EUSOMA quality indicators for non-metastatic breast cancer: An update.

INTRODUCTION: Quality care in breast cancer is higher if patients are treated in a Breast Center with a dedicated and specialized multidisciplinary team. Quality control is an essential activity to ensure quality care, which has to be based on the monitoring of specific quality indicators. Eusoma has proceeded with the up-dating of the 2017 Quality indicators for non-metastatic breast cancer based on the new diagnostic, locoregional and systemic treatment modalities. METHODS: To proceed with the updating, EUSOMA setup a multidisciplinary working group of BC experts and patients' representatives. It is a comprehensive set of QIs for early breast cancer care, which are classified as mandatory, recommended, or observational. For the first time patient reported outcomes (PROMs) have been included. As used in the 2017 EUSOMA QIs, evidence levels were based on the short version of the US Agency for Healthcare Research and Quality. RESULTS: This is a set of quality indicators representative for the different steps of the patient pathway in non-metastatic setting, which allow Breast Centres to monitor their performance with referring standards, i.e minimum standard and target. CONCLUSIONS: Monitoring these Quality Indicators, within the Eusoma datacentre will allow to have a state of the art picture at European Breast Centres level and the development of challenging research projects.

European guidelines for the diagnosis, treatment and follow-up of breast lesions with uncertain malignant potential (B3 lesions) developed jointly by EUSOMA, EUSOBI, ESP (BWG) and ESSO.

INTRODUCTION: Breast lesions of uncertain malignant potential (B3) include atypical ductal and lobular hyperplasias, lobular carcinoma in situ, flat epithelial atypia, papillary lesions, radial scars and fibroepithelial lesions as well as other rare miscellaneous lesions. They are challenging to categorise histologically, requiring specialist training and multidisciplinary input. They may coexist with in situ or invasive breast cancer (BC) and increase the risk of subsequent BC development. Management should focus on adequate classification and management whilst avoiding overtreatment. The aim of these guidelines is to provide updated information regarding the diagnosis and management of B3 lesions, according to updated literature review evidence. METHODS: These guidelines provide practical recommendations which can be applied in clinical practice which include recommendation grade and level of evidence. All sections were written according to an updated literature review and discussed at a consensus meeting. Critical appraisal by the expert writing committee adhered to the 23 items in the international Appraisal of Guidelines, Research and Evaluation (AGREE) tool. RESULTS: Recommendations for further management after core-needle biopsy (CNB) or vacuum-assisted biopsy (VAB) diagnosis of a B3 lesion reported in this guideline, vary depending on the presence of atypia, size of lesion, sampling size, and patient preferences. After CNB or VAB, the option of vacuum-assisted excision or surgical excision should be evaluated by a multidisciplinary team and shared decision-making with the patient is crucial for personalizing further treatment. De-escalation of surgical intervention for B3 breast lesions is ongoing, and the inclusion of vacuum-assisted excision (VAE) will decrease the need for surgical intervention in further approaches. Communication with patients may be different according to histological diagnosis, presence or absence of atypia, or risk of upgrade due to discordant imaging. Written information resources to help patients understand these issues alongside with verbal communication is recommended. Lifestyle interventions have a significant impact on BC incidence so lifestyle interventions need to be suggested to women at increased BC risk as a result of a diagnosis of a B3 lesion. CONCLUSIONS: These guidelines provide a state-of-the-art overview of the diagnosis, management and prognosis of B3 lesions in modern multidisciplinary breast practice.

Biomechanical characterization of the passive porcine stomach.

