Automated mitotic spindle hotspot counts are highly associated with clinical outcomes in systemically untreated early-stage triple-negative breast cancer.

Operable triple-negative breast cancer (TNBC) has a higher risk of recurrence and death compared to other subtypes. Tumor size and nodal status are the primary clinical factors used to guide systemic treatment, while biomarkers of proliferation have not demonstrated value. Recent studies suggest that subsets of TNBC have a favorable prognosis, even without systemic therapy. We evaluated the association of fully automated mitotic spindle hotspot (AMSH) counts with recurrence-free (RFS) and overall survival (OS) in two separate cohorts of patients with early-stage TNBC who did not receive systemic therapy. AMSH counts were obtained from areas with the highest mitotic density in digitized whole slide images processed with a convolutional neural network trained to detect mitoses. In 140 patients from the Mayo Clinic TNBC cohort, AMSH counts were significantly associated with RFS and OS in a multivariable model controlling for nodal status, tumor size, and tumor-infiltrating lymphocytes (TILs) (p < 0.0001). For every 10-point increase in AMSH counts, there was a 16% increase in the risk of an RFS event (HR 1.16, 95% CI 1.08-1.25), and a 7% increase in the risk of death (HR 1.07, 95% CI 1.00-1.14). We corroborated these findings in a separate cohort of systemically untreated TNBC patients from Radboud UMC in the Netherlands. Our findings suggest that AMSH counts offer valuable prognostic information in patients with early-stage TNBC who did not receive systemic therapy, independent of tumor size, nodal status, and TILs. If further validated, AMSH counts could help inform future systemic therapy de-escalation strategies.

Comparing deep learning and pathologist quantification of cell-level PD-L1 expression in non-small cell lung cancer whole-slide images.

Programmed death-ligand 1 (PD-L1) expression is currently used in the clinic to assess eligibility for immune-checkpoint inhibitors via the tumor proportion score (TPS), but its efficacy is limited by high interobserver variability. Multiple papers have presented systems for the automatic quantification of TPS, but none report on the task of determining cell-level PD-L1 expression and often reserve their evaluation to a single PD-L1 monoclonal antibody or clinical center. In this paper, we report on a deep learning algorithm for detecting PD-L1 negative and positive tumor cells at a cellular level and evaluate it on a cell-level reference standard established by six readers on a multi-centric, multi PD-L1 assay dataset. This reference standard also provides for the first time a benchmark for computer vision algorithms. In addition, in line with other papers, we also evaluate our algorithm at slide-level by measuring the agreement between the algorithm and six pathologists on TPS quantification. We find a moderately low interobserver agreement at cell-level level (mean reader-reader F1 score = 0.68) which our algorithm sits slightly under (mean reader-AI F1 score = 0.55), especially for cases from the clinical center not included in the training set. Despite this, we find good AI-pathologist agreement on quantifying TPS compared to the interobserver agreement (mean reader-reader Cohen's kappa = 0.54, 95% CI 0.26-0.81, mean reader-AI kappa = 0.49, 95% CI 0.27-0.72). In conclusion, our deep learning algorithm demonstrates promise in detecting PD-L1 expression at a cellular level and exhibits favorable agreement with pathologists in quantifying the tumor proportion score (TPS). We publicly release our models for use via the Grand-Challenge platform.

Multi-resolution deep learning characterizes tertiary lymphoid structures and their prognostic relevance in solid tumors.

BACKGROUND: Tertiary lymphoid structures (TLSs) are dense accumulations of lymphocytes in inflamed peripheral tissues, including cancer, and are associated with improved survival and response to immunotherapy in various solid tumors. Histological TLS quantification has been proposed as a novel predictive and prognostic biomarker, but lack of standardized methods of TLS characterization hampers assessment of TLS densities across different patients, diseases, and clinical centers. METHODS: We introduce an approach based on HookNet-TLS, a multi-resolution deep learning model, for automated and unbiased TLS quantification and identification of germinal centers in routine hematoxylin and eosin stained digital pathology slides. We developed HookNet-TLS using n = 1019 manually annotated TCGA slides from clear cell renal cell carcinoma, muscle-invasive bladder cancer, and lung squamous cell carcinoma. RESULTS: Here we show that HookNet-TLS automates TLS quantification across multiple cancer types achieving human-level performance and demonstrates prognostic associations similar to visual assessment. CONCLUSIONS: HookNet-TLS has the potential to be used as a tool for objective quantification of TLS in routine H&E digital pathology slides. We make HookNet-TLS publicly available to promote its use in research.

