In medical image segmentation and its semantic variant, annotations from multiple experts, when available, are used to generate consensus labels for training machine learning models. However, the annotator consistency is generally not captured in the consensus labels and can negatively impact the reliability of these models. In this study, we propose a novel approach based on the concept of self-consistency, which characterizes an expert's behavior by quantifying annotation certainty. The proposed method is model-agnostic and can be used as a universal preprocessing step for any segmentation backbone. To validate our approach, we apply it to semantic image segmentation in the context of prostate cancer grading, a domain known to be subject to both intra- and inter-expert variability. We use the PANDA dataset and generate synthetic experts to demonstrate that our approach provides valuable insights into intra-expert variability. Compared to other probabilistic approaches, such as soft and smooth labeling, our method improves the quality of the probabilistic consensus, thereby improving deep network training for semantic segmentation. We extend these findings to two real-world cancer datasets. The results highlight the potential of our approach to address challenges caused by annotation uncertainty in digital pathology. The code will be publicly available upon publication at https://github.com/lauragj95/SC_multi_expert.

Background: Tubulointerstitial hypoxia is a key factor for lupus nephritis progression to end-stage renal disease. Numerous aquaporins (AQPs) are expressed by renal tubules and are essential for their proper functioning. The aim of this study is to characterize the tubular expression of AQP1, AQP2 and AQP3, which could provide a better understanding of tubulointerstitial stress during lupus nephritis. Methods: This retrospective monocentric study was conducted at Erasme-HUB Hospital. We included 37 lupus nephritis samples and 9 healthy samples collected between 2000 and 2020, obtained from the pathology department. Immunohistochemistry was performed to target AQP1, AQP2 and AQP3 and followed by digital analysis. Results: No difference in AQP1, AQP2 and AQP3 staining location was found between healthy and lupus nephritis samples. However, we observed significant differences between these two groups, with a decrease in AQP1 expression in the renal cortex and in AQP3 expression in the cortex and medulla. In the subgroup of proliferative glomerulonephritis (class III/IV), this decrease in AQPs expression was more pronounced, particularly for AQP3. In addition, within this subgroup, we detected lower AQP2 expression in patients with higher interstitial inflammation score and lower AQP3 expression when higher interstitial fibrosis and tubular atrophy were present. Conclusions: We identified significant differences in the expression of aquaporins 1, 2, and 3 in patients with lupus nephritis. These findings strongly suggest that decreased AQP expression could serve as an indicator of tubular injury. Further research is warranted to evaluate AQP1, AQP2, and AQP3 as prognostic markers in both urinary and histological assessments of lupus nephritis.

Segmentation and classification of large numbers of instances, such as cell nuclei, are crucial tasks in digital pathology for accurate diagnosis. However, the availability of high-quality datasets for deep learning methods is often limited due to the complexity of the annotation process. In this work, we investigate the impact of noisy annotations on the training and performance of a state-of-the-art CNN model for the combined task of detecting, segmenting and classifying nuclei in histopathology images. In this context, we investigate the conditions for determining an appropriate number of training epochs to prevent overfitting to annotation noise during training. Our results indicate that the utilisation of a small, correctly annotated validation set is instrumental in avoiding overfitting and maintaining model performance to a large extent. Additionally, our findings underscore the beneficial role of pre-training.

Pancreatic ductal adenocarcinoma (PDAC) is reported to be amongst the cancers with the lowest survival rate at 5 years. In the present study we aimed to validate a targeted next-generation sequencing (tNGS) panel to use in clinical routine, investigating genes important for PDAC diagnostic, prognostic and potential theragnostic aspect. In this NGS panel we also designed target regions to inquire about loss of heterozygosity (LOH) of chromosome 18 that has been described to be possibly linked to a worse disease progression. Copy number alteration has also been explored for a subset of genes. The last two methods are not commonly used for routine diagnostic with tNGS panels and we investigated their possible contribution to better characterize PDAC. A series of 140 formalin-fixed paraffin-embedded (FFPE) PDAC samples from 140 patients was characterized using this panel. Ninety-two % of patients showed alterations in at least one of the investigated genes (most frequent KRAS, TP53, SMAD4, CDKN2A and RNF43). Regarding LOH evaluation, we were able to detect chr18 LOH starting at 20% cell tumor percentage. The presence of LOH on chr18 is associated with a worse disease- and metastasis-free survival, in uni- and multivariate analyses. The present study validates the use of a tNGS panel for PDAC characterization, also evaluating chr18 LOH status for prognostic stratification.

