Standardized Method for Defining a 1-mm² Region of Interest for Calculation of Mitotic Rate on Melanoma Whole Slide Images

After permission for this study was obtained from the review board of our institution, we retrospectively searched our institutional pathology database and chose 30 melanoma cases with reported mitotic figures ranging from 0 to 28. As recommended by Azzola et al,¹⁶  the reported mitotic counts of our 30 melanoma cases were divided into 3 groups: group 1 (0 or 1 mitosis/mm²), group 2 (2–4 mitoses/mm²), and group 3 (≥5 mitoses/mm²). The WSIs for these 30 melanoma cases were created by digitally scanning the original H&E-stained glass slides at ×20 magnification with a ScanScope digital pathology system (Aperio, Vista, California) with SVS format. This viewer system was selected because we have been using that software at our institution since 2012. However, most commercially available systems provide similar digital tools.

Four monitors used in our institution, ranging from 24- to 38-inch screen size with different native resolutions and aspect ratios, were tested (Table). The 24-inch monitor with 1920 × 1080 native resolution (1080p) is the most commonly used in our institution. The other 3 are high-resolution monitors used in sign-out areas or attending physicians' offices: 27-inch with native resolution 2560 × 1440 (1440p), 32-inch with 3840 × 2160 (4K), and 38-inch with 3840 × 1600. Native resolutions of 1080p, 1440p, and 4K also represent 3 commonly used resolution settings. The screen size of a monitor is measured in inches from corner to corner diagonally. The aspect ratio and resolution can be changed from the display setting of the computer or the monitor control. Once resolution is selected, the aspect ratio is automatically adjusted.

Different viewing magnifications were explored for identifying the hot spot (screening magnification) and counting mitotic figures (counting magnification). In the ImageScope software, the viewing magnification can be changed continuously by dragging the zoom slider or scrolling a mouse wheel. The viewing area, or digital visual field, was defined as the area of the scanned specimen that could be seen on screen, a designation similar to the visual field in microscopy. At ×10 and ×20 magnifications, the maximum width and height of the visual field on the monitor were measured, and the corresponding viewing area was calculated for each monitor (Figure 1). Notably, ImageScope automatically adjusts the measurement calibration for viewing magnification, which facilitates measuring viewing areas. The viewable width and height were measured on the monitors at different resolution and display scale settings. The minimum number of screen views needed to evaluate a 1-mm² ROI on monitors was calculated at different resolutions at 100% scale setting (Table).

ImageScope provides annotation tools that can be used to outline ROIs. Users can draw free-form shapes and fixed shapes, including rectangles, squares, circles, and ellipses. Different approaches to drawing a 1-mm² ROI using these annotation tools were evaluated in terms of convenience and efficiency.

Mitotic figures were defined as the unequivocal presence of extensions of chromatin (condensed chromosomes) extending from a condensed chromatin mass, corresponding to either a metaphase or telophase figure.¹  For each WSI, the mitotic rate was evaluated by first finding the hot spot (ie, the region of the lesion containing the most mitotic figures) and then counting mitotic figures beginning in the hot spot and then extending to the immediately adjacent nonoverlapping viewing fields until an area of tissue corresponding to 1 mm² was assessed. In ImageScope, the counter tool can be used to mark and automatically count mitotic figures, and any area with brisk mitotic activity can be annotated for future review (Figure 2). If the invasive component involved an area smaller than 1 mm², the mitotic figures were enumerated and recorded in the entire available area.

Fixed-shape annotations with 500 × 500-μm squares or circles were applied depending on the specimen orientation during mitotic figure counting, because this approach is able to achieve convenient annotation and efficient counting while ensuring easy transition from traditional glass slides, which will be discussed later in the Results section. One pathologist performed mitotic counting and recorded the counting time on 14 cases using available glass slides as well as WSIs on 4 monitors: 1080p (24-inch), 1440p (27-inch), 4K (32-inch), and 3840 × 1600 (38-inch). The 5 evaluation methods were applied in random order for mitotic counting with a washout period of 48 hours or greater. The counting times and mitotic rates from these 5 methods were compared. Comparison was also made between the mitotic rates obtained from recounting the glass slides and those listed in the pathology reports. Two pathologists (including the one above) reviewed the WSIs from all 30 cases selected for this study inclusive of the above-mentioned 14 cases using 4K monitors. For each melanoma case, the mitotic rate obtained from the WSI using 4K monitors was compared with that obtained from the glass slide as noted in the pathology report. For comparisons of mitotic rates and counting times, paired t tests were performed, and a 2-tailed P value <.05 was considered statistically significant. The interobserver agreement within the same modality and the agreement between glass slide and WSI modalities were evaluated using Pearson correlation coefficients (γ) with a 2-tailed P value <.05 considered statistically significant. All statistical analyses were performed using SPSS (Version 22; IBM SPSS Inc, Armonk, New York).