Tertiary lymphoid structures (TLS) are dense accumulations of immune cells within a cancer. They have been associated with patient survival and treatment effectiveness. Quantification of TLS in cancer microscopy images may therefore aid clinical decision-making. However, no consensus for defining TLS in such images exists leading to inconsistent and variable findings across different labs and studies. We developed a computational tool for automated and objective TLS quantification in cancer images. The tool, called HookNet-TLS, integrates information from multiple image resolutions, which resembles the process of how a pathologist would identify these structures using a microscope. HookNet-TLS detected TLS similarly to trained researchers in three different tumor types. We provided access to HookNet-TLS to facilitate its development and use for TLS assessment in clinical decision-making and research into the role of TLS in cancer.

Lymph node metastases and recurrence in pT1 colorectal cancer: Prediction with the International Budding Consortium Score-A retrospective, multi-centric study.

BACKGROUND: The International Collaboration on Cancer Reporting proposes histological tumour type, lymphovascular invasion, tumour grade, perineural invasion, extent, and dimensions of invasion as risk factors for lymph node metastases and tumour progression in completely endoscopically resected pT1 colorectal cancer (CRC). OBJECTIVE: The aim of the study was to propose a predictive and reliable score to optimise the clinical management of endoscopically resected pT1 CRC patients. METHODS: This multi-centric, retrospective International Budding Consortium (IBC) study included an international pT1 CRC cohort of 565 patients. All cases were reviewed by eight expert gastrointestinal pathologists. All risk factors were reported according to international guidelines. Tumour budding and immune response (CD8+ T-cells) were assessed with automated models using artificial intelligence. We used the information on risk factors and least absolute shrinkage and selection operator logistic regression to develop a prediction model and generate a score to predict the occurrence of lymph node metastasis or cancer recurrence. RESULTS: The IBC prediction score included the following parameters: lymphovascular invasion, tumour buds, infiltration depth and tumour grade. The score has an acceptable discrimination power (area under the curve of 0.68 [95% confidence intervals (CI) 0.61-0.75]; 0.64 [95% CI 0.57-0.71] after internal validation). At a cut-off of 6.8 points to discriminate high-and low-risk patients, the score had a sensitivity and specificity of 0.9 [95% CI 0.8-0.95] and 0.26 [95% 0.22, 0.3], respectively. CONCLUSION: The IBC score is based on well-established risk factors and is a promising tool with clinical utility to support the management of pT1 CRC patients.

Automatic data augmentation to improve generalization of deep learning in H&E stained histopathology.

In histopathology practice, scanners, tissue processing, staining, and image acquisition protocols vary from center to center, resulting in subtle variations in images. Vanilla convolutional neural networks are sensitive to such domain shifts. Data augmentation is a popular way to improve domain generalization. Currently, state-of-the-art domain generalization in computational pathology is achieved using a manually curated set of augmentation transforms. However, manual tuning of augmentation parameters is time-consuming and can lead to sub-optimal generalization performance. Meta-learning frameworks can provide efficient ways to find optimal training hyper-parameters, including data augmentation. In this study, we hypothesize that an automated search of augmentation hyper-parameters can provide superior generalization performance and reduce experimental optimization time. We select four state-of-the-art automatic augmentation methods from general computer vision and investigate their capacity to improve domain generalization in histopathology. We analyze their performance on data from 25 centers across two different tasks: tumor metastasis detection in lymph nodes and breast cancer tissue type classification. On tumor metastasis detection, most automatic augmentation methods achieve comparable performance to state-of-the-art manual augmentation. On breast cancer tissue type classification, the leading automatic augmentation method significantly outperforms state-of-the-art manual data augmentation.

LYSTO: The Lymphocyte Assessment Hackathon and Benchmark Dataset.

We introduce LYSTO, the Lymphocyte Assessment Hackathon, which was held in conjunction with the MICCAI 2019 Conference in Shenzhen (China). The competition required participants to automatically assess the number of lymphocytes, in particular T-cells, in images of colon, breast, and prostate cancer stained with CD3 and CD8 immunohistochemistry. Differently from other challenges setup in medical image analysis, LYSTO participants were solely given a few hours to address this problem. In this paper, we describe the goal and the multi-phase organization of the hackathon; we describe the proposed methods and the on-site results. Additionally, we present post-competition results where we show how the presented methods perform on an independent set of lung cancer slides, which was not part of the initial competition, as well as a comparison on lymphocyte assessment between presented methods and a panel of pathologists. We show that some of the participants were capable to achieve pathologist-level performance at lymphocyte assessment. After the hackathon, LYSTO was left as a lightweight plug-and-play benchmark dataset on grand-challenge website, together with an automatic evaluation platform.

