Automatic segmentation and supervised learning-based selection of nuclei in cancer tissue images

Area of Science:

• Biomedical imaging
• Computational pathology
• Genomics

Background:

• Accurate cell nuclei segmentation is crucial for gene localization analysis in breast cancer diagnosis.
• Manual segmentation is time-consuming, subjective, and impractical for large-scale studies.
• Existing imaging data often contains more nuclei than required for analysis.

Purpose of the Study:

• To develop an automated workflow for selecting accurately delineated cell nuclei from tissue images.
• To enable efficient and objective quantitative analysis of gene positioning for breast cancer diagnosis.
• To validate the performance of the automated segmentation and selection process.

Main Methods:

• A multistage watershed-based algorithm for automatic cell nuclei segmentation.
• An artificial neural network-based pattern recognition engine for screening segmented nuclei.
• Quantitative performance evaluation using visual confirmation and comparison with a 2D dynamic programming reference method.
• Application to discriminate normal and cancerous breast tissue based on HES5 gene positioning.

Main Results:

• The workflow successfully selected a subpopulation of accurately delineated nuclei.
• Performance was quantified by the fraction of confirmed well-segmented nuclei and boundary accuracy.
• The automated method showed high agreement with manual analysis across cancer, normal, and non-cancerous breast disease cases.

Conclusions:

• The integrated workflow provides an accurate and robust method for automated cell nuclei selection.
• This approach facilitates large-scale gene localization analysis for improved breast cancer diagnosis.
• The method demonstrates potential for objective and efficient pathological assessment.