Both resolution and display scale affect the viewable area. Under a fixed scale (such as 100%), the same resolution yields a fixed viewable area irrespective of the size of the monitor (Table). For a fixed-resolution setting, if the display scale is increased from 100% to 150%, the viewable width and height will decrease to 67% of their original values and the viewing area will shrink to 45% of its original value accordingly (data not shown).

At ×20 viewing magnification and 100% scale, the full-screen viewing areas were 0.47, 0.88, 1.81, and 1.42 mm², respectively (Table), for the 4 monitors tested (24-inch at 1080p, 27-inch at 1440p, 32-inch at 4K, and 38-inch at 3840 × 1660). For a 1080p monitor, at least 3 screen views were needed to see a 1-mm² ROI; for a 1440p monitor, more than 1 screen view was needed; and for 4K and 3840 × 1600 monitors, only 1 screen view was needed.

At ×10 viewing magnification and 100% scale, the viewing areas on full screen were 1.88, 3.52, 7.24, and 5.68 mm², respectively, at 1080p, 1440p, 4K, and 3840 × 1600 resolutions; these areas were 4 times the viewing areas at ×20 viewing magnification.

We found that continuous switching between ×10 and ×20 viewing magnifications was an efficient method for spotting mitotic figures and allowed efficient screening of the entire tumor area in a short time. After the hot spot was located, using ×20 to ×40 viewing magnifications ensured that mitotic figures could be properly identified and counted.

The ROI was freely delineated with the free-form shape annotation tool. In addition, several ROIs could be outlined at the same time and the total area calculated, as shown in the example in Figure 3, in which the total area of the invasive component of the tumor was less than 1 mm² and 1 mitotic figure was identified.

In some cases, such as the one shown in Figure 4, the invasive component of the tumor had an extended contiguous region, and thus a single 1000 × 1000-μm square could be used to annotate a 1-mm² ROI. However, in most of the cases in our series, a single-square annotation was not applicable because of the presence of other structures, such as skin adnexa or normal dermis; therefore, several fixed shapes were needed to enclose the ROI. Figure 5, A, shows an example in which a single square was not suitable for demarcating a 1-mm² ROI.

Figure 5, A through C, shows annotations of an image of a skin sample oriented obliquely at ×10 viewing magnification on a full 4K screen. Because ImageScope can rotate a WSI only by 90° or 180°, a single 1-mm² annotation does not fit the ROI irrespective of the annotation setting (Figure 5, A). A 4-square (Figure 5, B) or 5-circle annotation (Figure 5, C) is more suitable. Figure 6, A through F, shows different fixed-shape annotations commonly used to delineate a 1-mm² ROI at ×10 (Figure 6, A and B) and ×20 (Figure 6, C through F) viewing magnifications on a full 4K screen. The width and height of the rectangles, squares, and circles needed to delineate a 1-mm² ROI are listed in Table 1 of the supplemental digital content at https://meridian.allenpress.com/aplm in the October 2021 table of contents.

We also tested a 24-inch 1080p monitor, because this size is the most common in our institution. Figure 7, A and B, illustrates that a 1080p monitor is barely large enough to show two 500 × 500-μm circles (Figure 7, A) or two 500 × 500-μm squares (Figure 7, B) at ×20 magnification. To delineate a 1-mm² ROI, the user needs to view 2 to 3 screens at 1080p.

As for when to use 4 squares or 5 circles, Figure 8, A and B, shows that for more extended tumor regions, the use of 4 squares is more convenient, whereas for obliquely oriented specimens or irregular tumor regions, circles are more adaptable (Figure 5, C).

For the 14 cases with available glass slides, the mitotic rate obtained from recounting the slides was on average 7.2 mitoses/mm² (SD = 5.5 mitoses/mm²), similar to the numbers obtained from the pathology reports (7.7 mitoses/mm² [SD = 6.8 mitoses/mm²]; 95% CI, −1.3 to 2.3; P = .56). The interobserver correlations using glass slides exhibited good agreement among pathologists (γ = 0.89, P < .001). The mitotic rates from WSIs were similar for 1080p, 1440p, 4K, and 3840 × 1600 monitors.

