Singular Nuclei Segmentation for Automatic HER2 Quantification Using CISH Whole Slide Images.

Area of Science:

• Oncology
• Medical Imaging
• Computational Pathology

Background:

• Human epidermal growth factor receptor 2 (HER2) quantification is crucial for breast cancer treatment decisions.
• Current methods like chromogenic in situ hybridization (CISH) require manual nuclei selection, which is time-consuming and prone to errors.
• Existing automated methods using support vector machines (SVM) have limitations in accurately detecting singular nuclei.

Purpose of the Study:

• To develop a robust automated method for singular nuclei detection in CISH whole slide images (WSIs) for accurate HER2 quantification.
• To improve the reliability of HER2 grading for HER2-targeted therapy selection in breast cancer patients.
• To compare the performance of the proposed method against a previously established SVM-based approach.

Main Methods:

• A U-net based deep learning model was developed for singular nuclei detection.
• Complementary color correction and deconvolution techniques were integrated to enhance image quality and feature extraction.
• The method was applied to CISH whole slide images for HER2 quantification and grading.

Main Results:

• The proposed U-net based method demonstrated robust detection of singular nuclei.
• The automated HER2 quantification using the U-net model showed improved accuracy compared to the SVM-based method.
• The integration of color correction and deconvolution contributed to enhanced detection performance.

Conclusions:

• The developed U-net based method offers a reliable and accurate solution for automated HER2 quantification in breast cancer.
• This approach has the potential to streamline the pathological workflow and improve the precision of HER2-targeted therapy selection.
• Further validation on larger datasets is warranted to confirm its clinical utility.