Calibration-Aimed Comparison of Image-Cytometry- and Flow-Cytometry-Based Approaches of Ploidy Analysis

Abstract

Ploidy analysis is the fundamental method of measuring DNA content. For decades, the principal way of conducting ploidy analysis was through flow cytometry. A flow cytometer is a specialized tool for analyzing cells in a solution. This is convenient in laboratory environments, but prohibits measurement reproducibility and the complete detachment of sample preparation from data acquisition and analysis, which seems to have become paramount with the constant decrease in the number of pathologists per capita all over the globe. As more open computer-aided systems emerge in medicine, the demand for overcoming these shortcomings, and opening access to even more (and more flexible) options, has also emerged. Image-based analysis systems can provide an alternative to these types of workloads, placing the abovementioned problems in a different light. Flow cytometry data can be used as a reference for calibrating an image-based system. This article aims to show an approach to constructing an image-based solution for ploidy analysis, take measurements for a basic comparison of the data produced by the two methods, and produce a workflow with the ultimate goal of calibrating the image-based system.

Keywords: DNA content; calibration; cytology; cytometry; digital pathology; ploidy analysis.

Figure 1

Method of background intensity measurement: the resolution of the system is visible; the objects are only tens of pixels in size.

Figure 2

FCM (a) and ICM DNA content (b) histograms of one of the samples. Both x-axis values are intensity sums: in the FCM case, the maximal integral recorded at the detector during a nucleus passthrough; in the ICM case, the sum of the pixel intensities belonging to the segmented nucleus region (corrected by approximated background intensity). The dual peaks typical of samples from healthy donors are identifiable on the plots of both methods.

Figure 3

Flow cytometry data on a DNA content–object area scatterplot. Each data point represents a cell nucleus. The units of both axes are intensity-based: on the y-axis is the DNA content; on the x-axis is the forward scatter (a value proportional to the object size). The two populations of the 2N and 4N peaks are identifiable from looking at the y-axis.

Figure 4

Image cytometry data on a DNA content–object area scatterplot. Each data point represents a cell nucleus. The unit of both axes are intensity-based: on the y-axis is the integrated optical density—the proposed model for the DNA content; on the x-axis is the object area, measured after segmenting the nuclei from the image. Some subpopulations are identifiable, but not as clearly as on the FCM plot.

Figure 5

Principal component analysis (PCA) coefficient values across all 17 samples: The three input dimensions are area, granularity, and integrated fluorescence. Such high PCA coefficients mean that there is low dependence between these measured values, and they represent the associated measurements well. PC1 explains DNA content, PC2 explains granularity, and PC3 explains object area.

Figure 6

ICM data after the correction defined by the PCA: The resemblance to the FCM data plot (Figure 3) was improved. Each data point represents a cell nucleus, but now in the PC1–PC3 coefficient plane (representing the underlying, noise-filtered area, and DNA content measurements). The subpopulations of the 2N and 4N peaks are more prominent.

Figure 7

For each of the 17 samples, clustering gap analysis was conducted to approximate the possible object population counts. The results of a single sample are presented here with gap values (top), k-means method, with the different colors designating different populations (middle), and c-Means with the different colors designating different populations, with the cluster centers (bottom). These measurements were used to identify a model for the side scatter measurement of the FCM method that was proportional to the object granularity, with the internal structures identifiable.

Figure 8

The results of c-means clustering on a single sample, with the clusters projected back to the DNA content distribution of the sample (blue, green, orange). These measurements were used to identify the approximate locations of the subpopulations of the 2N and 4N peaks.

Figure 9

The DNA content distribution of sample #2, with the clusters projected back to the DNA content distribution of the sample (blue, green, orange), and normal distributions fitted to the latter two populations (magenta, red).

Figure 10

The original ICM data scatterplot recolored based on the clustering results: Data point radii represent the cluster membership value. Classification results identified the peak populations, and provided the basis of the measurement of their DNA content values.

Figure 11

Overview of the reference (FCM) and the proposed (ICM) methods: The two approaches attempt to measure the same physical property through two different appliances. The ICM method was shown to have a linear response to uniform-intensity sources (i.e., fluorescent beads) with the parameter of exposure time (within the detection range of the appliance). Measuring nuclei, the same method is proposed to also have a linear relationship with the values produced by the reference method.

Figure 12

The ratio of the 2N and the 4N peak DNA content ratios for the histogram bin-based image cytometry (top), for flow cytometry (middle), and the reference and normal distribution fit-based image cytometry (bottom) results: The horizontal axis contains the 17 samples by index; the vertical axis shows the peak ratio value (the actual data with blue, the average with red, yellow and purple lines represent the ± 1 SD interval). The diagrams show the same range for comparison. The lower standard deviation of the bottom plot compared to the original (top) one is prominent.

Figure 13

Graphical comparison of the peak ratios for the 17 samples in two equivalent representations: The datasets are marked with the same color on both plots. Peak ratio is the ratio of the DNA content at the 4N peak to the DNA content of the 2N peak. This theoretically equals 2.0 because, during mitosis, a normal body cell first replicates/doubles its DNA content before dividing into two new daughter cells.

Figure 14

Graphical results of the linear regression for the more exact, ICM-fit dataset with the standard regression model: The data do not seem to align well with linear regression. Residuals seem to be non-normal. Removing hand-selected outliers based on overall image intensity did not improve the dataset considerably in this regard.

Figure 15

The DNA content of the first peaks of the samples (as a very rough approximation of the image intensity), the data points labeled with the sample indices. Samples #1, #12, and #18 are of lower intensity than the other samples. This is probably due to extended exposure to excitation light during multiple attempts to select the digitization parameters. These samples were treated as outliers for the rest of the process. Samples #7, #8, and #9 were not digitized in the expected quality (unfocused), and were omitted from the study altogether.