Computational pathology improves risk stratification of a multi-gene assay for early stage ER+ breast cancer

Abstract

Prognostic markers currently utilized in clinical practice for estrogen receptor-positive (ER+) and lymph node-negative (LN-) invasive breast cancer (IBC) patients include the Nottingham grading system and Oncotype Dx (ODx). However, these biomarkers are not always optimal and remain subject to inter-/intra-observer variability and high cost. In this study, we evaluated the association between computationally derived image features from H&E images and disease-free survival (DFS) in ER+ and LN- IBC. H&E images from a total of n = 321 patients with ER+ and LN- IBC from three cohorts were employed for this study (Training set: D1 (n = 116), Validation sets: D2 (n = 121) and D3 (n = 84)). A total of 343 features relating to nuclear morphology, mitotic activity, and tubule formation were computationally extracted from each slide image. A Cox regression model (IbRiS) was trained to identify significant predictors of DFS and predict a high/low-risk category using D1 and was validated on independent testing sets D2 and D3 as well as within each ODx risk category. IbRiS was significantly prognostic of DFS with a hazard ratio (HR) of 2.33 (95% confidence interval (95% CI) = 1.02-5.32, p = 0.045) on D2 and a HR of 2.94 (95% CI = 1.18-7.35, p = 0.0208) on D3. In addition, IbRiS yielded significant risk stratification within high ODx risk categories (D1 + D2: HR = 10.35, 95% CI = 1.20-89.18, p = 0.0106; D1: p = 0.0238; D2: p = 0.0389), potentially providing more granular risk stratification than offered by ODx alone.

Fig. 1. Illustration of the overall workflow for the experimental design.

a Three deep-learning models: (a1) a CNN, (a2) a pixel2pixel GAN, and (a3) a U-Net model, were trained to detect mitosis, nuclei, and tubules, respectively. b Tumor tiles were exhaustively extracted from the tumor regions delineated by an experienced breast pathologist. c After detection of mitosis, nuclei, and tubules, quantitative patient-level features were extracted to describe mitotic rates, nuclear pleomorphism, and tubule formation, respectively. d The four most prognostic features were selected from each feature category by their association with disease-free survival (DFS) using a Cox regression model. e The top features identified from individual feature families were ensembled to train a final combined Cox proportional hazards model to stratify the ER+ and LN− breast cancer patients into high- and low-risk categories on the training set D1 with group differences assessed by two-sided log-rank test. f The prognostic model was subsequently locked down (g) and evaluated on two independent validation cohorts, D2 and D3 with the differences between high- and low-risk categories measured by two-sided log-rank test.

Fig. 2. Representative H&E WSI comparison of a recurrent (top row) and a censored (bottom row) patient.

The first column (a, f) shows the original WSI with the pathologist-annotated tumor region. The second column (b, g) illustrates the distribution of mitotic counts on the WSI with warmer color in the scale bar indicating a higher mitosis number. The third column (c, h) is a magnified view of a tumor tile. The fourth column (d, i) demonstrates the top-identified nuclear feature, which quantifies the number of connected nuclei clusters (connected in green line). The fifth column (e, j) shows the tubule feature “ratio of tubule nuclei count to non-tubule nuclei count” with tubule nuclei highlighted in cyan.

Fig. 3. Prognostic performance of IbRiS on D1-D3.

KM curve estimates for DFS for IbRiS^H (red) versus IbRiS^L (blue) across D1–D3 (a–c), with IbRiS^H demonstrating a significantly worse prognosis compared to IbRiS^L on D1, D2, and D3 using two-sided log-rank approach.

Fig. 4. Prognostic performance of IbRiS within individual ODx risk category in D1-D2.

KM curve estimates for DFS for IbRiS^H (red) versus IbRiS^L (blue) in the low, intermediate, and high ODx risk categories, respectively across D₁₊₂ (a, d, g), D1 (b, e, h) and D2 (c, f, i) with the differences between the risk categories assessed by two-sided log-rank test. IbRiS was significantly prognostic within high ODx risk category for both D1 and D2.

Fig. 5. Prognostic performance of IbRiS within individual histology grade in D1-D3.

KM curve estimates for DFS for IbRiS^H (red) versus IbRiS^L (blue) in the low, intermediate, and high histologic grades, respectively across D₁₊₂₊₃ (a, d, h), D1 (b, e, i), D2 (c, f, j) and D3 (g, k) with the differences between the risk categories assessed by two-sided log-rank test. IbRiS was significantly prognostic within high histologic grade groups for all three cohorts.