The 4K monitor yielded a mean mitotic counting time of 48.6 seconds (SD = 12.4 seconds), similar to the 3840 × 1600 monitor (50 seconds [SD = 10.9 seconds], P = .16). Both 4K and 3840 × 1600 monitors produced statistically shorter counting time than the 1440p monitor (55.5 seconds [SD = 13.8 seconds], P < .001 and P = .004, respectively). The 1080p monitor produced a mean mitotic counting time of 65.4 seconds (SD = 20.5 seconds), significantly longer than the 1440p (P = .01), 4K (P < .001), and 3840 × 1600 (P = .001) monitors. The mean counting time with glass slides was 69.1 seconds (SD = 22.0 seconds), which was longer than those produced with WSI on 1440p, 4K, or 3840 × 1600 monitors (P = .001, P < .001, and P < .001, respectively), but similar to that produced on a 1080p monitor (P = .16).

Of the 30 cases chosen for this study, based on the mitotic rates from the pathology reports, there were 4 cases classified in group 1 (0 or 1 mitoses/mm²), 8 cases in group 2 (2–4 mitoses/mm²), and 18 cases in group 3 (≥5 mitoses/mm²) (Figure 9). Two cases in group 2 changed groups when counted on WSI. One case was downgraded from group 2 on glass slide to group 1 on WSI and the other case was upgraded from group 2 on glass slides to group 3 on WSI, as evaluated by both pathologists. An additional case was upgraded to group 3 by one pathologist after using WSI, but was kept in group 2 by the second pathologist. The mean mitotic rate of the 30 cases was 9.8 (SD = 9.1) and 9.4 (SD = 8.6) mitoses/mm² on digital images as evaluated by the 2 pathologists, compared with 7.5 (SD = 6.7) mitoses/mm² noted in pathology reports. In this series, evaluation via WSIs yielded a higher mitotic rate than did traditional H&E-stained slides by both pathologists (95% CI, 1.05–3.55; P < .001; and 95% CI, 0.84–2.96; P < .001, respectively). There was a tendency that more mitotic figures were counted for group 2 and group 3 cases with WSI. The cases in group 3, the high-mitotic-rate group, showed the highest variance in mitotic rate difference between WSI and glass slides (Figure 9).

The interobserver correlation of WSI-based mitotic counting exhibited substantial agreement between the 2 pathologists (γ = 0.99, P < .001). The correlation between the 2 counting modalities (glass slides and WSI) demonstrated good agreement as well for both pathologists (γ = 0.95, P < .001, and γ = 0.96, P < .001).

Our findings showed that of the 4 monitors examined, a 32-inch 4K monitor in its native resolution and 100% scale appeared to be the best for mitotic figure counting within a 1-mm² ROI in melanoma. When WSIs were viewed in ImageScope viewer, a hot spot could be efficiently located under ×10 to ×20 viewing magnifications during screening and mitotic figures accurately identified during counting at ×20 to ×40. A precise 1-mm² ROI could be produced using a square annotation measuring 1000 × 1000 μm, 4 square annotations each measuring 500 × 500 μm, or 5 circle annotations each with height and width (diameter) of 500 μm. Interobserver correlation using WSI and intermodality correlation between WSIs and glass slides both demonstrated substantial agreements, showing that our method of counting mitotic figures on WSIs was equivalent to using glass slides.

In many tumors, quantification of mitotic counts is critical in making an accurate and reliable diagnosis. In melanoma, mitotic count is an essential part of the pathology report.²  In other tumors, for example gastrointestinal neuroendocrine tumors,¹⁷  invasive ductal carcinomas of the breast,¹⁸  meningiomas,¹⁹  and gastrointestinal stromal tumors,²⁰  mitotic rate is crucial in the determination of the tumor's histologic grade. Standardization of the counting practice is an important step in achieving accurate and consistent mitotic rate counting.^4,21 

According to the 8th edition of the American Joint Committee on Cancer staging system for melanoma and the College of American Pathologists protocol, a hot spot should be identified before mitotic figures are counted.^2,3  With traditional microscopy, tissue sections on glass slides are generally screened at ×100 or ×200 magnification (×10 or ×20 objective lens) to identify the hot spot and then at ×400 magnification (×40 objective lens) to count mitotic figures. Our findings confirm that when WSIs are viewed in the ImageScope viewer, viewing magnifications from ×10 to ×20 for screening and from ×20 to ×40 for counting can help locate a hot spot and identify mitotic figures accurately and efficiently. When reviewing slides scanned at ×20 magnification, it may be difficult to distinguish a mitotic figure from an apoptotic body or a hyperchromatic nucleus. Under such circumstances, based on our experience, the digital ×40 zooming may help with the distinction, particularly in high-resolution monitors. One advantage of digital pathology is that in viewers such as ImageScope, a screen view can be conveniently magnified or shrunk in a continuous fashion by scrolling a mouse wheel; there is no need to rotate the objectives and then readjust the focus, which is required when glass slides are viewed under a microscope.

