Impact of Stain Normalization on Pathologist Assessment of Prostate Cancer: A Comparative Study

Abstract

In clinical routine, the quality of whole-slide images plays a key role in the pathologist's diagnosis, and suboptimal staining may be a limiting factor. The stain normalization process helps to solve this problem through the standardization of color appearance of a source image with respect to a target image with optimal chromatic features. The analysis is focused on the evaluation of the following parameters assessed by two experts on original and normalized slides: (i) perceived color quality, (ii) diagnosis for the patient, (iii) diagnostic confidence and (iv) time required for diagnosis. Results show a statistically significant increase in color quality in the normalized images for both experts (p < 0.0001). Regarding prostate cancer assessment, the average times for diagnosis are significantly lower for normalized images than original ones (first expert: 69.9 s vs. 77.9 s with p < 0.0001; second expert: 37.4 s vs. 52.7 s with p < 0.0001), and at the same time, a statistically significant increase in diagnostic confidence is proven. The improvement of poor-quality images and greater clarity of diagnostically important details in normalized slides demonstrate the potential of stain normalization in the routine practice of prostate cancer assessment.

Keywords: Gleason score; color quality; digital pathology; prostate cancer; stain normalization.

Figure 1

Prostate cancer grade groups. Normal prostatic tissue is shown for reference. Grade Group 1 (GG1) is a low-grade tumor with well-formed individual glands. With increasing GG, glands become fused, disorganized, solid, cribriform, or they are not formed at all (GG4, GG5).

Figure 2

Stain normalization process for a prostate WSI. Original slide (a) and normalized slide (b) with zoomed-in views showing the effects at high magnification of color normalization with respect to target image (c).

Figure 3

Workflow adopted in this study: an example of original WSI (a) and the corresponding normalized slide (b) processed by STAINS tool. Both pathologists (P1 and P2) provided evaluations on stain color quality, diagnosis, confidence, and time for diagnosis, for both datasets, i.e., the original images (an example of pathologist’s evaluation is reported in green) and the corresponding normalized ones (an example of pathologist’s evaluation is reported in blue).

Figure 4

Boxplots of distributions of color quality values of original images (red) and normalized images (blue). The evaluation of the first pathologist (P1) is shown on the left, and the evaluation of the second pathologist (P2) is represented on the right. The plus symbol (+) indicates outliers in the distribution, while the asterisk (*) denotes a statistically significant difference between original and normalized images.

Figure 5

Visual performance of normalization process. In the first row, there are examples of tiles with suboptimal staining: reddish (sample #1), greyish color (sample #2) and weak staining (sample #3). The second row shows the corresponding normalized tiles.

Figure 6

Confusion matrices on Gleason Grade Group (GG). First row shows the intra-agreement of P1 (a) and P2 (b) on the GG values of the original and normalized images. Second row shows the inter-agreement (P1 vs. P2) on the original (c) and normalized (d) WSIs.

Figure 7

Comparison between time for diagnosis of the original and normalized WSIs. Boxplot of diagnosis time values of original (red) and normalized images (blue) according to the evaluations of the first pathologist (P1) on the left and second one (P2) on the right. The plus symbol (+) indicates outliers in the distribution, while the asterisk (*) denotes a statistically significant difference between original and normalized images.

Figure 8

Boxplots of confidence values. Overall distribution of confidence values of original (red) and normalized images (blue) according to the evaluations of the first pathologist (P1) on the left and second one (P2) on the right. The plus symbol (+) indicates outliers in the distribution, while the asterisk (*) denotes a statistically significant difference between original and normalized images.