Effects of segmentation errors on downstream-analysis in highly-multiplexed tissue imaging

Area of Science:

• Biomedical Imaging
• Computational Biology
• Single-cell Analysis

Background:

• Highly multiplexed single-cell imaging advances tissue analysis by capturing spatial protein expression.
• Accurate cell segmentation is fundamental for generating reliable single-cell expression profiles.
• The impact of segmentation inaccuracies on downstream analyses remains poorly quantified.

Purpose of the Study:

• To introduce a framework for simulating segmentation errors using affine transformations.
• To evaluate the robustness of downstream analyses, including cell clustering and phenotyping, to segmentation inaccuracies.
• To quantify the propagation of segmentation errors in multiplexed single-cell imaging data.

Main Methods:

• Developed a framework employing affine transformations to simulate realistic cell segmentation errors.
• Applied simulated segmentation perturbations to multiplexed single-cell imaging data.
• Assessed the impact of segmentation errors on unsupervised k-Means, graph-based Leiden clustering, and Gaussian Mixture Model phenotyping.

Main Results:

• Moderate segmentation errors significantly distort single-cell protein profiles and cellular neighborhood relationships.
• Clustering analyses (k-Means, Leiden) showed reduced consistency with increased segmentation error, particularly with smaller neighborhood sizes.
• Cell phenotyping using Gaussian Mixture Models resulted in misclassifications between similar cell types due to segmentation inaccuracies.

Conclusions:

• Segmentation quality is critical for reliable downstream analysis in multiplexed tissue imaging.
• Mitigating spurious results requires careful data processing and consideration of segmentation inaccuracies.
• Probabilistic modeling frameworks may enhance the reliability and reproducibility of findings in spatial biology studies.