Efficient Brain Tumor Segmentation With Multiscale Two-Pathway-Group Conventional Neural Networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Manual segmentation of brain tumors from MRI is challenging and time-consuming.
• Accurate segmentation is vital for diagnosis, treatment planning, and outcome evaluation.
• Existing automatic methods often rely on hand-crafted features or large annotated datasets, which are scarce in medicine.

Purpose of the Study:

• To develop an efficient and accurate automatic brain tumor segmentation model.
• To address the limitations of traditional deep learning methods in medical imaging.
• To exploit both local and global features for improved segmentation performance.

Main Methods:

• Proposed a novel two-pathway-group convolutional neural network (CNN) architecture.
• Incorporated equivariance to reduce model instability and overfitting.
• Employed a cascade architecture, integrating outputs from a basic CNN as an additional feature source.

Main Results:

• The two-pathway-group CNN architecture effectively exploits local and global contextual features.
• The model demonstrated improved performance over state-of-the-art methods on BRATS2013 and BRATS2015 datasets.
• Achieved attractive computational complexity alongside enhanced segmentation accuracy.

Conclusions:

• The developed two-pathway-group CNN offers a robust and efficient solution for brain tumor segmentation.
• This approach overcomes data scarcity issues in medical deep learning.
• The model shows significant potential for improving clinical diagnosis and treatment planning.