The Banff 2022 Kidney Meeting Work Plan: Data-driven refinement of the Banff Classification for renal allografts.

The XVIth Banff Meeting for Allograft Pathology was held in Banff, Alberta, Canada, from September 19 to 23, 2022, as a joint meeting with the Canadian Society of Transplantation. In addition to a key focus on the impact of microvascular inflammation and biopsy-based transcript analysis on the Banff Classification, further sessions were devoted to other aspects of kidney transplant pathology, in particular T cell-mediated rejection, activity and chronicity indices, digital pathology, xenotransplantation, clinical trials, and surrogate endpoints. Although the output of these sessions has not led to any changes in the classification, the key role of Banff Working Groups in phrasing unanswered questions, and coordinating and disseminating results of investigations addressing these unanswered questions was emphasized. This paper summarizes the key Banff Meeting 2022 sessions not covered in the Banff Kidney Meeting 2022 Report paper and also provides an update on other Banff Working Group activities relevant to kidney allografts.

PROACTING: predicting pathological complete response to neoadjuvant chemotherapy in breast cancer from routine diagnostic histopathology biopsies with deep learning.

BACKGROUND: Invasive breast cancer patients are increasingly being treated with neoadjuvant chemotherapy; however, only a fraction of the patients respond to it completely. To prevent overtreatment, there is an urgent need for biomarkers to predict treatment response before administering the therapy. METHODS: In this retrospective study, we developed hypothesis-driven interpretable biomarkers based on deep learning, to predict the pathological complete response (pCR, i.e., the absence of tumor cells in the surgical resection specimens) to neoadjuvant chemotherapy solely using digital pathology H&E images of pre-treatment breast biopsies. Our approach consists of two steps: First, we use deep learning to characterize aspects of the tumor micro-environment by detecting mitoses and segmenting tissue into several morphology compartments including tumor, lymphocytes and stroma. Second, we derive computational biomarkers from the segmentation and detection output to encode slide-level relationships of components of the tumor microenvironment, such as tumor and mitoses, stroma, and tumor infiltrating lymphocytes (TILs). RESULTS: We developed and evaluated our method on slides from n = 721 patients from three European medical centers with triple-negative and Luminal B breast cancers and performed external independent validation on n = 126 patients from a public dataset. We report the predictive value of the investigated biomarkers for predicting pCR with areas under the receiver operating characteristic curve between 0.66 and 0.88 across the tested cohorts. CONCLUSION: The proposed computational biomarkers predict pCR, but will require more evaluation and finetuning for clinical application. Our results further corroborate the potential role of deep learning to automate TILs quantification, and their predictive value in breast cancer neoadjuvant treatment planning, along with automated mitoses quantification. We made our method publicly available to extract segmentation-based biomarkers for research purposes.

The Latent Doctor Model for Modeling Inter-Observer Variability.

Many inherently ambiguous tasks in medical imaging suffer from inter-observer variability, resulting in a reference standard defined by a distribution of labels with high variance. Training only on a consensus or majority vote label, as is common in medical imaging, discards valuable information on uncertainty amongst a panel of experts. In this work, we propose to train on the full label distribution to predict the uncertainty within a panel of experts and the most likely ground-truth label. To do so, we propose a new stochastic classification framework based on the conditional variational auto-encoder, which we refer to as the Latent Doctor Model (LDM). In an extensive comparative analysis, we compare the LDM with a model trained on the majority vote label and other methods capable of learning a distribution of labels. We show that the LDM is able to reproduce the reference-standard distribution significantly better than the majority vote baseline. Compared to the other baseline methods, we demonstrate that the LDM performs best at modeling the label distribution and its corresponding uncertainty in two prostate tumor grading tasks. Furthermore, we show competitive performance of the LDM with the more computationally demanding deep ensembles on a tumor budding classification task.

Pathologists' first opinions on barriers and facilitators of computational pathology adoption in oncological pathology: an international study.

