IDCC-SAM: A Zero-Shot Approach for Cell Counting in Immunocytochemistry Dataset Using the Segment Anything Model

Area of Science:

• Biomedical research
• Microscopy
• Computational biology

Background:

• Manual cell counting in immunocytochemistry is laborious and prone to errors.
• Deep learning methods require extensive, costly labeled datasets, hindering scalability.
• Accurate cell quantification is crucial for disease diagnosis and treatment research.

Purpose of the Study:

• To introduce IDCC-SAM, a novel application of the Segment Anything Model (SAM) for zero-shot cell counting in immunocytochemistry.
• To leverage SAM's pre-trained capabilities to eliminate the need for manual annotations.
• To enhance the scalability and efficiency of cell counting in microscopic imaging.

Main Methods:

• Utilized Meta AI's Segment Anything Model (SAM), pre-trained on a large image dataset.
• Adapted SAM for zero-shot cell counting specifically for fluorescent microscopic immunocytochemistry data.
• Evaluated IDCC-SAM on three public datasets: IDCIA, ADC, and VGG.

Main Results:

• IDCC-SAM achieved the lowest Mean Absolute Error on VGG (26) and ADC (28) datasets.
• Demonstrated the highest Acceptable Absolute Error across all tested datasets (28%, 26%, 33%).
• Outperformed state-of-the-art supervised (U-Net, Mask R-CNN) and zero-shot (NP-SAM, SAM4Organoid) models.

Conclusions:

• IDCC-SAM offers a scalable and efficient solution for cell counting in immunocytochemistry.
• The zero-shot approach significantly reduces the reliance on manual annotations and specialized models.
• This method has the potential to improve accuracy and accessibility in biomedical image analysis.