An adaptive digital stain separation method for deep learning-based automatic cell profile counts.

Area of Science:

• Digital pathology
• Computational biology
• Histology

Background:

• Accurate cell quantification in biological tissues is vital for pathology, toxicology, cancer, and behavioral studies.
• Automated methods, including deep learning, can quantify cells in Extended Depth of Field (EDF) images with high accuracy and reproducibility.
• Current automated methods are primarily designed for single-immunostained tissue sections.

Purpose of the Study:

• To develop an adaptive method for separating stain color channels in dual-stained tissue images.
• To enable the application of automated cell counting methods to complex dual-staining protocols.
• To enhance the accuracy and efficiency of cell quantification in histological analysis.

Main Methods:

• Developed an adaptive stain color separation method for dual-stained tissue sections.
• Applied deep learning-based and hand-crafted algorithms for automatic cell counting.
• Utilized images from sections immunostained for neurons (Neu-N) and microglial cells (Iba-1) with cresyl violet counterstain.

Main Results:

• The proposed stain separation method overcomes limitations of existing techniques, such as requiring pure stain color basis.
• Deep learning methods demonstrated more accurate cell counts compared to the hand-crafted method.
• Stain-separated images serve as effective input for deep learning-based quantification of single-stained tissue sections.

Conclusions:

• The developed stain separation technique successfully enables automated cell quantification in dual-stained tissues.
• Deep learning-based automated cell counting provides superior accuracy over traditional methods.
• This approach expands the utility of automated quantification tools for complex histological samples.