Abstract 6233: A deep learning pipeline to capture the prognostic immune responses in lymph nodes of breast cancer patients

Abstract

Capturing tumor infiltrating leucocytes (TILS) and systemic immune responses in breast cancer informs disease progression and optimal treatment management. We have previously shown that morphological alterations in axillary lymph nodes (LNs), namely the formation of germinal centers in cancer-free LNs, adds prognostic value to TILs in triple negative breast cancer patients (TNBC) for the development of distant metastasis. Extending manual assessment of LNs beyond the detection of cancer requires the integration of robust deep learning pipelines into the digital pathology workflow. In this retrospective study, we used 1,100 Haematoxylin & Eosin-stained (H&E) Whole Slide Images (WSI) from Guy’s Hospital (London, UK) of metastatic and cancer-free LNs from 151 patients (100 N+) enriched for triple-negative or HER2-positive breast cancer to implement a supervised deep learning pipeline. A subset of 114 WSI, along with 5 breast cancer LN WSIs from each of Barts Hospital (London, UK) and Tianjin University Hospital (Tianjin, China), and 5 head and neck squamous cell carcinomas LN WSI (Guy’s Hospital) were used to develop, train and evaluate the segmentation task. For training Fully Convolutional Networks (FCNs), WSIs manually annotated for both germinal centers and sinuses formed a ground-truth set. Three FCNs were implemented: (i) a standard U-Net architecture; (ii) a U-Net model with an attention gate mechanism; and (iii) a multiscale-U-Net network (MSA-U-Net) that encodes, in parallel, a feature representation of the image at multiple resolutions. The MSA-U-Net achieved the best performance with an average dice score of 0.85 for germinal centers and 0.75 for sinuses. In comparison, the average dice score amongst 4 pathologists assessing 25 LN WSI for germinal centers and sinuses, was 0.67 and 0.61 respectively, demonstrating the robustness of the MSA-U-Net model. To quantify germinal centers and sinuses in LNs across the entire cohort, the trained MSA-U-Net was used in an inference step on all 1,100 WSI. The detected morphological features were initially localized within LNs using image thresholding and contouring techniques, and quantitatively assessed based on their number, area, shape, and Shannon diversity. We found significant morphological differences in metastatic and cancer-free LNs between N0 and N+ patients, with the latter displaying larger germinal centers with more irregular shapes especially in their metastatic LNs. In addition, we found differences in the Sinus area between LNs containing GCs and those without. Here, we propose a robust deep learning pipeline based on a multiscale FCN framework to automatically detect, localize and quantify histopathological immune features in WSI of LNs. By applying our pipeline to LNs of cancer patients, such as breast or head and neck, in prospective studies or clinical trials, we will further evaluate their prognostic and predictive values. Citation Format: Gregory Verghese, Mengyuan Li, Amit Lohan, Nikhil Cherian, Swapnil Rane, Fangfang Liu, Aekta Shah, Pat Gazinska, Selvam Thavaraj, Amit Sethi, Anita Grigoriadis. A deep learning pipeline to capture the prognostic immune responses in lymph nodes of breast cancer patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 6233.

Computational pathology in cancer diagnosis, prognosis, and prediction – present day and prospects

“…In head and neck malignancies, AI-based methodologies for detecting TILs have shown superiority in separating patients' outcomes compared with manual TILs scoring and a better delineation of stromal TILs from native lymphocytes in lymphoepithelial tissues such as the oropharynx [46]. Our group has repeatedly demonstrated that assessing immune responses in lymph nodes of triple-negative breast cancer (TNBC) patients adds prognostic value [43,47,48]. In particular, the formation of germinal centres and an expanded sinus surface area in a patient's lymph nodes are associated with longer intervals of disease recurrence [43].…”

“…In particular, the formation of germinal centres and an expanded sinus surface area in a patient's lymph nodes are associated with longer intervals of disease recurrence [43]. By implementing a multi-scale CNN-based framework, germinal centres and sinuses on digitised H&E-stained axillary lymph node sections were robustly quantified, comparable with inter-pathologist assessments [48,49].…”

“…Standardised public datasets have become the norm for developing and evaluating AI models in other fields, such as the ImageNet Large Scale Visual Recognition Challenge in computer vision [1,75,76]. As computational pathology scales, public large-scale datasets [48] may become more readily available and virtual biobanks comprising large multinational cohorts can be created to advance reliable prospective and retrospective tissue-based cancer research globally. Increased robustness, consistency, and validity of AI models will further fuel confidence in computational pathology.…”

B Cells in Breast Cancer Pathology

“…Morphological features indicative of B cell activation, namely the formation of GC in LNs and TLS in the primary tumour, are ideally suited for robust quantification using deep learning approaches on digitised whole slide images of H&E-stained tumours and LNs. We and others have begun to implement such deep learning frameworks, and demonstrated that an increased frequency of GCs in cancer-free and involved LNs is associated with longer time to distant metastasis in TNBC patients [86]. Tools to automatically annotate TLS in breast carcinomas have not yet been developed.…”

Spatial analyses of immune cell infiltration in cancer: current methods and future directions: A report of the International Immuno‐Oncology Biomarker Working Group on Breast Cancer

“…Superpixels is another method that segments tumors into raster cells according to similarities in adjacent pixel colors or other features [17,18]. Finally, approaches can be implemented that analyze tumors on different scales (multiscaling) or to optimize raster cell size as a hyperparameter, as it is effectively changing the resolution and accuracy of the statistical modeling [19].…”