Multi-Pass Adaptive Voting for Nuclei Detection in Histopathological Images

Area of Science:

• Digital Pathology
• Medical Image Analysis
• Computational Biology

Background:

• Nuclei detection is crucial for computer-aided diagnosis and prognosis in digital pathology.
• Existing nuclei detection methods often fail with poor staining quality and noise from tissue preparation artifacts.
• There is a need for robust nuclei detection algorithms that can handle challenging image conditions.

Purpose of the Study:

• To introduce a novel Multi-Pass Adaptive Voting (MPAV) algorithm for nuclei detection.
• To address the limitations of current methods in handling poor staining and noise in digital pathology images.
• To evaluate the performance of MPAV on diverse tissue staining cohorts.

Main Methods:

• The Multi-Pass Adaptive Voting (MPAV) method utilizes the symmetric property of nuclear boundaries.
• It adaptively selects gradient information from edge fragments for nucleus location voting.
• MPAV was tested on three cohorts with different staining methods (Hematoxylin &Eosin, CD31 &Hematoxylin, Ki-67).

Main Results:

• MPAV demonstrated superior performance in nuclei detection across 47 images with approximately 17,700 manually labeled nuclei.
• The method achieved an area under the precision-recall curve (AUC) of 0.73.
• MPAV outperformed three state-of-the-art nuclei detection methods, including voting-based and deep learning approaches.

Conclusions:

• The Multi-Pass Adaptive Voting (MPAV) method offers a robust solution for nuclei detection in digital pathology images with suboptimal staining.
• MPAV's adaptive approach effectively handles noise and staining variations, outperforming existing methods.
• This technique has the potential to enhance the reliability of computer-aided diagnosis and prognosis systems.