The complex mechanics of the gastric wall facilitates the main digestive tasks of the stomach. However, the interplay between the mechanical properties of the stomach, its microstructure, and its vital functions is not yet fully understood. Importantly, the pig animal model is widely used in biomedical research for preliminary or ethically prohibited studies of the human digestion system. Therefore, this study aims to thoroughly characterize the mechanical behavior and microstructure of the porcine stomach. For this purpose, multiple quasi-static mechanical tests were carried out with three different loading modes, i.e., planar biaxial extension, radial compression, and simple shear. Stress-relaxation tests complemented the quasi-static experiments to evaluate the deformation and strain-dependent viscoelastic properties. Each experiment was conducted on specimens of the complete stomach wall and two separate layers, mucosa and muscularis, from each of the three gastric regions, i.e., fundus, body, and antrum. The significant preconditioning effects and the considerable regional and layer-specific differences in the tissue response were analyzed. Furthermore, the mechanical experiments were complemented with histology to examine the influence of the microstructural composition on the macrostructural mechanical response and vice versa. Importantly, the shear tests showed lower stresses in the complete wall compared to the single layers which the loose network of submucosal collagen might explain. Also, the stratum arrangement of the muscularis might explain mechanical anisotropy during tensile tests. This study shows that gastric tissue is characterized by a highly heterogeneous microstructure with regional variations in layer composition reflecting not only functional differences but also diverse mechanical behavior. STATEMENT OF SIGNIFICANCE: Unfortunately, only few experimental data on gastric tissue are available for an adequate material parameter and model estimation. The present study therefore combines layer- and region-specific stomach wall mechanics obtained under multiple loading conditions with histological insights into the heterogeneous microstructure. On the one hand, the extensive data sets of this study expand our understanding of the interplay between gastric mechanics, motility and functionality, which could help to identify and treat associated pathologies. On the other hand, such data sets are of high relevance for the constitutive modeling of stomach tissue, and its application in the field of medical engineering, e.g., in the development of surgical staplers and the improvement of bariatric surgical interventions.

Third International Consensus Conference on lesions of uncertain malignant potential in the breast (B3 lesions).

The heterogeneous group of B3 lesions in the breast harbors lesions with different malignant potential and progression risk. As several studies about B3 lesions have been published since the last Consensus in 2018, the 3rd International Consensus Conference discussed the six most relevant B3 lesions (atypical ductal hyperplasia (ADH), flat epithelial atypia (FEA), classical lobular neoplasia (LN), radial scar (RS), papillary lesions (PL) without atypia, and phyllodes tumors (PT)) and made recommendations for diagnostic and therapeutic approaches. Following a presentation of current data of each B3 lesion, the international and interdisciplinary panel of 33 specialists and key opinion leaders voted on the recommendations for further management after core-needle biopsy (CNB) and vacuum-assisted biopsy (VAB). In case of B3 lesion diagnosis on CNB, OE was recommended in ADH and PT, whereas in the other B3 lesions, vacuum-assisted excision was considered an equivalent alternative to OE. In ADH, most panelists (76%) recommended an open excision (OE) after diagnosis on VAB, whereas observation after a complete VAB-removal on imaging was accepted by 34%. In LN, the majority of the panel (90%) preferred observation following complete VAB-removal. Results were similar in RS (82%), PL (100%), and FEA (100%). In benign PT, a slim majority (55%) also recommended an observation after a complete VAB-removal. VAB with subsequent active surveillance can replace an open surgical intervention for most B3 lesions (RS, FEA, PL, PT, and LN). Compared to previous recommendations, there is an increasing trend to a de-escalating strategy in classical LN. Due to the higher risk of upgrade into malignancy, OE remains the preferred approach after the diagnosis of ADH.

Explainability and causability in digital pathology.

The current move towards digital pathology enables pathologists to use artificial intelligence (AI)-based computer programmes for the advanced analysis of whole slide images. However, currently, the best-performing AI algorithms for image analysis are deemed black boxes since it remains - even to their developers - often unclear why the algorithm delivered a particular result. Especially in medicine, a better understanding of algorithmic decisions is essential to avoid mistakes and adverse effects on patients. This review article aims to provide medical experts with insights on the issue of explainability in digital pathology. A short introduction to the relevant underlying core concepts of machine learning shall nurture the reader's understanding of why explainability is a specific issue in this field. Addressing this issue of explainability, the rapidly evolving research field of explainable AI (XAI) has developed many techniques and methods to make black-box machine-learning systems more transparent. These XAI methods are a first step towards making black-box AI systems understandable by humans. However, we argue that an explanation interface must complement these explainable models to make their results useful to human stakeholders and achieve a high level of causability, i.e. a high level of causal understanding by the user. This is especially relevant in the medical field since explainability and causability play a crucial role also for compliance with regulatory requirements. We conclude by promoting the need for novel user interfaces for AI applications in pathology, which enable contextual understanding and allow the medical expert to ask interactive 'what-if'-questions. In pathology, such user interfaces will not only be important to achieve a high level of causability. They will also be crucial for keeping the human-in-the-loop and bringing medical experts' experience and conceptual knowledge to AI processes.

