Clinical evaluation of deep learning-based risk profiling in breast cancer histopathology and comparison to an established multigene assay

Abstract

ABSTRACTA significant proportion of oestrogen receptor (ER)-positive and human epidermal growth factor receptor 2 (HER2)-negative early breast cancer patients are categorised as intermediate risk based on classic clinicopathological variables, thus providing limited information to guide treatment decisions. The Prosigna assay is one of the established prognostic multigene assays in clinical practice for risk profiling. Stratipath Breast is a novel deep learning-based image analysis tool that utilises haematoxylin and eosin (HE)-stained histopathological images for risk profiling. In this study, we aimed to evaluate the Stratipath Breast tool for image-based risk profiling and compare it with the Prosigna assay. In a real-world breast cancer case series comprising 234 invasive tumours from patients with early ER+/HER2-breast cancer, clinically intermediate risk and eligible for chemotherapy, clinicopathological data including Prosigna results and corresponding HE-stained tissue slides were retrieved. The digitised HE slides were analysed by Stratipath Breast. Our findings showed that the Stratipath Breast analysis identified 49.6% of the clinically intermediate tumours as low risk and 50.4% as high risk. The Prosigna assay classified 32.5%, 47.0% and 20.5% tumours as low, intermediate and high risk, respectively. Among Prosigna intermediate-risk tumours, 47.3% were stratified as Stratipath low risk and 52.7% as high risk. In addition, 89.7% of Stratipath low-risk cases were classified as Prosigna low/intermediate risk. The overall agreement between the two tests for low-risk and high-risk groups was 71.0%, with a Cohen’s kappa of 0.42. For both risk profiling tests, grade and Ki67 differed significantly between risk groups. In conclusion, for the first time, we here present the results from a clinical evaluation of image-based risk stratification and show a considerable agreement to an established gene expression assay in routine breast pathology. The findings demonstrate that image-based risk profiling may aid in the identification of low-risk patients who could potentially be spared adjuvant chemotherapy.