Background Previous breast carcinoma studies focused on the evaluation of tumour-infiltrating lymphocytes (TILs) or of tumoural stroma via the tumour stroma ratio (TSR). Few studies assessed peritumoural lymphocytes and almost no studies investigated a possible relationship between lymphocytes and stroma. This prompted us to evaluate the amount of tumour cells, intra- and peritumoural lymphocytes, and stroma in breast cancer to support the hypothesis that the stroma may block the infiltration of lymphocytes inside the tumour. Methods We collected a retrospective series of 158 breast cancers (<25 mm). In addition to standard TILs and TSR evaluations, we assessed the percentages of tumour cells, stromal myofibroblasts, intra- and peritumoural lymphocytes on full-section tumours with haematoxylin and eosin and immunohistochemical staining. Results We showed significant negative correlations between the amounts of stroma and both intra- and peritumoural lymphocyte percentages. Considering the estrogen receptor positive invasive breast cancer of no special type cases, we showed that TSR had a positive prognostic value with an optimal threshold of 10%. Conclusions This study is one of the first to show inverse correlations between tumoural stroma amount and intra- and peritumoural lymphocyte percentages, which supports the hypothesis that tumoural stroma can prevent the recruitment of lymphocytes around and within the tumour.

BACKGROUND:In the era of "precision medicine," the availability of high-quality tumor biomarker tests is critical and tumor proliferation evaluated by Ki-67 antibody is one of the most important prognostic factors in breast cancer. But the evaluation of Ki-67 index has been shown to suffer from some interobserver variability. The goal of the study is to develop an easy, automated, and reliable Ki-67 assessment approach for invasive breast carcinoma in routine practice.PATIENTS AND METHODS:A total of 151 biopsies of invasive breast carcinoma were analyzed. The Ki-67 index was evaluated by 2 pathologists with MIB-1 antibody as a global tumor index and also in a hotspot. These 2 areas were also analyzed by digital image analysis (DIA).RESULTS:For Ki-67 index assessment, in the global and hotspot tumor area, the concordances were very good between DIA and pathologists when DIA focused on the annotations made by pathologist (0.73 and 0.83, respectively). However, this was definitely not the case when DIA was not constrained within the pathologist's annotations and automatically established its global or hotspot area in the whole tissue sample (concordance correlation coefficients between 0.28 and 0.58).CONCLUSIONS:The DIA technique demonstrated a meaningful concordance with the indices evaluated by pathologists when the tumor area is previously identified by a pathologist. In contrast, basing Ki-67 assessment on automatic tissue detection was not satisfactory and provided bad concordance results. A representative tumoral zone must therefore be manually selected prior to the measurement made by the DIA.

Background: SMAD4 is inactivated in 50–55% of pancreatic ductal adenocarcinomas (PDACs). SMAD4 loss of expression has been described as a negative prognostic factor in PDAC associated with an increased rate of metastasis and resistance to therapy. However, the impact of SMAD4 inactivation in patients receiving neoadjuvant therapy (NAT) is not well characterized. The aim of our study was to investigate whether SMAD4 status is a prognostic and predictive factor in patients receiving NAT. Methods: We retrospectively analyzed 59 patients from a single center who underwent surgical resection for primary PDAC after NAT. SMAD4 nuclear expression was assessed by immunohistochemistry, and its relationship to clinicopathologic variables and survival parameters was evaluated. Interaction testing was performed between SMAD4 status and the type of NAT. Results: 49.15% of patients presented loss of SMAD4. SMAD4 loss was associated with a higher positive lymph node ratio (p = 0.03), shorter progression-free survival (PFS) (p = 0.02), and metastasis-free survival (MFS) (p = 0.02), but it was not an independent prognostic biomarker in multivariate analysis. Interaction tests demonstrated that patients with SMAD4-positive tumors receiving FOLFIRINOX-based NAT showed the best outcome. Conclusion: This study highlights the potential prognostic and predictive role of SMAD4 status in PDAC patients receiving FOLFIRINOX-based NAT.