Computational pathology (CPath) algorithms detect, segment or classify cancer in whole slide images, approaching or even exceeding the accuracy of pathologists. Challenges have to be overcome before these algorithms can be used in practice. We therefore aim to explore international perspectives on the future role of CPath in oncological pathology by focusing on opinions and first experiences regarding barriers and facilitators. We conducted an international explorative eSurvey and semi-structured interviews with pathologists utilizing an implementation framework to classify potential influencing factors. The eSurvey results showed remarkable variation in opinions regarding attitude, understandability and validation of CPath. Interview results showed that barriers focused on the quality of available evidence, while most facilitators concerned strengths of CPath. A lack of consensus was present for multiple factors, such as the determination of sufficient validation using CPath, the preferred function of CPath within the digital workflow and the timing of CPath introduction in pathology education. The diversity in opinions illustrates variety in influencing factors in CPath adoption. A next step would be to quantitatively determine important factors for adoption and initiate validation studies. Both should include clear case descriptions and be conducted among a more homogenous panel of pathologists based on sub specialization.

Gigapixel end-to-end training using streaming and attention.

Current hardware limitations make it impossible to train convolutional neural networks on gigapixel image inputs directly. Recent developments in weakly supervised learning, such as attention-gated multiple instance learning, have shown promising results, but often use multi-stage or patch-wise training strategies risking suboptimal feature extraction, which can negatively impact performance. In this paper, we propose to train a ResNet-34 encoder with an attention-gated classification head in an end-to-end fashion, which we call StreamingCLAM, using a streaming implementation of convolutional layers. This allows us to train end-to-end on 4-gigapixel microscopic images using only slide-level labels. We achieve a mean area under the receiver operating characteristic curve of 0.9757 for metastatic breast cancer detection (CAMELYON16), close to fully supervised approaches using pixel-level annotations. Our model can also detect MYC-gene translocation in histologic slides of diffuse large B-cell lymphoma, achieving a mean area under the ROC curve of 0.8259. Furthermore, we show that our model offers a degree of interpretability through the attention mechanism.

Deep learning for multi-class semantic segmentation enables colorectal cancer detection and classification in digital pathology images.

In colorectal cancer (CRC), artificial intelligence (AI) can alleviate the laborious task of characterization and reporting on resected biopsies, including polyps, the numbers of which are increasing as a result of CRC population screening programs ongoing in many countries all around the globe. Here, we present an approach to address two major challenges in the automated assessment of CRC histopathology whole-slide images. We present an AI-based method to segment multiple ([Formula: see text]) tissue compartments in the H &E-stained whole-slide image, which provides a different, more perceptible picture of tissue morphology and composition. We test and compare a panel of state-of-the-art loss functions available for segmentation models, and provide indications about their use in histopathology image segmentation, based on the analysis of (a) a multi-centric cohort of CRC cases from five medical centers in the Netherlands and Germany, and (b) two publicly available datasets on segmentation in CRC. We used the best performing AI model as the basis for a computer-aided diagnosis system that classifies colon biopsies into four main categories that are relevant pathologically. We report the performance of this system on an independent cohort of more than 1000 patients. The results show that with a good segmentation network as a base, a tool can be developed which can support pathologists in the risk stratification of colorectal cancer patients, among other possible uses. We have made the segmentation model available for research use on https://grand-challenge.org/algorithms/colon-tissue-segmentation/ .

Fully Automated Tumor Bud Assessment in Hematoxylin and Eosin-Stained Whole Slide Images of Colorectal Cancer.

Tumor budding (TB), the presence of single cells or small clusters of up to 4 tumor cells at the invasive front of colorectal cancer (CRC), is a proven risk factor for adverse outcomes. International definitions are necessary to reduce interobserver variability. According to the current international guidelines, hotspots at the invasive front should be counted in hematoxylin and eosin (H&E)-stained slides. This is time-consuming and prone to interobserver variability; therefore, there is a need for computer-aided diagnosis solutions. In this study, we report an artificial intelligence-based method for detecting TB in H&E-stained whole slide images. We propose a fully automated pipeline to identify the tumor border, detect tumor buds, characterize them based on the number of tumor cells, and produce a TB density map to identify the TB hotspot. The method outputs the TB count in the hotspot as a computational biomarker. We show that the proposed automated TB detection workflow performs on par with a panel of 5 pathologists at detecting tumor buds and that the hotspot-based TB count is an independent prognosticator in both the univariate and the multivariate analysis, validated on a cohort of n = 981 patients with CRC. Computer-aided detection of tumor buds based on deep learning can perform on par with expert pathologists for the detection and quantification of tumor buds in H&E-stained CRC histopathology slides, strongly facilitating the introduction of budding as an independent prognosticator in clinical routine and clinical trials.

Automated Deep Learning-Based Classification of Wilms Tumor Histopathology.

