study:

• To develop an advanced deep learning model, SF-Diff, for precise whole tumor (WT) segmentation in brain MRI.
• To improve the accuracy of brain tumor segmentation by addressing challenges in boundary definition.

Main Methods:

• SF-Diff utilizes a conditional diffusion network incorporating a spatial channel attention infusion (SCAI) module and a frequency spatial attention (FSA) mechanism.
• The model extracts multiscale features and employs a diffusion process for boundary and detail restoration.
• A cross-model loss function is used to optimize feature extraction and noise distribution.

Main Results:

• Experiments on the Multimodal BraTS Challenge 2020 (BraTS2020) dataset demonstrated SF-Diff's superior performance.
• SF-Diff achieved a Dice score of 91.87%, Hausdorff 95 of 5.47 mm, and IoU of 84.96%.
• High sensitivity (92.29%) and specificity (99.95%) were recorded, indicating robust segmentation capabilities.

Conclusions:

• SF-Diff effectively segments the whole tumor (WT) region, outperforming state-of-the-art models, especially in complex boundary cases.
• The model's ability to delineate non-contiguous lesions is clinically significant.
• Future work will focus on extending SF-Diff to a three-class brain tumor segmentation task.