SM-SegNet: A Lightweight Squeeze M-SegNet for Tissue Segmentation in Brain MRI Scans

Area of Science:

• Medical Imaging
• Computer Vision
• Neuroscience

Background:

• Accurate brain segmentation in Magnetic Resonance Imaging (MRI) is crucial for neurological disorder diagnosis and treatment planning.
• Existing segmentation methods often face challenges with speed, memory efficiency, and accuracy.

Purpose of the Study:

• To introduce a novel Squeeze M-SegNet (SM-SegNet) architecture for efficient and accurate brain MRI segmentation.
• To leverage a fire module and combined-connections to enhance segmentation performance.

Main Methods:

• Developed the SM-SegNet architecture incorporating a fire module with squeeze-expand convolutional layers.
• Utilized uniform input patches, combined-connections, and long skip connections for improved feature transfer and gradient stability.
• Implemented multi-scale deep networks on the encoder and deep supervision on the decoder for robust feature extraction and training.

Main Results:

• The SM-SegNet achieved high segmentation accuracies: 95% for cerebrospinal fluid, 95% for gray matter, and 96% for white matter.
• The model demonstrated significantly faster training and more efficient memory usage, with 83% fewer parameters compared to existing methods.
• Outperformed previous methods in both subjective and objective metrics on OASIS and IBSR datasets.

Conclusions:

• The proposed SM-SegNet architecture offers a significant advancement in brain MRI segmentation.
• The novel approach provides a faster, more memory-efficient, and highly accurate solution for segmenting brain tissues from MRI scans.
• SM-SegNet shows great potential for clinical applications requiring precise brain structure analysis.