With digital viewers, as with microscopes, the higher the viewing magnification, the smaller the visual field. A monitor with a higher resolution can display a larger visual field for a given viewing magnification. Of the monitors we tested, the 32-inch monitor with 4K resolution is optimal for counting mitotic figures with a 1-mm² region for melanoma. The 38-inch monitor with 3840 × 1600 resolution is fairly uncommon, although it is almost as good as the 4K monitor in terms of counting time. The 1440p monitor is a good alternative considering its reasonable counting efficiency and low cost. The 1080p monitor is not optimal given the significantly longer counting time. It should be pointed out that monitor size is also a major factor to consider. With a given resolution (4K, for example), larger screen dimension increases the size of native pixels and makes images appear larger, similar to the experience of watching a movie on a cell phone versus the large screen of a movie theater. Our experience shows that on a 27-inch monitor at 100% scale and 4K resolution, the text interface is almost unreadable, and consequently the display scale has to be increased to 150%, which decreases the viewing area by 55% and diminishes the advantage of the 4K resolution. Therefore, at 4K resolution, monitors with a 32-inch screen or larger are justified. A 27-inch monitor is a practical size at 1440p resolution, however, for achieving good image quality and text readability, based on our experience.

The advantage of annotating ROIs with free-form shapes is obvious: this approach provides the user with the freedom to delineate regions of any shape. Free-form shapes are useful when an irregular area is examined, such as blood vessels or adnexal structures intermixed with melanocytes. The disadvantage of free-form annotation is that the user has to demarcate the interesting area by hand, and during the process, any error will necessitate the user's restarting the whole process. Thus, the use of free-form shapes can make ROI delineation a time-consuming and labor-intensive task. Moreover, in practice, the user can only delineate an area of approximately, but almost never precisely, 1 mm². For this reason, delineation with free-form shapes does not appear to be feasible for routine application, even though it may be useful for delineating and measuring an irregular small invasive component. If the area of the invasive component is smaller than 1 mm², all mitotic figures in the entire area are counted.

Theoretically, if a hot spot is located, using one rather than multiple annotations should be time-saving. However, the specimen section may be obliquely arranged on the slide or have an irregularly shaped tumor region, in which case more than one annotation will be required. The process of manually adjusting the width and height of annotations is time-consuming. Mathematically, the smaller the annotations and the more annotations performed, the better the annotated areas will match irregular outlines of the tumor sections. In practice, however, smaller numbers of annotations are more efficient. Our experience indicates that annotations with 4 squares or 5 circles can achieve a good balance between efficiency and flexibility. In many cases, all of the annotations can fit on a 4K screen, and we can easily count mitotic figures in a 1-mm² region without having to move around the tissue.

Regarding the choice between 4 squares and 5 circles, a 4-square approach seems preferable because it is easier for the user to alter the position of each square without leaving a blind spot than would be the case with circles. However, for nonoriented specimens or lesions with irregularly shaped tumor regions, circles can be more easily adjustable than squares. The 5-circle annotation also mimics the traditional approach in microscopy of examining 4 to 5 high-power fields (×400) to cover a 1-mm² area. It is noteworthy that the width and height settings for 4 squares and 5 circles to precisely enclose a 1-mm² area are remarkably close, at 500 × 500 μm and 505 × 505 μm, respectively. Therefore, if the width and height are just set and kept at 500 × 500 μm in practice, one can easily switch between the 4 squares and 5 circles annotations without needing to reset the numbers.