(1) Background: Histopathological assessment of Wilms tumors (WT) is crucial for risk group classification to guide postoperative stratification in chemotherapy pre-treated WT cases. However, due to the heterogeneous nature of the tumor, significant interobserver variation between pathologists in WT diagnosis has been observed, potentially leading to misclassification and suboptimal treatment. We investigated whether artificial intelligence (AI) can contribute to accurate and reproducible histopathological assessment of WT through recognition of individual histopathological tumor components. (2) Methods: We assessed the performance of a deep learning-based AI system in quantifying WT components in hematoxylin and eosin-stained slides by calculating the Sørensen-Dice coefficient for fifteen predefined renal tissue components, including six tumor-related components. We trained the AI system using multiclass annotations from 72 whole-slide images of patients diagnosed with WT. (3) Results: The overall Dice coefficient for all fifteen tissue components was 0.85 and for the six tumor-related components was 0.79. Tumor segmentation worked best to reliably identify necrosis (Dice coefficient 0.98) and blastema (Dice coefficient 0.82). (4) Conclusions: Accurate histopathological classification of WT may be feasible using a digital pathology-based AI system in a national cohort of WT patients.

Semi-Supervised Learning to Automate Tumor Bud Detection in Cytokeratin-Stained Whole-Slide Images of Colorectal Cancer.

Tumor budding is a histopathological biomarker associated with metastases and adverse survival outcomes in colorectal carcinoma (CRC) patients. It is characterized by the presence of single tumor cells or small clusters of cells within the tumor or at the tumor-invasion front. In order to obtain a tumor budding score for a patient, the region with the highest tumor bud density must first be visually identified by a pathologist, after which buds will be counted in the chosen hotspot field. The automation of this process will expectedly increase efficiency and reproducibility. Here, we present a deep learning convolutional neural network model that automates the above procedure. For model training, we used a semi-supervised learning method, to maximize the detection performance despite the limited amount of labeled training data. The model was tested on an independent dataset in which human- and machine-selected hotspots were mapped in relation to each other and manual and machine detected tumor bud numbers in the manually selected fields were compared. We report the results of the proposed method in comparison with visual assessment by pathologists. We show that the automated tumor bud count achieves a prognostic value comparable with visual estimation, while based on an objective and reproducible quantification. We also explore novel metrics to quantify buds such as density and dispersion and report their prognostic value. We have made the model available for research use on the grand-challenge platform.

Consensus based recommendations for the diagnosis of serous tubal intraepithelial carcinoma: an international Delphi study.

AIM: Reliably diagnosing or safely excluding serous tubal intraepithelial carcinoma (STIC), a precursor lesion of tubo-ovarian high-grade serous carcinoma (HGSC), is crucial for individual patient care, for better understanding the oncogenesis of HGSC, and for safely investigating novel strategies to prevent tubo-ovarian carcinoma. To optimize STIC diagnosis and increase its reproducibility, we set up a three-round Delphi study. METHODS AND RESULTS: In round 1, an international expert panel of 34 gynecologic pathologists, from 11 countries, was assembled to provide input regarding STIC diagnosis, which was used to develop a set of statements. In round 2, the panel rated their level of agreement with those statements on a 9-point Likert scale. In round 3, statements without previous consensus were rated again by the panel while anonymously disclosing the responses of the other panel members. Finally, each expert was asked to approve or disapprove the complete set of consensus statements. The panel indicated their level of agreement with 64 statements. A total of 27 statements (42%) reached consensus after three rounds. These statements reflect the entire diagnostic work-up for pathologists, regarding processing and macroscopy (three statements); microscopy (eight statements); immunohistochemistry (nine statements); interpretation and reporting (four statements); and miscellaneous (three statements). The final set of consensus statements was approved by 85%. CONCLUSION: This study provides an overview of current clinical practice regarding STIC diagnosis amongst expert gynecopathologists. The experts' consensus statements form the basis for a set of recommendations, which may help towards more consistent STIC diagnosis.

Continual learning strategies for cancer-independent detection of lymph node metastases.

Recently, large, high-quality public datasets have led to the development of convolutional neural networks that can detect lymph node metastases of breast cancer at the level of expert pathologists. Many cancers, regardless of the site of origin, can metastasize to lymph nodes. However, collecting and annotating high-volume, high-quality datasets for every cancer type is challenging. In this paper we investigate how to leverage existing high-quality datasets most efficiently in multi-task settings for closely related tasks. Specifically, we will explore different training and domain adaptation strategies, including prevention of catastrophic forgetting, for breast, colon and head-and-neck cancer metastasis detection in lymph nodes. Our results show state-of-the-art performance on colon and head-and-neck cancer metastasis detection tasks. We show the effectiveness of adaptation of networks from one cancer type to another to obtain multi-task metastasis detection networks. Furthermore, we show that leveraging existing high-quality datasets can significantly boost performance on new target tasks and that catastrophic forgetting can be effectively mitigated.Last, we compare different mitigation strategies.