ONEST (Observers Needed to Evaluate Subjective Tests) Analysis of Stromal Tumour-Infiltrating Lymphocytes (sTILs) in Breast Cancer and Its Limitations.

Tumour-infiltrating lymphocytes (TILs) reflect antitumour immunity. Their evaluation of histopathology specimens is influenced by several factors and is subject to issues of reproducibility. ONEST (Observers Needed to Evaluate Subjective Tests) helps in determining the number of observers that would be sufficient for the reliable estimation of inter-observer agreement of TIL categorisation. This has not been explored previously in relation to TILs. ONEST analyses, using an open-source software developed by the first author, were performed on TIL quantification in breast cancers taken from two previous studies. These were one reproducibility study involving 49 breast cancers, 23 in the first circulation and 14 pathologists in the second circulation, and one study involving 100 cases and 9 pathologists. In addition to the estimates of the number of observers required, other factors influencing the results of ONEST were examined. The analyses reveal that between six and nine observers (range 2-11) are most commonly needed to give a robust estimate of reproducibility. In addition, the number and experience of observers, the distribution of values around or away from the extremes, and outliers in the classification also influence the results. Due to the simplicity and the potentially relevant information it may give, we propose ONEST to be a part of new reproducibility analyses.

Global and local stiffening of ex vivo-perfused stented human thoracic aortas: A mock circulation study.

The effects of thoracic endovascular repair (TEVAR) on the biomechanical properties of aortic tissue have not been adequately studied. Understanding these features is important for the management of endograft-triggered complications of a biomechanical nature. This study aims to examine how stent-graft implantation affects the elastomechanical behavior of the aorta. Non-pathological human thoracic aortas (n=10) were subjected to long-standing perfusion (8h) within a mock circulation loop under physiological conditions. To quantify compliance and its mismatch in the test periods without and with a stent, the aortic pressure and the proximal cyclic circumferential displacement were measured. After perfusion, biaxial tension tests (stress-stretch) were carried out to examine the stiffness profiles between non-stented and stented tissue, followed by a histological assessment. Experimental evidence shows: (i) a significant reduction in aortic distensibility after TEVAR, indicating aortic stiffening and compliance mismatch, (ii) a stiffer behavior of the stented samples compared to the non-stented samples with an earlier entry into the nonlinear part of the stress-stretch curve and (iii) strut-induced histological remodeling of the aortic wall. The biomechanical and histological comparison of the non-stented and stented aortas provides new insights into the interaction between the stent-graft and the aortic wall. The knowledge gained could refine the stent-graft design to minimize the stent-induced impacts on the aortic wall and the resulting complications. STATEMENT OF SIGNIFICANCE: Stent-related cardiovascular complications occur the moment the stent-graft expands on the human aortic wall. Clinicians base their diagnosis on the anatomical morphology of CT scans while neglecting the endograft-triggered biomechanical events that compromise aortic compliance and wall mechanotransduction. Experimental replication of endovascular repair in cadaver aortas within a mock circulation loop may have a catalytic effect on biomechanical and histological findings without an ethical barrier. Demonstrating interactions between the stent and the wall can help clinicians make a broader diagnosis such as ECG-triggered oversizing and stent-graft characteristics based on patient-specific anatomical location and age. In addition, the results can be used to optimize towards more aortophilic stent grafts.

Changes in the microstructure of the human aortic adventitia under biaxial loading investigated by multi-photon microscopy.

