Impact of Image Segmentation on High-Content Screening Data Quality for SK-BR-3 Cells
Author Information
Author(s): Hill Andrew A, LaPan Peter, Li Yizheng, Haney Steve
Primary Institution: Department of Biological Technologies, Wyeth Research
Hypothesis
Are segmentation errors common for the SK-BR-3 cell line, do they affect data quality, and can data quality be improved by identifying well-segmented cells?
Conclusion
Image segmentation has a measurable impact on HCS data, and using a multivariate shape-based filter can improve data quality.
Supporting Evidence
- Segmentation errors were found in over 64% of the segmented objects.
- The SVM-RBF classifier achieved 81% accuracy in identifying well-segmented cells.
- Eliminating poorly-segmented cells improved the purity of DNA content distributions.
- Well-segmented cells showed a clearer DNA content distribution compared to poorly-segmented cells.
- Significant morphological changes were detected in well-segmented cells but not in poorly-segmented cells.
Takeaway
This study shows that if we don't segment images of cells properly, we can get bad data, but we can fix this by using a special filter to find the well-segmented cells.
Methodology
The study involved human review of segmented cell images, training classifiers to distinguish well- and poorly-segmented cells, and applying these classifiers to improve data quality.
Potential Biases
Potential bias due to the subjective nature of human review in classifying segmentation quality.
Limitations
The study primarily focused on one cell line (SK-BR-3), which may limit the generalizability of the findings to other cell lines.
Participant Demographics
The study involved SK-BR-3 breast carcinoma cells.
Statistical Information
P-Value
2.5 × 10^-9
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website