Deep learning based tumor-stroma ratio scoring in colon cancer correlates with microscopic assessment.

Background: The amount of stroma within the primary tumor is a prognostic parameter for colon cancer patients. This phenomenon can be assessed using the tumor-stroma ratio (TSR), which classifies tumors in stroma-low (≤50% stroma) and stroma-high (>50% stroma). Although the reproducibility for TSR determination is good, improvement might be expected from automation. The aim of this study was to investigate whether the scoring of the TSR in a semi- and fully automated method using deep learning algorithms is feasible. Methods: A series of 75 colon cancer slides were selected from a trial series of the UNITED study. For the standard determination of the TSR, 3 observers scored the histological slides. Next, the slides were digitized, color normalized, and the stroma percentages were scored using semi- and fully automated deep learning algorithms. Correlations were determined using intraclass correlation coefficients (ICCs) and Spearman rank correlations. Results: 37 (49%) cases were classified as stroma-low and 38 (51%) as stroma-high by visual estimation. A high level of concordance between the 3 observers was reached, with ICCs of 0.91, 0.89, and 0.94 (all P < .001). Between visual and semi-automated assessment the ICC was 0.78 (95% CI 0.23-0.91, P-value 0.005), with a Spearman correlation of 0.88 (P < .001). Spearman correlation coefficients above 0.70 (N=3) were observed for visual estimation versus the fully automated scoring procedures. Conclusion: Good correlations were observed between standard visual TSR determination and semi- and fully automated TSR scores. At this point, visual examination has the highest observer agreement, but semi-automated scoring could be helpful to support pathologists.

A whole-slide imaging based workflow reduces the reading time of pathologists.

Even though entirely digitized microscopic tissue sections (whole slide images, WSIs) are increasingly being used in histopathology diagnostics, little data is still available on the effect of this technique on pathologists' reading time. This study aimed to compare the time required to perform the microscopic assessment by pathologists between a conventional workflow (an optical microscope) and digitized WSIs. WSI was used in primary diagnostics at the Laboratory for Pathology Eastern Netherlands for several years (LabPON, Hengelo, The Netherlands). Cases were read either in a traditional workflow, with the pathologist recording the time required for diagnostics and reporting, or entirely digitally. Reading times were extracted from image management system log files, and the digitized workflow was fully integrated into the laboratory information system. The digital workflow saved time in the majority of case categories, with prostate biopsies saving the most (68% time gain). Taking into account case distribution, the digital workflow produced an average gain of 12.3%. Using WSI instead of conventional microscopy significantly reduces pathologists' reading times. Pathologists must work in a fully integrated environment to fully reap the benefits of a digital workflow.

Deep learning for fully-automated nuclear pleomorphism scoring in breast cancer.

To guide the choice of treatment, every new breast cancer is assessed for aggressiveness (i.e., graded) by an experienced histopathologist. Typically, this tumor grade consists of three components, one of which is the nuclear pleomorphism score (the extent of abnormalities in the overall appearance of tumor nuclei). The degree of nuclear pleomorphism is subjectively classified from 1 to 3, where a score of 1 most closely resembles epithelial cells of normal breast epithelium and 3 shows the greatest abnormalities. Establishing numerical criteria for grading nuclear pleomorphism is challenging, and inter-observer agreement is poor. Therefore, we studied the use of deep learning to develop fully automated nuclear pleomorphism scoring in breast cancer. The reference standard used for training the algorithm consisted of the collective knowledge of an international panel of 10 pathologists on a curated set of regions of interest covering the entire spectrum of tumor morphology in breast cancer. To fully exploit the information provided by the pathologists, a first-of-its-kind deep regression model was trained to yield a continuous scoring rather than limiting the pleomorphism scoring to the standard three-tiered system. Our approach preserves the continuum of nuclear pleomorphism without necessitating a large data set with explicit annotations of tumor nuclei. Once translated to the traditional system, our approach achieves top pathologist-level performance in multiple experiments on regions of interest and whole-slide images, compared to a panel of 10 and 4 pathologists, respectively.