Self-Supervised Wavelet-Based Attention Network for Semantic Segmentation of MRI Brain Tumor

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Accurate brain tumor segmentation is crucial for treatment planning.
• Manual segmentation is time-consuming and prone to inaccuracies.
• Gliomas present challenges in MRI segmentation due to low contrast and variable appearance.

Purpose of the Study:

• To develop an automated method for precise brain tumor segmentation in MRI scans.
• To address limitations of existing segmentation techniques, such as susceptibility to noise.
• To improve the analysis of pathological conditions by evaluating tumor characteristics.

Main Methods:

• Introduction of a Self-Supervised Wavelet-based Attention Network (SSW-AN).
• Utilizes a novel attention module with self-supervised activation functions and dynamic weights.
• Employs 2D Wavelet transform for input and labels, segmenting data into low and high-frequency channels.
• Incorporates channel and spatial attention modules within the self-supervised attention block (SSAB).

Main Results:

• The SSW-AN effectively captures global context information for improved segmentation.
• The method demonstrates superior performance compared to current state-of-the-art algorithms.
• Achieved higher accuracy, enhanced dependability, and reduced redundancy in medical image segmentation tasks.

Conclusions:

• The proposed SSW-AN offers a more accurate and reliable approach to brain tumor segmentation.
• This automated method aids radiologists in determining appropriate patient treatment plans.
• The technique shows significant promise for advancing medical image analysis in neuro-oncology.