Among the three layers of the aortic wall, the media is primarily responsible for its mechanical properties, but the adventitia prevents the aorta from overstretching and rupturing. The role of the adventitia is therefore crucial with regard to aortic wall failure, and understanding the load-induced changes in tissue microstructure is of high importance. Specifically, the focus of this study is on the changes in collagen and elastin microstructure in response to macroscopic equibiaxial loading applied to the aortic adventitia. To observe these changes, multi-photon microscopy imaging and biaxial extension tests were performed simultaneously. In particular, microscopy images were recorded at 0.02 stretch intervals. The microstructural changes of collagen fiber bundles and elastin fibers were quantified with the parameters of orientation, dispersion, diameter, and waviness. The results showed that the adventitial collagen was divided from one into two fiber families under equibiaxial loading conditions. The almost diagonal orientation of the adventitial collagen fiber bundles remained unchanged, but the dispersion was substantially reduced. No clear orientation of the adventitial elastin fibers was observed at any stretch level. The waviness of the adventitial collagen fiber bundles decreased under stretch, but the adventitial elastin fibers showed no change. These original findings highlight differences between the medial and adventitial layers and provide insight into the stretching process of the aortic wall. STATEMENT OF SIGNIFICANCE: To provide accurate and reliable material models, it is essential to understand the mechanical behavior of the material and its microstructure. Such understanding can be enhanced with tracking of the microstructural changes caused by mechanical loading of the tissue. This study provides therefore a unique dataset of structural parameters of the human aortic adventitia obtained under equibiaxial loading. The structural parameters describe orientation, dispersion, diameter, and waviness of collagen fiber bundles and elastin fibers. Eventually, the microstructural changes in the human aortic adventitia are compared with the microstructural changes in the human aortic media from a previous study. This comparison reveals the cutting-edge findings on the differences in the response to the loading between these two human aortic layers.

Tumor Microenvironment in Male Breast Carcinoma with Emphasis on Tumor Infiltrating Lymphocytes and PD-L1 Expression.

Male breast cancer (MBC) is rare and usually presents as a locally advanced disease. Stromal tumor-infiltrating lymphocytes (sTILs) are associated with a better response to neoadjuvant chemotherapy and improved prognosis in all molecular subtypes of female breast cancer, but their role in MBC is less clear. We studied sTILs and the expression of programmed cell death ligand 1 (PD-L1) and pan-TRK in MBC. We retrospectively studied 113 cases of MBC surgically treated between 1988 and 2015. The tumors were evaluated for histological type and grade, stage, intrinsic subtype and sTILs. We performed immunohistochemistry for PD-L1 (clone SP142) and pan-TRK (clone EPR17341) on tissue microarrays. Pan-TRK positive cases were further analyzed by next-generation sequencing. The median age was 69 years (range 60−77). Invasive carcinoma of no special type was found in 94.7% of cases, of which 53.1% were grade 2. Estrogen receptor was positive in 92% of the tumors, progesterone receptor in 85.8%, androgen receptor in 70.8%; 4.4% were human epidermal growth factor receptor 2 (HER2)-positive, and 55.8% HER2-low. 40.7% of tumors were luminal A and 51.3% luminal B, 4.4% HER2-enriched and 3.5% triple negative carcinoma. sTILs density was <50% in 96.4% of the tumors, >50% in 3.6% of the tumors. PD-L1 immune cell score >1% was found in 7.1% of the tumors (all of luminal subtype). A weak focal cytoplasmic pan-TRK staining was present in 8.8% but without NTRK fusion. Neither sTILs nor PD-L1 had statistically significant outcomes. Our findings suggest that a subset of MBC patients harbors an immunological environment characterized by increased sTILs with PD-L1 expression. These patients may potentially benefit from immune checkpoint inhibitor therapy. Frequent HER2-low may offer novel anti-HER2 treatment options.

Resilience and Protection of Health Care and Research Laboratory Workers During the SARS-CoV-2 Pandemic: Analysis and Case Study From an Austrian High Security Laboratory.

The SARS-CoV-2 pandemic has highlighted the interdependency of healthcare systems and research organizations on manufacturers and suppliers of personnel protective equipment (PPE) and the need for well-trained personnel who can react quickly to changing working conditions. Reports on challenges faced by research laboratory workers (RLWs) are rare in contrast to the lived experience of hospital health care workers. We report on experiences gained by RLWs (e.g., molecular scientists, pathologists, autopsy assistants) who significantly contributed to combating the pandemic under particularly challenging conditions due to increased workload, sickness and interrupted PPE supply chains. RLWs perform a broad spectrum of work with SARS-CoV-2 such as autopsies, establishment of virus cultures and infection models, development and verification of diagnostics, performance of virus inactivation assays to investigate various antiviral agents including vaccines and evaluation of decontamination technologies in high containment biological laboratories (HCBL). Performance of autopsies and laboratory work increased substantially during the pandemic and thus led to highly demanding working conditions with working shifts of more than eight hours working in PPE that stressed individual limits and also the ergonomic and safety limits of PPE. We provide detailed insights into the challenges of the stressful daily laboratory routine since the pandemic began, lessons learned, and suggest solutions for better safety based on a case study of a newly established HCBL (i.e., BSL-3 laboratory) designed for autopsies and research laboratory work. Reduced personal risk, increased resilience, and stress resistance can be achieved by improved PPE components, better training, redundant safety measures, inculcating a culture of safety, and excellent teamwork.

