Nuclei detection in breast histopathology images with iterative correction

Area of Science:

• Digital pathology
• Computational biology
• Medical image analysis

Background:

• Accurate nuclei detection and localization are crucial for diagnosing breast cancer from histopathology images.
• Existing methods face challenges with cluttered nuclei and achieving high localization precision.

Purpose of the Study:

• To develop an advanced deep network architecture for improved nuclei detection and precise localization in breast cancer histopathology images.
• To enhance the accuracy of nuclei identification and spatial mapping in digital pathology.

Main Methods:

• A two-part deep network: a nuclear candidate generation module and a nuclear localization refinement module.
• Novel patch learning with added location representations for multi-task learning (classification and localization).
• Deep supervision mechanism for coherent multi-scale layer contributions and an iterative correction strategy for localization refinement.

Main Results:

• The proposed method significantly improves nuclei detection performance on H&E stained histopathology datasets.
• Demonstrates superior accuracy in localizing nuclei compared to state-of-the-art approaches.
• Achieves better results than existing techniques, particularly in detecting multiple, cluttered nuclei.

Conclusions:

• The developed deep network architecture effectively enhances nuclei detection and localization accuracy in breast cancer histopathology.
• The novel patch learning and iterative refinement strategies contribute to superior performance, especially in challenging imaging scenarios.
• This approach offers a promising tool for more precise analysis of histopathological images in cancer diagnostics.