Automatic segmentation of skin cells in multiphoton data using multi-stage merging.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Dermatology

Background:

• Accurate segmentation of human skin cells in 3D multiphoton tomography data is crucial for diagnostic applications.
• Existing methods often struggle with depth-dependent variations in contrast and cell size, limiting robustness.

Purpose of the Study:

• To develop a novel, robust automatic segmentation algorithm for human skin cell components (cytoplasm and nuclei) in 3D multiphoton tomography data.
• To introduce new features for improved cell classification and to enable automated analysis for potential skin cancer detection.

Main Methods:

• A multi-stage superpixel merging approach was employed to overcome limitations of global thresholds and handle depth-dependent data characteristics.
• A cell model utilizing four features, including two novel metrics (Outer Cell Inner Nucleus relationship and stability index), was developed for fuzzy classification.
• The algorithm was validated on a 3D image stack of human skin layers (stratum spinosum and stratum basale).

Main Results:

• The proposed algorithm demonstrated robust segmentation of skin cell components, independent of empirical global thresholds.
• The novel features (OCIN and stability index) combined with existing ones improved the model-based fuzzy evaluation of cell segments.
• Successful application on a 3D image dataset of healthy human skin confirmed the pipeline's efficacy.

Conclusions:

• The developed image processing pipeline enables fully automated classification of human skin cells in multiphoton data.
• This approach provides a foundational tool for non-invasive optical biopsy and the early detection of skin cancer.