Changes in the microstructure of the human aortic medial layer under biaxial loading investigated by multi-photon microscopy.

Understanding the correlation between tissue architecture, health status, and mechanical properties is essential for improving material models and developing tissue engineering scaffolds. Since structural-based material models are state of the art, there is an urgent need for experimentally obtained structural parameters. For this purpose, the medial layer of nine human abdominal aortas was simultaneously subjected to equibiaxial loading and multi-photon microscopy. At each loading interval of 0.02, collagen and elastin fibers were imaged based on their second-harmonic generation signal and two-photon excited autofluorescence, respectively. The structural alterations in the fibers were quantified using the parameters of orientation, diameter, and waviness. The results of the mechanical tests divided the sample cohort into the ruptured and non-ruptured, and stiff and non-stiff groups, which were covered by the findings from histological investigations. The alterations in structural parameters provided an explanation for the observed mechanical behavior. In addition, the waviness parameters of both collagen and elastin fibers showed the potential to serve as indicators of tissue strength. The data provided address deficiencies in current material models and bridge multiscale mechanisms in the aortic media. STATEMENT OF SIGNIFICANCE: Available material models can reproduce, but cannot predict, the mechanical behavior of human aortas. This deficiency could be overcome with the help of experimentally validated structural parameters as provided in this study. Simultaneous multi-photon microscopy and biaxial extension testing revealed the microstructure of human aortic media at different stretch levels. Changes in the arrangement of collagen and elastin fibers were quantified using structural parameters such as orientation, diameter and waviness. For the first time, structural parameters of human aortic tissue under continuous loading conditions have been obtained. In particular, the waviness parameters at the reference configuration have been associated with tissue stiffness, brittleness, and the onset of atherosclerosis.

Host and microbiome features of secondary infections in lethal covid-19.

Secondary infections contribute significantly to covid-19 mortality but driving factors remain poorly understood. Autopsies of 20 covid-19 cases and 14 controls from the first pandemic wave complemented with microbial cultivation and RNA-seq from lung tissues enabled description of major organ pathologies and specification of secondary infections. Lethal covid-19 segregated into two main death causes with either dominant diffuse alveolar damage (DAD) or secondary pneumonias. The lung microbiome in covid-19 showed a reduced biodiversity and increased prototypical bacterial and fungal pathogens in cases of secondary pneumonias. RNA-seq distinctly mirrored death causes and stratified DAD cases into subgroups with differing cellular compositions identifying myeloid cells, macrophages and complement C1q as strong separating factors suggesting a pathophysiological link. Together with a prominent induction of inhibitory immune-checkpoints our study highlights profound alterations of the lung immunity in covid-19 wherein a reduced antimicrobial defense likely drives development of secondary infections on top of SARS-CoV-2 infection.

Sensitive and robust liquid biopsy-based detection of PIK3CA mutations in hormone-receptor-positive metastatic breast cancer patients.

BACKGROUND: The benefit of alpelisib in hormone-receptor-positive (HR+) metastatic breast cancer patients provided clinical evidence for the increasing importance of PIK3CA testing. We performed a comparison of liquid biopsy and tissue-based detection of PIK3CA mutations. MATERIALS AND METHODS: PIK3CA hotspot mutation analysis using a high-resolution SiMSen-Seq assay was performed in plasma from 93/99 eligible patients with HR+/HER2- breast cancer. Additionally, mFAST-SeqS was used to estimate the tumour fractions in plasma samples. In 72/93 patients, matched tissue was available and analysed using a customised Ion Torrent panel. RESULTS: PIK3CA mutations were detected in 48.6% of tissue samples and 47.3% of plasma samples, with identical PIK3CA mutation detected in 24/72 (33.3%) patients both in tissue and plasma. In 10 (13.9%) patients, mutations were only found in plasma, and in 6 (8.3%) patients, PIK3CA mutations found in tissue were not detectable in ctDNA. In 49/93 plasma samples without detectable PIK3CA mutations, 22 (44.9%) samples had elevated tumour fractions, implying true negative results. CONCLUSION: SiMSen-Seq-based detection of PIK3CA mutations in plasma shows advantageous concordance with the tissue analyses. A combination with an untargeted approach for detecting ctDNA fractions may confirm a negative PIK3CA result and enhance the performance of the SiMSen-Seq test.

