Cell segmentation for immunofluorescence multiplexed images using two-stage domain adaptation and weakly labeled data...

Area of Science:

• Computational Biology
• Biomedical Imaging
• Artificial Intelligence

Background:

• Multiplexed immunofluorescence (MxIF) imaging enables detailed cellular profiling for patient stratification in immunotherapy.
• Accurate cell segmentation is a critical prerequisite for reliable MxIF image analysis.
• Current segmentation methods may require extensive manual annotation, limiting scalability.

Purpose of the Study:

• To develop and validate a deep learning pipeline for automated cell segmentation in MxIF images.
• To improve the efficiency and accuracy of cell segmentation, particularly in scenarios with limited annotated data.
• To enhance patient stratification for immunotherapy through precise cellular analysis.

Main Methods:

• A Mask R-CNN deep learning model was trained for cell segmentation using nuclear (DAPI) and membrane (Na⁺K⁺ATPase) markers.
• A two-stage domain adaptation strategy was employed, involving pre-training on a weakly labeled dataset followed by fine-tuning on a manually annotated dataset.
• The model's performance was validated against manual annotations across three distinct datasets, including ovarian cancer and mouse pancreatic tissues.

Main Results:

• The proposed deep learning method achieved cell segmentation performance comparable to multi-observer agreement on an ovarian cancer dataset.
• The method demonstrated state-of-the-art improvements on a publicly available mouse pancreatic tissue dataset.
• Two-stage domain adaptation with a weakly labeled dataset significantly boosted performance, especially with smaller training sample sizes.
• The model achieved comparable results with reduced training data compared to larger sample sizes.

Conclusions:

• The developed deep learning pipeline effectively performs cell segmentation in MxIF images.
• Two-stage domain adaptation is a powerful strategy for enhancing segmentation accuracy, particularly when labeled data is scarce.
• This approach facilitates more accurate patient stratification for immunotherapy by improving cellular image analysis.
• The model has been integrated into CellProfiler, a widely used platform for cellular image analysis.