Blur-Robust Nuclei Segmentation for Immunofluorescence Images.

Area of Science:

• Digital Pathology
• Biomedical Imaging
• Machine Learning

Background:

• Automated nuclei segmentation is vital for digital pathology.
• Existing methods struggle with poor-quality, out-of-focus images.
• Robust algorithms are needed for blurry immunofluorescence (IF) data.

Purpose of the Study:

• To evaluate nuclei segmentation performance on out-of-focus images.
• To develop a robust deep learning method for nuclei segmentation in blurry images.
• To improve digital pathology workflows with enhanced image analysis.

Main Methods:

• A deep learning encoder-decoder framework with a Y-forked decoder was proposed.
• The model performs simultaneous nuclei segmentation and image deblurring.
• A deblurring task was integrated to regularize the network for blurry inputs.

Main Results:

• The proposed method achieved accurate instance nuclei segmentation on both sharp and blurry images.
• The model demonstrated robustness across different levels of image blur.
• Experimental analysis showed superior performance compared to state-of-the-art methods on the Human U2OS cells dataset.

Conclusions:

• The novel deep learning approach effectively segments nuclei in out-of-focus microscopic images.
• Integrating a deblurring task enhances segmentation performance on low-quality data.
• This method offers a robust solution for digital pathology applications with challenging image conditions.