UNSUPERVISED SHAPE PRIOR MODELING FOR CELL SEGMENTATION IN NEUROENDOCRINE TUMOR

Area of Science:

• Medical Imaging
• Computational Pathology
• Biomedical Engineering

Background:

• Accurate cell segmentation is crucial for quantitative analysis of neuroendocrine tumor (NET) images.
• Automated segmentation is challenging due to significant variations in cell morphology.

Purpose of the Study:

• To develop an automated and accurate cell segmentation method for NET images.
• To overcome the limitations of supervised methods requiring extensive annotated data.

Main Methods:

• Incorporation of unsupervised shape priors into a repulsive deformable model.
• Application of group similarity for shape constraints, eliminating the need for manual annotation.
• Testing the algorithm on 51 NET images.

Main Results:

• The proposed unsupervised method achieves superior performance in cell segmentation on NET images.
• Comparative experiments demonstrate the effectiveness against state-of-the-art techniques.
• The algorithm successfully avoids labor-intensive data annotation.

Conclusions:

• Unsupervised shape priors integrated into a repulsive deformable model offer an efficient solution for automated NET cell segmentation.
• This method provides a robust alternative to supervised approaches, particularly when annotated data is scarce.
• The findings support enhanced quantitative analysis of digitized NET images.