In digital pathology, segmentation between tissue and glass slide is a very common pre-processing step in image processing pipelines. It is often presented as relatively trivial, and solved using ad-hoc heuristics that are not always precisely defined nor justified. Most tissue segmentation pipelines start by reducing the color image to a single-channel representation, grayscale being the most common. We show in this study that representations that focus on the colorfulness or entropy offer better separability between tissue and background, and lead to better results in simple thresholding pipelines.

Panoptic Quality (PQ), designed for the task of “Panoptic Segmentation” (PS), has been used in several digital pathology challenges and publications on cell nucleus instance segmentation and classification (ISC) since its introduction in 2019. Its purpose is to encompass the detection and the segmentation aspects of the task in a single measure, so that algorithms can be ranked according to their overall performance. A careful analysis of the properties of the metric, its application to ISC and the characteristics of nucleus ISC datasets, shows that is not suitable for this purpose and should be avoided. Through a theoretical analysis we demonstrate that PS and ISC, despite their similarities, have some fundamental differences that make PQ unsuitable. We also show that the use of the Intersection over Union as a matching rule and as a segmentation quality measure within PQ is not adapted for such small objects as nuclei. We illustrate these findings with examples taken from the NuCLS and MoNuSAC datasets. The code for replicating our results is available on GitHub ( https://github.com/adfoucart/panoptic-quality-suppl ).

Background Lupus nephritis is a severe and life-threatening manifestation of systemic lupus erythematosus. Tubulointerstitial hypoxia is a key factor in the progression to end-stage renal disease. Numerous aquaporins are expressed by renal tubules and are essential for their proper functioning. Several mouse models have shown their involvement in the regulation of renal inflammation during acute stress episodes. However, the expression of aquaporins (AQP) in human lupus nephritis has been poorly studied. Objectives The aim of this study is to characterise the tubular expression of AQP1, AQP2 and AQP3 which could provide a better understanding of tubulointerstitial stress during lupus nephritis. Methods This retrospective monocentric study was conducted at Erasme-HUB Hospital, Brussels with the approval of the ULB-Erasme Ethical Committee (P2020/710). A total of 37 lupus nephritis samples and 9 healthy samples collected between 2000 and 2020 were obtained from the biobank of the pathology department. Kidney biopsy sections were reviewed according to the ISN/RPS 2018 classification. Immunohistochemistry was performed to detect AQP1, AQP2 and AQP3 and followed by digital image analysis. Digital image quantification analysis was performed using ImageJ software. Results We observed weak expression of AQP1 in glomeruli and strong expression on the apical side of proximal convoluted tubules in the cortex. In the medulla, it was confined to the descending loop of Henle and the vasa recta. AQP2 was exclusively expressed on the apical side of collecting tubules in both the cortex and the medulla. AQP3 was not detected in glomeruli, weakly expressed on the basolateral side of proximal convoluted tubules, and strongly expressed on the basolateral side of distal convoluted tubules and collecting tubules. No difference in staining location was found between healthy and lupus nephritis kidney biopsies. By digital quantitative analysis, we observed a significant decrease in AQP1 expression in the renal cortex (p<0.0001), a non-significant trend towards a decrease in AQP2 expression and a significant cortical and medullary decrease in AQP3 expression (p<0.001). This decrease was more pronounced in the subgroup of membranoproliferative glomerulonephritis (class III/IV), particularly for AQP3 (p<0.05). Within the subgroup of membranoproliferative glomerulonephritis, there was a strong negative correlation between both cortical and medullary AQP2 expression and interstitial inflammation (r=-0.5238; p<0.01). Decreased cortical AQP3 expression was strongly negatively correlated with interstitial fibrosis (r=-0.6651; p<0.001) and tubular atrophy (r=–0.6651; p<0.001). Conclusion We found a significant decrease in the expression of several AQPs in the parenchyma of lupus nephritis. We believe that this decrease is a feature of tubulointerstitial damage. Immunohistochemical analyses of AQPs on lupus nephritis samples could help to assess the tubulointerstitial hypoxia and cell damage and therefore renal prognosis. References [1]Maria, N.I., Davidson, A. Protecting the kidney in systemic lupus erythematosus: from diagnosis to therapy. Nat Rev Rheumatol 16, 255–267 (2020).[2]Matsuzaki, T., Yaguchi, T., Shimizu, K. et al. The distribution and function of aquaporins in the kidney: resolved and unresolved questions. Anat Sci Int 92, 187–199 (2017). Acknowledgements We thank DIAPath for their technical assistance. The CMMI is supported by the European Regional Development Fund and the Walloon Region. Disclosure of Interests None Declared.

