Interactive cytometry, chance or evil of bias?

Area of Science:

• Oncology
• Biomedical Imaging
• Quantitative Pathology

Background:

• DNA histograms of breast cancer cells are key prognosticators.
• Interactive cell selection in imaging cytometry presents sampling challenges.
• Statistical variations are expected due to limited cell counts (e.g., 100 cells per case).

Purpose of the Study:

• To investigate statistical and systematic sampling problems in interactive cell selection for breast cancer DNA histogram analysis.
• To compare malignancy grading from controlled systematic sampling versus selective biased sampling.
• To evaluate the necessity of expert pathologists for reliable interactive sampling.

Main Methods:

• Analysis of DNA histograms from 361 breast cancer aspirate specimens.
• Measurements conducted across two independent laboratories.
• Comparison of sampling strategies: controlled systematic sampling vs. selective biased sampling.

Main Results:

• High statistical variations were observed, consistent with the low number of cells analyzed per case.
• Slight systematic differences were detected between laboratories.
• Controlled systematic sampling without pathological bias resulted in a higher malignancy grading compared to selective biased sampling.
• No significant contrary findings were observed to support the necessity of expert pathologists for interactive sampling.

Conclusions:

• The study highlights significant sampling issues in interactive cell selection for breast cancer prognostic analysis.
• Controlled, unbiased sampling may provide more accurate malignancy grading than selective methods.
• The findings challenge the assumption that expert pathologists are essential for reliable interactive cell sampling in this context.