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

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.

Explainable nucleus classification using Decision Tree Approximation of Learned Embeddings

MOTIVATION

Nucleus detection, segmentation and classification are fundamental to high-resolution mapping of the tumor microenvironment using whole-slide histopathology images. The growing interest in leveraging the power of deep learning to achieve state-of-the-art performance often comes at the cost of explainability, yet there is general consensus that explainability is critical for trustworthiness and widespread clinical adoption. Unfortunately, current explainability paradigms that rely on pixel saliency heatmaps or superpixel importance scores are not well-suited for nucleus classification. Techniques like Grad-CAM or LIME provide explanations that are indirect, qualitative and/or nonintuitive to pathologists.

RESULTS

In this article, we present techniques to enable scalable nuclear detection, segmentation and explainable classification. First, we show how modifications to the widely used Mask R-CNN architecture, including decoupling the detection and classification tasks, improves accuracy and enables learning from hybrid annotation datasets like NuCLS, which contain mixtures of bounding boxes and segmentation boundaries. Second, we introduce an explainability method called Decision Tree Approximation of Learned Embeddings (DTALE), which provides explanations for classification model behavior globally, as well as for individual nuclear predictions. DTALE explanations are simple, quantitative, and can flexibly use any measurable morphological features that make sense to practicing pathologists, without sacrificing model accuracy. Together, these techniques present a step toward realizing the promise of computational pathology in computer-aided diagnosis and discovery of morphologic biomarkers.

AVAILABILITY AND IMPLEMENTATION

Relevant code can be found at github.com/CancerDataScience/NuCLS.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Circulating soluble urokinase plasminogen activator receptor levels and peripheral arterial disease outcomes

BACKGROUND AND AIMS

Circulating soluble urokinase plasminogen activator receptor (suPAR) is a marker of immune activation associated with atherosclerosis. Whether suPAR levels are associated with prevalent peripheral arterial disease (PAD) and its adverse outcomes remains unknown and is the aim of the study.

METHODS

SuPAR levels were measured in 5810 patients (mean age 63 years, 63% male, 77% with obstructive coronary artery disease [CAD]) undergoing cardiac catheterization. The presence of PAD (n = 967, 17%) was classified as carotid (36%), lower/upper extremities (30%), aortic (15%) and multisite disease (19%). Multivariable logistic and Cox regression models were used to determine independent predictors of prevalent PAD and outcomes including all-cause death, cardiovascular death and PAD-related events after adjustment for age, gender, race, body mass index, smoking, diabetes, hypertension, hyperlipidemia, renal function, heart failure history, and obstructive CAD.

RESULTS

Plasma suPAR levels were 22.5% (p < 0.001) higher in patients with PAD compared to those without PAD. Plasma suPAR was higher in patients with more extensive PAD (≥2 compared to single site) p < 0.001. After multivariable adjustment, suPAR was associated with prevalent PAD; odds ratio (OR) for highest compared to lowest tertile of 2.0, 95% CI (1.6-2.5) p < 0.001. In Cox survival analyses adjusted for clinical characteristics and medication regimen, suPAR (in the highest vs. lowest tertile) remained an independent predictor of all-cause death [HR 3.1, 95% CI (1.9-5.3)], cardiovascular death [HR 3.5, 95% CI (1.8-7.0)] and PAD-related events [HR = 1.8, 95% CI (1.3-2.6) p < 0.001 for all].

CONCLUSIONS

Plasma suPAR level is predictive of prevalent PAD and of incident cardiovascular and PAD-related events. Whether SuPAR measurement can help screen, risk stratify, or monitor therapeutic responses in PAD requires further investigation.

Double staining: diagnostic utility in non-small cell lung carcinoma in the era of tissue conservation

INTRODUCTION

In an era of precision medicine distinguishing pulmonary squamous cell carcinoma (SQCC) from adenocarcinoma (ADC) is vital for treatment. Immunohistochemical (IHC) staining for p40, p63 and Cytokeratin 5 (CK5) are useful for SQCC, while TTF-1 and Napsin-A can be used for confirming ADC. Fine needle aspiration (FNA) cell blocks (CB) have limited tissue, hence, double IHC staining is helpful for tissue conservation for molecular analysis.

MATERIALS AND METHODS

Thirty six confirmed lung SQCC and 45 ADC CB were selected for IHC. Double staining was performed with p40/CK5 and p63/CK5 on all SQCC, and with TTF-1/Napsin-A on all ADC. Results were positive if at least 5% of malignant cells were immunoreactive for the antigen.

RESULTS

P40/CK5 had (92%) sensitivity, (100%) specificity, (100%) positive predictive value (PPV), (91%) negative predictive value (NPV) and an overall diagnostic accuracy of (96%). By contrast, P63/CK5 double stains showed (92%) sensitivity, (80%) specificity, (85%) PPV, (89%) NPV and (86%) overall diagnostic accuracy, respectively. TTF-1/Napsin A staining for ADC showed a sensitivity of 80%, specificity of 96%, PPV of 97%, NPV of 71% and accuracy of 85%.

CONCLUSION

P40/CK5 double stain has higher specificity, PPV, NPV, and overall accuracy than P63/CK5 double stain in the diagnosis of lung SQCC. TTF-1/Napsin-A double staining is a valuable marker with high specificity, PPV, and diagnostic accuracy in diagnosing lung ADC. The usage of P40/CK5 and TTF-1/Napsin-A as a panel can be recommended for characterizing non-small cell carcinoma (NSCC) of the lung and for conserving tissue for molecular testing.

Telomere Shortening, Regenerative Capacity, and Cardiovascular Outcomes

RATIONALE

Leukocyte telomere length (LTL) is a biological marker of aging, and shorter LTL is associated with adverse cardiovascular outcomes. Reduced regenerative capacity has been proposed as a mechanism. Bone marrow-derived circulating progenitor cells are involved in tissue repair and regeneration.

OBJECTIVE

Main objective of this study was to examine the relationship between LTL and progenitor cells and their impact on adverse cardiovascular outcomes.

METHODS AND RESULTS

We measured LTL by quantitative polymerase chain reaction in 566 outpatients (age: 63±9 years; 76% men) with coronary artery disease. Circulating progenitor cells were enumerated by flow cytometry. After adjustment for age, sex, race, body mass index, smoking status, and previous myocardial infarction, a shorter LTL was associated with a lower CD34⁺ cell count: for each 10% shorter LTL, CD34⁺ levels were 5.2% lower (P<0.001). After adjustment for the aforementioned factors, both short LTL (<Q1) and low CD34+ levels (<Q1) predicted adverse cardiovascular outcomes (death, myocardial infarction, coronary revascularization, or cerebrovascular events) independently of each other, with a hazard ratio of 1.8 and 95% confidence interval of 1.1 to 2.0, and a hazard ratio of 2.1 and 95% confidence interval of 1.3 to 3.0, respectively, comparing Q1 to Q2-4. Patients who had both short LTL (<Q1) and low CD34+ cell count (<Q1) had the greatest risk of adverse outcomes (hazard ratio =3.5; 95% confidence interval, 1.7-7.1).

CONCLUSIONS

Although shorter LTL is associated with decreased regenerative capacity, both LTL and circulating progenitor cell levels are independent and additive predictors of adverse cardiovascular outcomes in coronary artery disease patients. Our results suggest that both biological aging and reduced regenerative capacity contribute to cardiovascular events, independent of conventional risk factors.