Cell Nuclei Segmentation in Cytological Images Using Convolutional Neural Network and Seeded Watershed Algorithm.

Area of Science:

• Computational pathology
• Medical image analysis
• Biomedical engineering

Background:

• Nuclei segmentation is vital for cytological analysis but difficult due to clustered nuclei.
• Overlapping nuclei in cytological samples pose significant segmentation challenges.

Purpose of the Study:

• To develop and evaluate an automated approach for accurate nuclei segmentation in breast cancer cytological images.
• To compare the efficacy of a convolutional neural network (CNN)-based approach against traditional thresholding methods for nuclei segmentation.

Main Methods:

• A hybrid method combining CNN for semantic segmentation and seeded watershed transform for separating clustered nuclei.
• Image preprocessing using color deconvolution to enhance nuclei visualization.
• CNN identifies nuclei, cytoplasm, and background; watershed transform refines segmentation of overlapping nuclei.

Main Results:

• The CNN-watershed approach demonstrated superior performance in segmenting nuclei, particularly in cases with significant overlap.
• CNN outperformed Otsu and adaptive thresholding methods in generating topographical maps for watershed segmentation.
• Quantitative evaluation using Jaccard and Hausdorff distances confirmed the accuracy of the proposed method against manual segmentations.

Conclusions:

• The integration of CNN with watershed transform offers a robust solution for nuclei segmentation in challenging cytological images.
• This automated approach enhances the reliability of morphometric analysis in breast cancer diagnosis.
• The study highlights the advantage of using CNN-based semantic segmentation over traditional thresholding for improved nuclei separation.