Figure 1.
Eye-tracking scan paths of fixated nuclei.
The high resolution computer screen, on which the above described slide presentation was displayed, was interfaced with an eye-tracking device, which allowed to follow the eye movements of the pathologists and hence to record exactly which nuclei they looked at and for how long. Scan paths and attention maps were computed with the aid of SMI BeGaze Analysis software. All nuclei that were fixated for a minimum of 100 milliseconds were recorded. The hatched green circles depict the location of the fixation within the HPF and the size of the circle the duration of the fixation. The lines between the circles depict the scan path. All nuclei that were selected by the eyetracking program were later analyzed by Photoshop-based image analysis in terms of nuclear size, hyperchromasia, heterochromasia and roundness.
Figure 2.
Slide show of HPFs before different architectural backgrounds.
A slide presentation was displayed on a 24″ computer screen that depicted HPFs of 20 different prostate carcinomas in central round windows before the background of a low power image of the tumor architecture. What the pathologists did not know was that from each of the 20 prostate carcinomas, two HPFs were shown, once before a low power image with well-formed tubular structures (corresponding to a combined Gleason grade 2–3) and again some time later before a low power image showing solid tumor architecture (corresponding to a combined Gleason grade 4–5). The presentation was automatically timed to show each slide fir exactly 8 seconds. Pathologists were asked to assign nuclear grades for each HPF. This part of the presentation lasted for a little over 10 minutes. Immediately after that, 40 slides were shown that depicted 10 random nuclei from each of the displayed HPFs, arranged in a 2×5 matrix (see figure 2).
Figure 3.
10 random nuclei isolated out of their architectural context.
Ten nuclei were randomly selected from each HPF displayed in the first part of the presentation (see figure 2), yet this time isolated out of their architectural context, and displayed in an orderly arranged 2×5 matrix on the computer screen for a duration of 8 seconds. Pathologists were asked to assign nuclear grades for each group of 10 nuclei. Later, these same nuclei were analyzed by Photoshop-based image analysis in terms of nuclear size, hyperchromasia, heterochromasia and roundness of the nuclear contour.
Figure 4.
Nuclear grades assigned to prostate carcinomas are biased by the architectural growth pattern.
The y-axes of the nine graphs depict the nuclear grades assigned by each of 20 pathologists on HPFs that were displayed on a computer screen for 8 seconds. Each line shows the two grades assigned by one pathologist for the same HPF depicted before a low grade architecture, rich in tubules (“tub”) and a high grade, solid architecture (“solid”). Note that for every pathologist, lower nuclear grades were assigned when the HPFs were depicted before a tubule-rich carcinoma and higher nuclear grades when the HPFs were presented before a solid carcinoma. The left panels show data for all 20 pathologists (upper panels), for 12 board-certified pathologists (“faculty”, middle panel) and for 8 residents (lower panel). The right three panels show the data calculated on those pairs of HPFs where the background images differed by more than 1.5 Gleason points and the middle panels the data ro those HPFs where the background images differed by only 1.5 Gleason points.
Figure 5.
The architectural bias influences how pathologists select nuclei (eye-tracking experiments).
Nuclear morphometric features of nuclei selected during eye-tracking experiments. Each line shows the morphometric features for nuclei selected by one pathologist for HPFs of each case then displayed either before a low grade architecture, rich in tubules (“tub”) and a high grade, solid architecture (“solid”). Note that for every pathologist, larger (size, upper left panel), darker (hyperchromasia, lower left panel), and coarser nuclei (heterochromasia, upper right panels) were fixated when the HPFs were shown before a solid carcinoma. The differences were statistically significant at the P<0.01 level (size) and the P<0.001 level (chromasia). In contrast, nuclear selection appeared not to be based on nuclear roundness (lower right panel, not significant).