High throughput automatic muscle image segmentation using parallel framework.

Area of Science:

• Biomedical Imaging
• Computational Pathology
• Digital Health

Background:

• Accurate skeletal muscle cell segmentation is vital for diagnosing muscle diseases.
• Current methods are often complex, slow, and unsuitable for large-scale images.
• Manual annotation is labor-intensive and time-consuming.

Purpose of the Study:

• To develop a fast and accurate automatic segmentation method for skeletal muscle cell images.
• To address the limitations of existing methods in handling large-scale specimens.
• To improve the efficiency of muscle-related disease diagnosis.

Main Methods:

• A distributed computing approach using data and model parallelism on the Spark platform.
• Master-worker parallelism for distributing image data across multiple nodes.
• Utilizing a structured random forest (SRF) contour detector and superpixel technique for region candidate generation.
• Implementing a hierarchical tree-based region selection algorithm with conditional random field (CRF) for segmentation, parallelized using multi-core programming.

Main Results:

• Achieved over 10x speed improvement compared to standalone implementations on large-scale muscle images.
• Demonstrated high-quality segmentation results comparable to state-of-the-art methods.
• Successfully processed images containing hundreds to thousands of cells efficiently.

Conclusions:

• The proposed parallel muscle image segmentation method effectively utilizes data and model parallelism.
• The parallel strategy is highly compatible with the muscle segmentation framework.
• The method enables high-throughput and effective cell segmentation on large-scale muscle images.