Intra-Tumour Heterogeneity Is One of the Main Sources of Inter-Observer Variation in Scoring Stromal Tumour Infiltrating Lymphocytes in Triple Negative Breast Cancer.

Stromal tumour infiltrating lymphocytes (sTILs) are a strong prognostic marker in triple negative breast cancer (TNBC). Consistency scoring sTILs is good and was excellent when an internet-based scoring aid developed by the TIL-WG was used to score cases in a reproducibility study. This study aimed to evaluate the reproducibility of sTILs assessment using this scoring aid in cases from routine practice and to explore the potential of the tool to overcome variability in scoring. Twenty-three breast pathologists scored sTILs in digitized slides of 49 TNBC biopsies using the scoring aid. Subsequently, fields of view (FOV) from each case were selected by one pathologist and scored by the group using the tool. Inter-observer agreement was good for absolute sTILs (ICC 0.634, 95% CI 0.539-0.735, p < 0.001) but was poor to fair using binary cutpoints. sTILs heterogeneity was the main contributor to disagreement. When pathologists scored the same FOV from each case, inter-observer agreement was excellent for absolute sTILs (ICC 0.798, 95% CI 0.727-0.864, p < 0.001) and good for the 20% (ICC 0.657, 95% CI 0.561-0.756, p < 0.001) and 40% (ICC 0.644, 95% CI 0.546-0.745, p < 0.001) cutpoints. However, there was a wide range of scores for many cases. Reproducibility scoring sTILs is good when the scoring aid is used. Heterogeneity is the main contributor to variance and will need to be overcome for analytic validity to be achieved.

The OncoMasTR Test Predicts Distant Recurrence in Estrogen Receptor-Positive, HER2-Negative Early-Stage Breast Cancer: A Validation Study in ABCSG Trial 8.

PURPOSE: To validate the clinical performance of the OncoMasTR Risk Score in the biomarker cohort of Austrian Breast and Colorectal Cancer Study Group (ABCSG) Trial 8. EXPERIMENTAL DESIGN: We evaluated the OncoMasTR test in 1,200 formalin-fixed, paraffin-embedded (FFPE) surgical specimens from postmenopausal women with estrogen receptor (ER)-positive, human epidermal growth factor receptor 2 (HER2)-negative primary breast cancer with 0 to 3 involved lymph nodes in the prospective, randomized ABCSG Trial 8. Time to distant recurrence (DR) was analyzed by Cox models. RESULTS: The OncoMasTR Risk Score categorized 850 of 1,087 (78.2%) evaluable patients as "low risk". At 10 years, the DR rate for patients in the low-risk group was 5.8% versus 21.1% for patients in the high-risk group (P < 0.0001, absolute risk reduction 15.3%). The OncoMasTR Risk Score was highly prognostic for prediction of DR in years 0 to 10 in all patients [HR 1.91, 95% confidence interval (CI) 1.62-2.26, P < 0.0001; C-index 0.73], in patients that were node negative (HR 1.79, 95% CI, 1.43-2.24, P < 0.0001; C-index 0.72), and in patients with 1 to 3 involved lymph nodes (HR 1.93, 95% CI, 1.44-2.58, P < 0.0001; C-index 0.71). The OncoMasTR Risk Score provided significant additional prognostic information beyond clinical parameters, Ki67, Nottingham Prognostic Index, and Clinical Treatment Score. CONCLUSIONS: OncoMasTR Risk Score is highly prognostic for DR in postmenopausal women with ER-positive, HER2-negative primary breast cancer with 0 to 3 involved lymph nodes. In combination with prior validation studies, this fully independent validation in ABCSG Trial 8 provides level 1B evidence for the prognostic capability of the OncoMasTR Risk Score.

Interpretable survival prediction for colorectal cancer using deep learning.

