Semi-Automatic Neuron Segmentation in Electron Microscopy Images Via Sparse Labeling.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Cell Biology

Background:

• Accurate segmentation of cell membranes in electron microscopy images is crucial for understanding cellular structures and functions.
• Existing automated methods often struggle with accuracy and require extensive manual annotation.
• Sparse labeling offers a potential solution to reduce manual effort.

Purpose of the Study:

• To introduce a novel, user-guided method for sparse labeling of membranes in electron microscopy images.
• To evaluate the efficiency and accuracy of the proposed method compared to existing techniques.
• To demonstrate the potential of minimal user input for accurate image segmentation.

Main Methods:

• User interaction involves marking sparse labels on membrane crossings using gridlines as guides.
• A best path algorithm is employed to connect sparse labels and generate a complete membrane segmentation.
• The method utilizes electron microscopy image data.

Main Results:

• The proposed sparse labeling method significantly outperforms automatic methods in terms of segmentation accuracy (lower Rand error).
• The technique requires minimal user input, with as little as 2% of the image being labeled.
• The best path algorithm effectively connects sparse labels to create accurate membrane segmentations.

Conclusions:

• This novel method offers an efficient and accurate approach for segmenting cell membranes in electron microscopy images.
• The user-guided sparse labeling technique drastically reduces the annotation effort while achieving high-quality results.
• The method holds promise for accelerating biological research that relies on detailed cellular imaging.