Deep learning based tumor-stroma ratio scoring in colon cancer correlates with microscopic assessment

Abstract

Background: The amount of stroma within the primary tumor is a prognostic parameter for colon cancer patients. This phenomenon can be assessed using the tumor-stroma ratio (TSR), which classifies tumors in stroma-low (≤50% stroma) and stroma-high (>50% stroma). Although the reproducibility for TSR determination is good, improvement might be expected from automation. The aim of this study was to investigate whether the scoring of the TSR in a semi- and fully automated method using deep learning algorithms is feasible.

Methods: A series of 75 colon cancer slides were selected from a trial series of the UNITED study. For the standard determination of the TSR, 3 observers scored the histological slides. Next, the slides were digitized, color normalized, and the stroma percentages were scored using semi- and fully automated deep learning algorithms. Correlations were determined using intraclass correlation coefficients (ICCs) and Spearman rank correlations.

Results: 37 (49%) cases were classified as stroma-low and 38 (51%) as stroma-high by visual estimation. A high level of concordance between the 3 observers was reached, with ICCs of 0.91, 0.89, and 0.94 (all P < .001). Between visual and semi-automated assessment the ICC was 0.78 (95% CI 0.23-0.91, P-value 0.005), with a Spearman correlation of 0.88 (P < .001). Spearman correlation coefficients above 0.70 (N=3) were observed for visual estimation versus the fully automated scoring procedures.

Conclusion: Good correlations were observed between standard visual TSR determination and semi- and fully automated TSR scores. At this point, visual examination has the highest observer agreement, but semi-automated scoring could be helpful to support pathologists.

Keywords: Artificial intelligence; Colon carcinoma; Computational pathology; Deep learning; Tumor–stroma ratio; Visual estimation.

Fig. 1

An example of the workflow output for semi-automated scoring algorithm. In (A) H&E stained sections in the spot chosen by microscopic assessment. (B) The first step was making a segmentation output, before in (C) the class labels can be displayed.

Fig. 2

The output from the full automated workflow. In (A) the tumor bulk is annotated. In (B) the heatmap is created. The biggest dot corresponds with the highest stroma-percentage (TSR-1), the second biggest with the second highest (TSR-2), etcetera. In (C) the class output of the highest spot (TSR-1), in (D) the second highest spot (TSR-2) and in (E) the third highest spot (TSR-3).

Fig. 3

Bubble plot of assessed stroma percentage of the 75 cases. In (A) for the visual (consensus) scores plotted against the semi-automated scores. The lines indicate the cut-off between stroma-low and stroma-high. Green dots indicate cases where agreement was reached and dots in red indicate disagreement. In (B) the visual scores plotted against the highest fully automated scores (TSR-1), and in (C) the semi-automated scores plotted against the highest fully automated scores (TSR-1).

Fig. 4

Three examples for the semi-automated output. (A) Agreement between visual estimation and semi-automated score (20% stroma), (B) agreement between visual estimation and semi-automated score (60% stroma), (C) by visual estimation 50% stroma was scored, 80% stroma was the semi-automated outcome due to not always recognized tumor cells.