NuCLS: A scalable crowdsourcing approach and dataset for nucleus classification and segmentation in breast cancer

Abstract

Background Deep learning enables accurate high-resolution mapping of cells and tissue structures that can serve as the foundation of interpretable machine-learning models for computational pathology. However, generating adequate labels for these structures is a critical barrier, given the time and effort required from pathologists. Results This article describes a novel collaborative framework for engaging crowds of medical students and pathologists to produce quality labels for cell nuclei. We used this approach to produce the NuCLS dataset, containing >220,000 annotations of cell nuclei in breast cancers. This builds on prior work labeling tissue regions to produce an integrated tissue region- and cell-level annotation dataset for training that is the largest such resource for multi-scale analysis of breast cancer histology. This article presents data and analysis results for single and multi-rater annotations from both non-experts and pathologists. We present a novel workflow that uses algorithmic suggestions to collect accurate segmentation data without the need for laborious manual tracing of nuclei. Our results indicate that even noisy algorithmic suggestions do not adversely affect pathologist accuracy and can help non-experts improve annotation quality. We also present a new approach for inferring truth from multiple raters and show that non-experts can produce accurate annotations for visually distinctive classes. Conclusions This study is the most extensive systematic exploration of the large-scale use of wisdom-of-the-crowd approaches to generate data for computational pathology applications.

Beyond pathologist-level annotation of large-scale cancer histology for semantic segmentation using immunofluorescence restaining

“…b, Annotated areas for each tissue and c, the number of annotated cells for each cell type in SegPath. Those in publicly available datasets (BCSS 13 , GlaS 15 , CoNIC 5 , MoNuSAC2020 14 , and NuCLS 7 ) are also shown. Organs in brackets represent the target organs of the dataset.…”

“…A mixture of anti-CD3 and anti-CD20 antibodies was used for lymphocytes. Because our workflow can generate segmentation masks in a high-throughput manner without the need for manual annotation, the size of the dataset is over one order of magnitude larger than the currently available segmentation mask datasets for tumour tissues 5,7,[13][14][15] (Fig. 1b, c).…”

“…However, the high performance of deep neural networks is because of their abundant annotations, which are often difficult to obtain in medical imaging. For the semantic segmentation of H&E images, there have been many efforts to annotate tissues or cells, most of which rely on human annotators 6, 7 . However, manually annotating tumour tissues is not feasible by non-pathologists and requires considerable time and labour, limiting the generation of large- scale annotated datasets.…”

A Deep Learning Approach for Histology-Based Nuclei Segmentation and Tumor Microenvironment Characterization

“…The publicly available NuCLS dataset 28 consists of many image patches extracted from breast cancer images from The Cancer Genome Atlas (TCGA) dataset 25 . Image patches were split into single-rater and multi-rater datasets.…”

Why Panoptic Quality should be avoided as a metric for assessing cell nuclei segmentation and classification in digital pathology

“…The NuCLS dataset 14 proposes a "crowdsourced" dataset where the annotations are made by non-pathologists from algorithmic suggestions, and with corrections by junior and senior pathologists. It also provides a "multi-rater" dataset, where detailed individual annotations from experts and non-experts are provided on selected FOVs.…”