Deep Voting: A Robust Approach Toward Nucleus Localization in Microscopy Images.

Area of Science:

• Medical Imaging
• Computational Biology
• Computer-Aided Diagnosis

Background:

• Accurate nuclei localization in microscopy is vital for computer-aided diagnosis.
• Microscopy images often present challenges like heavy cluttering and morphologic variations.

Purpose of the Study:

• To propose a robust and accurate nuclei localization method for microscopy images.
• To develop a convolutional neural network (CNN) based Hough voting approach named 'deep voting'.

Main Methods:

• The 'deep voting' method utilizes a CNN to assign voting offset vectors and confidence scores to image patches.
• It constructs an implicit Hough-voting codebook.
• Nucleus positions are identified by analyzing a voting density map generated via weighted votes and Parzen-window estimation.

Main Results:

• The method achieves state-of-the-art performance in nuclei localization.
• It demonstrates effectiveness on Neuroendocrine Tumor (NET) microscopy images.
• Requires minimal annotation effort (one click near the nucleus center).

Conclusions:

• The proposed 'deep voting' method offers a robust solution for nuclei localization in challenging microscopy images.
• This technique has significant potential to improve computer-aided diagnosis systems.
• The approach is efficient and requires minimal user input for annotation.