Computer-assisted stereology and automated image analysis for quantification of tumor infiltrating lymphocytes in colon cancer

Abstract

Background: Precise prognostic and predictive variables allowing improved post-operative treatment stratification are missing in patients treated for stage II colon cancer (CC). Investigation of tumor infiltrating lymphocytes (TILs) may be rewarding, but the lack of a standardized analytic technique is a major concern. Manual stereological counting is considered the gold standard, but digital pathology with image analysis is preferred due to time efficiency. The purpose of this study was to compare manual stereological estimates of TILs with automatic counts obtained by image analysis, and at the same time investigate the heterogeneity of TILs.

Methods: From 43 patients treated for stage II CC in 2002 three paraffin embedded, tumor containing tissue blocks were selected one of them representing the deepest invasive tumor front. Serial sections from each of the 129 blocks were immunohistochemically stained for CD3 and CD8, and the slides were scanned. Stereological estimates of the numerical density and area fraction of TILs were obtained using the computer-assisted newCAST stereology system. For the image analysis approach an app-based algorithm was developed using Visiopharm Integrator System software. For both methods the tumor areas of interest (invasive front and central area) were manually delineated by the observer.

Results: Based on all sections, the Spearman's correlation coefficients for density estimates varied from 0.9457 to 0.9638 (p < 0.0001), whereas the coefficients for area fraction estimates ranged from 0.9400 to 0.9603 (P < 0.0001). Regarding heterogeneity, intra-class correlation coefficients (ICC) for CD3+ TILs varied from 0.615 to 0.746 in the central area, and from 0.686 to 0.746 in the invasive area. ICC for CD8+ TILs varied from 0.724 to 0.775 in the central area, and from 0.746 to 0.765 in the invasive area.

Conclusions: Exact objective and time efficient estimates of numerical densities and area fractions of CD3+ and CD8+ TILs in stage II colon cancer can be obtained by image analysis and are highly correlated to the corresponding estimates obtained by the gold standard based on stereology. Since the intra-tumoral heterogeneity was low, this method may be recommended for quantifying TILs in only one histological section representing the deepest invasive tumor front.

Keywords: Colon cancer; Heterogeneity; Image analysis; Stereology; Tumor infiltrating lymphocytes.

Fig. 1

Region of interest. For each tissue section the region of interest (ROI) was manually marked. Green line demarcates central tumor area (CA) and blue line the invasive area (IA), including the invasive front of the adenocarcinoma

Fig. 2

Workflow for stereology and image analysis. The regions of interest were outlined manually and the exact same areas were used for both technical approaches. The stereological analysis was performed, using a computer assisted software. The field of views were selected by the software by systematic random sampling, while the counting was carried out manually by the observer. The image analysis was performed automatic using, an image analysis algoritm

Fig. 3

Systematic, uniform random sampling of fields of vision using newCAST software. The yellow frame represents the current field of vision (FOV)

Fig. 4

Field of vision in a CD3-stained section magnified ×40. The density estimation was performed using the 2D unbiased counting frame with left and bottom edges, and their extensions, serving as exclusion lines (red), and with the upper and right edges of the frame as inclusion lines (green). Cell profiles were counted when completely inside the counting frame or partly inside the frame, provided that they did not touch the exclusion lines or their extensions. Thus, three cell profiles were counted (red crosses). The area fraction estimation was performed using the point grid. Points hitting CD3 positive cells = 2 (red ring) and points hitting tumor = 30, giving an area fraction of 0.07 in this field of vision

Fig. 5

Example of the processing of the image analysis. a) Part of a CD3 stained section. b) CD3+ lymphocytes labelled with the red color are counted by the software

Fig. 6

Correlation between cell counts as obtained by stereology and image analysis. a) Correlation for CD3+ tumor infiltrating lymphocytes (TILs) numerical density in the central area of the tumor. b) Correlation for CD3+ TILs numerical density in the invasive tumor front. c) Correlation for CD8+ TILs numerical density in the central area of the tumor. d) Correlation for CD8+ TILs numerical density in the invasive tumor front

Fig. 7

Correlation between area fractions as estimated by stereology and image analysis. a) Correlation for CD3+ tumor infiltrating lymphocytes (TILs) area fraction in the central tumor area. b) Correlation for CD3+ TILs area fraction in the invasive tumor front. c) Correlation for CD8+ TILs area fraction in the central tumor area. d) Correlation for CD8+ TILs area fraction in the invasive tumor front

Fig. 8

Bland Altman plots showing the differences in densities of CD3+ and CD8+ tumor infiltrating lymphocytes for the central and invasive area measured by image analysis. The horizontal red line corresponds to zero difference, the blue dashed line shows mean and the dashed red lines show ±1.96 standard deviation. a and b) Differences for densities of CD3+ and CD8+ tumor infiltrating lymphocytes (TILs) for all sections (n = 129). c and d) Differences for densities of CD3+ and CD8+ TILs for “the deepest invasive sections” (n = 43). e and f) Differences for densities of CD3+ and CD8+ TILs per tumor, where density is calculated as an average of the densities obtained from each of the three sections from each tumor (n = 43)