MF-Net: Automated Muscle Fiber Segmentation From Immunofluorescence Images Using a Local-Global Feature Fusion...

Area of Science:

• Histopathology
• Biomedical Imaging
• Deep Learning

Background:

• Accurate histological assessment of skeletal muscle is vital for evaluating muscle atrophy, particularly under conditions like weightlessness.
• Segmentation of muscle fiber boundaries is a critical prerequisite for this evaluation.
• Challenges in segmenting muscle fibers from immunofluorescence images include low contrast and background noise, limiting traditional methods.

Purpose of the Study:

• To develop an effective method for segmenting macaque muscle fibers in immunofluorescence images.
• To address the limitations of existing methods in capturing global information and handling image noise and low contrast.

Main Methods:

• Proposed a muscle fiber segmentation network (MF-Net) utilizing a dual encoder with convolutional neural networks and transformers to capture both local and global features.
• Incorporated a low-level feature decoder to enhance global context by integrating multi-scale information.
• Validated the method on immunofluorescence datasets from six macaque weightlessness models.

Main Results:

• The MF-Net demonstrated effective segmentation of macaque muscle fibers, highlighting foreground features and suppressing background noise.
• The dual encoder successfully captured local and global feature information, improving segmentation accuracy.
• The low-level feature decoder aided in supplementing missing pixel details by combining different image scales.

Conclusions:

• The proposed MF-Net provides an accurate and effective automatic segmentation method for muscle fibers in immunofluorescence images.
• The approach successfully overcomes challenges associated with low contrast and background noise.
• The study validates the applicability of MF-Net for muscle fiber segmentation in weightlessness models.