An appropriate reading environment may be beneficial to improving the efficiency of mitotic counting. In our study, the counting time through conventional microscopy was longer than those achieved via WSI with 1440p, 4K, and 3840 × 1600 monitors, but similar to that with a 1080p monitor. Interestingly, one previous study reported the mitotic counting time for WSI was longer than that of conventional microscopy²² ; the authors explained that cumbersome software requiring multiple mouse clicks for annotating one mitotic figure and the lack of a counter tool may have contributed to the longer counting time. With our digital pathology software, one mitotic figure can be marked with a single mouse click, and there is no need to manually count mitotic figures. The 4K and 3840 × 1600 monitors used in our study were able to reduce mouse scroll and clicks because they are able to cover more than 1 mm² of ROI on one full screen. In addition, our pathologists were experienced with WSI. All the above may have facilitated the counting process and helped shorten the counting time. It is worth mentioning that the time needed for reviewing WSI is also dependent upon the speed of the Internet, as it takes time for digital images to be uploaded to the user terminal from the pathology database.⁷ 

Our study showed that mitotic rates obtained with WSIs were higher than those from glass slides. A possible reason may be that, with the digital approach, mitotic figures can be easily marked during slide screening, which may increase the possibility of finding the most mitotically active area, that is, the best hot spot. The higher count on WSI may also reflect a bias of increased counting effort in this study compared with clinical routine.

We classified mitotic rates into 3 groups based on the previously reported finding that significant difference in survival exists among each of the tumor mitotic rate groups.¹⁶  In this study, the mitotic rates counted using WSI produced highly consistent groupings with those obtained using glass slides. All the cases in group 1 (4 of 4) and group 3 (18 of 18) remained in the same group after counting with WSI. Two of 8 cases in group 2 changed groups, with one moving to group 1 and the other moving to group 3. The clinical significance of such group change after using WSI is unclear considering our study is limited by a small series of cases. To date, all the studies related to the clinical significance of mitotic rate on melanoma have been performed on traditional H&E-stained glass slides.^23–25  In this light, future studies involving large samples are needed to shed light on how mitotic rates counted on WSI may affect tumor staging and/or survival. A recent study reported a first evaluation of WSI for mitotic figure prognostic grading using canine melanoma.²⁶  This prognosis-associated topic, however, is beyond the main purpose of this study. It is also noteworthy that mitosis-related studies are all prone to complication by the variability in observing mitotic figures.²⁷  Our methodology for ROI delineation may therefore assist in improving reproducibility when digital counting is used. Moreover, all annotations and counting marks can be conveniently saved in digital slides, which is beneficial to improving interobserver reproducibility and diagnostic consensus. Such annotated ROIs have the potential to be used as templates to train and validate artificial algorithms for identifying mitoses and ultimately help with future developments of automated computational tools targeting mitotic figure identification and hot-spot mapping on H&E-stained slides.

It should be pointed out that only 4 monitors in relatively common resolutions were tested in our study among the numerous monitors of various resolutions available on the market. A second limitation is that our study did not examine the image acquisition component (ie, scanner). All the WSIs in our study were scanned with the Aperio ScanScope digital pathology system at ×20 magnification, which is routinely used in our institution. We have had 4 different Aperio scanners during a period of 5 years and have not noticed significant differences among them. However, viewable areas of WSIs produced with digital pathology platforms from other vendors at different scanning magnifications can be easily evaluated with our methodology described in the Methods section. For users of other monitor resolutions, the same methodology for evaluating viewable areas can also be applied, by which the preset annotations (circles and squares) can be drawn onscreen under the user's specific resolution to confirm the adequacy of the viewable area.

In the evaluation of mitotic rate in gastrointestinal neuroendocrine tumors, screening of a 2-mm² area is needed. In addition to the previously mentioned setting of 8 squares of 500 × 500 μm, we may also consider using 4 squares each measuring 707 × 707 μm to cover a required 2-mm² screening area if the specimen has an adequate contiguous area. For a gastrointestinal stromal tumor, however, a total area of 5 mm² is needed for screening. Therefore, 5 squares of 1000 × 1000 μm each may be used; if the tissue is fragmented, then annotating with 10 squares in a 707 × 707-μm setting may be considered. The counter tool provided with the digital pathology software is even more useful in these larger screening areas as it avoids the error-prone manual counting.

Our study shows that mitotic figure counting in melanoma using WSIs is equivalent to using glass slides and can be efficiently done in real practice. In terms of annotation methodology, we recommend fixed-shape annotations with 4 squares or 5 circles in a setting of 500 × 500 μm to cover a 1-mm² region. The pathologist can easily enclose the ROI of the tumor and effectively match up irregular tumor regions with position adjustments of the 4 squares or 5 circles. If the tumor has a large contiguous area, a single-square annotation of 1000 × 1000 μm can be used. Our methodology can be potentially extended to calculating mitotic rate in other tumors.

Stephanie Deming, ELS, senior scientific editor, Scientific Publication Services, Research Medical Library, The University of Texas MD Anderson Cancer Center, assisted us with editing of this manuscript.