Deep Learning Nuclei Detection in Digitized Histology Images by Superpixels

Area of Science:

• Digital Pathology
• Computational Biology
• Medical Image Analysis

Background:

• Accurate nuclei detection is crucial for classifying cervical intraepithelial neoplasia (CIN) grades.
• Histopathology image analysis benefits from advanced computational techniques for feature identification.

Purpose of the Study:

• To develop and evaluate a deep learning (DL)-based nuclei segmentation method for CIN classification.
• To enhance the accuracy of nuclei detection in histopathology images.

Main Methods:

• Utilized a deep learning approach for nuclei segmentation.
• Employed superpixel generation via simple linear iterative clustering (SLIC) to gather localized image information.
• Trained a convolutional neural network (CNN) for nuclei segmentation.

Main Results:

• Achieved an overall nuclei detection accuracy of 95.97% on a dataset of 133 digitized histology images.
• Demonstrated superior performance compared to traditional imaging-based and clustering-based benchmark methods.
• The DL-based method with superpixel analysis showed improved segmentation results.

Conclusions:

• The proposed deep learning nuclei segmentation method with superpixel analysis offers enhanced accuracy for CIN classification.
• This approach represents an advancement over current state-of-the-art methods in histopathology image analysis.