Biomedical image analysis competitions often rank the participants based on a single metric that com-bines assessments of different aspects of the task at hand. While this is useful for declaring a single winner for a competition, it makes it difficult to assess the strengths and weaknesses of participating al-gorithms. By involving multiple capabilities (detection, segmentation and classification) and releasing the prediction masks provided by several teams, the MoNuSAC 2020 challenge provides an interesting op-portunity to look at what information may be lost by using entangled metrics. We analyse the challenge results based on the "Panoptic Quality" (PQ) used by the organizers, as well as on disentangled metrics that assess the detection, classification and segmentation abilities of the algorithms separately. We show that the PQ hides interesting aspects of the results, and that its sensitivity to small changes in the pre-diction masks makes it hard to interpret these results and to draw useful insights from them. Our results also demonstrate the necessity to have access, as much as possible, to the raw predictions provided by the participating teams so that challenge results can be more easily analysed and thus more useful to the research community.(c) 2023 Elsevier Ltd. All rights reserved.

Digital pathology image analysis challenges have been organised regularly since 2010, often with events hosted at major conferences and results published in high-impact journals. These challenges mobilise a lot of energy from organisers, participants, and expert annotators (especially for image segmentation challenges). This study reviews image segmentation challenges in digital pathology and the top-ranked methods, with a particular focus on how reference annotations are generated and how the methods' predictions are evaluated. We found important shortcomings in the handling of inter-expert disagreement and the relevance of the evaluation process chosen. We also noted key problems with the quality control of various challenge elements that can lead to uncertainties in the published results. Our findings show the importance of greatly increasing transparency in the reporting of challenge results, and the need to make publicly available the evaluation codes, test set annotations and participants' predictions. The aim is to properly ensure the reproducibility and interpretation of the results and to increase the potential for exploitation of the substantial work done in these challenges.

In digital pathology, the annotation process can be complex and time-consuming, the reason why the available annotations are often imperfect. In this paper, we focus on the detection, segmentation, and classification of multiple instances within histopathology images in the presence of noisy annotations. After highlighting such noise in the training set provided by the Multi-Organ Nuclei Segmentation and Classification Challenge (MoNuSAC 2020), we develop and compare two strategies of model prediction-based filtering, in order to obtain a cleaner dataset for training a deep neural network (HoVer-Net). We show the superiority and positive impact on performance of a filtering method based on analysing model output at instance level rather than pixel level.

Netrin-1 is upregulated in cancers as a protumoural mechanism 1 . Here we describe netrin-1 upregulation in a majority of human endometrial carcinomas (ECs) and demonstrate that netrin-1 blockade, using an anti-netrin-1 antibody (NP137), is effective in reduction of tumour progression in an EC mouse model. We next examined the efficacy of NP137, as a first-in-class single agent, in a Phase I trial comprising 14 patients with advanced EC. As best response we observed 8 stable disease (8 out of 14, 57.1%) and 1 objective response as RECIST v.1.1 (partial response, 1 out of 14 (7.1%), 51.16% reduction in target lesions at 6 weeks and up to 54.65% reduction during the following 6 months). To evaluate the NP137 mechanism of action, mouse tumour gene profiling was performed, and we observed, in addition to cell death induction, that NP137 inhibited epithelial-to-mesenchymal transition (EMT). By performing bulk RNA sequencing (RNA-seq), spatial transcriptomics and single-cell RNA-seq on paired pre- and on-treatment biopsies from patients with EC from the NP137 trial, we noted a net reduction in tumour EMT. This was associated with changes in immune infiltrate and increased interactions between cancer cells and the tumour microenvironment. Given the importance of EMT in resistance to current standards of care 2 , we show in the EC mouse model that a combination of NP137 with carboplatin-paclitaxel outperformed carboplatin-paclitaxel alone. Our results identify netrin-1 blockade as a clinical strategy triggering both tumour debulking and EMT inhibition, thus potentially alleviating resistance to standard treatments.