Deriving interpretable prognostic features from deep-learning-based prognostic histopathology models remains a challenge. In this study, we developed a deep learning system (DLS) for predicting disease-specific survival for stage II and III colorectal cancer using 3652 cases (27,300 slides). When evaluated on two validation datasets containing 1239 cases (9340 slides) and 738 cases (7140 slides), respectively, the DLS achieved a 5-year disease-specific survival AUC of 0.70 (95% CI: 0.66-0.73) and 0.69 (95% CI: 0.64-0.72), and added significant predictive value to a set of nine clinicopathologic features. To interpret the DLS, we explored the ability of different human-interpretable features to explain the variance in DLS scores. We observed that clinicopathologic features such as T-category, N-category, and grade explained a small fraction of the variance in DLS scores (R2 = 18% in both validation sets). Next, we generated human-interpretable histologic features by clustering embeddings from a deep-learning-based image-similarity model and showed that they explained the majority of the variance (R2 of 73-80%). Furthermore, the clustering-derived feature most strongly associated with high DLS scores was also highly prognostic in isolation. With a distinct visual appearance (poorly differentiated tumor cell clusters adjacent to adipose tissue), this feature was identified by annotators with 87.0-95.5% accuracy. Our approach can be used to explain predictions from a prognostic deep learning model and uncover potentially-novel prognostic features that can be reliably identified by people for future validation studies.

Predicting prostate cancer specific-mortality with artificial intelligence-based Gleason grading.

Background: Gleason grading of prostate cancer is an important prognostic factor, but suffers from poor reproducibility, particularly among non-subspecialist pathologists. Although artificial intelligence (A.I.) tools have demonstrated Gleason grading on-par with expert pathologists, it remains an open question whether and to what extent A.I. grading translates to better prognostication. Methods: In this study, we developed a system to predict prostate cancer-specific mortality via A.I.-based Gleason grading and subsequently evaluated its ability to risk-stratify patients on an independent retrospective cohort of 2807 prostatectomy cases from a single European center with 5-25 years of follow-up (median: 13, interquartile range 9-17). Results: Here, we show that the A.I.'s risk scores produced a C-index of 0.84 (95% CI 0.80-0.87) for prostate cancer-specific mortality. Upon discretizing these risk scores into risk groups analogous to pathologist Grade Groups (GG), the A.I. has a C-index of 0.82 (95% CI 0.78-0.85). On the subset of cases with a GG provided in the original pathology report (n = 1517), the A.I.'s C-indices are 0.87 and 0.85 for continuous and discrete grading, respectively, compared to 0.79 (95% CI 0.71-0.86) for GG obtained from the reports. These represent improvements of 0.08 (95% CI 0.01-0.15) and 0.07 (95% CI 0.00-0.14), respectively. Conclusions: Our results suggest that A.I.-based Gleason grading can lead to effective risk stratification, and warrants further evaluation for improving disease management.

Determining breast cancer biomarker status and associated morphological features using deep learning.

Background: Breast cancer management depends on biomarkers including estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 (ER/PR/HER2). Though existing scoring systems are widely used and well-validated, they can involve costly preparation and variable interpretation. Additionally, discordances between histology and expected biomarker findings can prompt repeat testing to address biological, interpretative, or technical reasons for unexpected results. Methods: We developed three independent deep learning systems (DLS) to directly predict ER/PR/HER2 status for both focal tissue regions (patches) and slides using hematoxylin-and-eosin-stained (H&E) images as input. Models were trained and evaluated using pathologist annotated slides from three data sources. Areas under the receiver operator characteristic curve (AUCs) were calculated for test sets at both a patch-level (>135 million patches, 181 slides) and slide-level (n = 3274 slides, 1249 cases, 37 sites). Interpretability analyses were performed using Testing with Concept Activation Vectors (TCAV), saliency analysis, and pathologist review of clustered patches. Results: The patch-level AUCs are 0.939 (95%CI 0.936-0.941), 0.938 (0.936-0.940), and 0.808 (0.802-0.813) for ER/PR/HER2, respectively. At the slide level, AUCs are 0.86 (95%CI 0.84-0.87), 0.75 (0.73-0.77), and 0.60 (0.56-0.64) for ER/PR/HER2, respectively. Interpretability analyses show known biomarker-histomorphology associations including associations of low-grade and lobular histology with ER/PR positivity, and increased inflammatory infiltrates with triple-negative staining. Conclusions: This study presents rapid breast cancer biomarker estimation from routine H&E slides and builds on prior advances by prioritizing interpretability of computationally learned features in the context of existing pathological knowledge.