Preliminary results from a crowdsourcing experiment in immunohistochemistry

Area of Science:

• Biomedical Image Analysis
• Computational Pathology
• Crowdsourcing in Science

Background:

• Crowdsourcing, outsourcing tasks to a large group, has proven effective in various fields.
• Immunohistochemistry (IHC) quantification is a time-consuming task for expert pathologists.
• This study explores crowdsourcing for IHC image analysis.

Purpose of the Study:

• To evaluate the efficacy of crowdsourcing techniques for quantifying immunohistochemistry.
• To assess the feasibility of using a large crowd for complex image analysis tasks.
• To determine if crowdsourced IHC quantification can provide reliable results comparable to expert pathologists.

Main Methods:

• Fourteen MIB1-stained breast specimen images were analyzed.
• Images were evaluated by a pathologist (gold standard) and then crowdsourced via a custom application.
• 10 crowdsourced positivity evaluations per image were collected and summarized using the median, then compared to the gold standard via Spearman correlation.

Main Results:

• 28 contributors evaluated an average of 4.64 images each.
• A high Spearman correlation of 0.946 (p < 0.001) was found between gold standard and crowdsourced positivity percentages.
• The median positivity percentage, a basic indicator, yielded good results.

Conclusions:

• Crowdsourced data aggregation can minimize identification errors in image analysis.
• This preliminary study demonstrates the potential of crowdsourcing for IHC quantification.
• Crowdsourcing offers a rapid solution for evaluating numerous images, particularly in research contexts.