Epithelial-to-mesenchymal transition (EMT) regulates tumour initiation, progression, metastasis and resistance to anti-cancer therapy1–7. Although great progress has been made in understanding the role of EMT and its regulatory mechanisms in cancer, no therapeutic strategy to pharmacologically target EMT has been identified. Here we found that netrin-1 is upregulated in a primary mouse model of skin squamous cell carcinoma (SCC) exhibiting spontaneous EMT. Pharmacological inhibition of netrin-1 by administration of NP137, a netrin-1-blocking monoclonal antibody currently used in clinical trials in human cancer (ClinicalTrials.gov identifier NCT02977195 ), decreased the proportion of EMT tumour cells in skin SCC, decreased the number of metastases and increased the sensitivity of tumour cells to chemotherapy. Single-cell RNA sequencing revealed the presence of different EMT states, including epithelial, early and late hybrid EMT, and full EMT states, in control SCC. By contrast, administration of NP137 prevented the progression of cancer cells towards a late EMT state and sustained tumour epithelial states. Short hairpin RNA knockdown of netrin-1 and its receptor UNC5B in EPCAM+ tumour cells inhibited EMT in vitro in the absence of stromal cells and regulated a common gene signature that promotes tumour epithelial state and restricts EMT. To assess the relevance of these findings to human cancers, we treated mice transplanted with the A549 human cancer cell line—which undergoes EMT following TGFβ1 administration8,9—with NP137. Netrin-1 inhibition decreased EMT in these transplanted A549 cells. Together, our results identify a pharmacological strategy for targeting EMT in cancer, opening up novel therapeutic interventions for anti-cancer therapy. Netrin-1 is upregulated in cancer models that undergo spontaneous epithelial-to-mesenchymal transition, and its targeting blocks the progression of tumour cells to a late mesenchymal state, suggesting possible therapeutic applications.

Introduction COVID-19-infected patients harbour neurological symptoms such as stroke and anosmia, leading to the hypothesis that there is direct invasion of the central nervous system (CNS) by SARS-CoV-2. Several studies have reported the neuropathological examination of brain samples from patients who died from COVID-19. However, there is still sparse evidence of virus replication in the human brain, suggesting that neurologic symptoms could be related to mechanisms other than CNS infection by the virus. Our objective was to provide an extensive review of the literature on the neuropathological findings of postmortem brain samples from patients who died from COVID-19 and to report our own experience with 18 postmortem brain samples. Material and methods We used microscopic examination, immunohistochemistry (using two different antibodies) and PCR-based techniques to describe the neuropathological findings and the presence of SARS-CoV-2 virus in postmortem brain samples. For comparison, similar techniques (IHC and PCR) were applied to the lung tissue samples for each patient from our cohort. The systematic literature review was conducted from the beginning of the pandemic in 2019 until June 1st, 2022. Results In our cohort, the most common neuropathological findings were perivascular haemosiderin-laden macrophages and hypoxic-ischaemic changes in neurons, which were found in all cases (n = 18). Only one brain tissue sample harboured SARS-CoV-2 viral spike and nucleocapsid protein expression, while all brain cases harboured SARS-CoV-2 RNA positivity by PCR. A colocalization immunohistochemistry study revealed that SARS-CoV-2 antigens could be located in brain perivascular macrophages. The literature review highlighted that the most frequent neuropathological findings were ischaemic and haemorrhagic lesions, including hypoxic/ischaemic alterations. However, few studies have confirmed the presence of SARS-CoV-2 antigens in brain tissue samples. Conclusion This study highlighted the lack of specific neuropathological alterations in COVID-19-infected patients. There is still no evidence of neurotropism for SARS-CoV-2 in our cohort or in the literature.

Feature-based registration has become increasingly popular in digital pathology for achieving initial global alignment between image pairs. However, the selection of algorithms used in this approach is often not well-justified. Specifically, the choice of local feature descriptor is rarely, if ever, discussed in the context of digital pathology. The majority of feature-based whole-slide image registration methods rely on the SIFT descriptor. In this study, we demonstrate that the choice of descriptor significantly influences the quality of registration results and that the BRIEF descriptor captures more optimal information for histological image registration.

Fibroblast activation protein-α (FAPα) is a marker of activated fibroblasts that can be selectively targeted by an inhibitor (FAPI) and visualised by PET/CT imaging. We evaluated whether the measurement of FAPα in bronchoalveolar lavage fluids (BALF) and the uptake of FAPI by PET/CT could be used as biomarkers of fibrogenesis. The dynamics of lung uptake of 18F-labeled FAPI ([18F]FAPI-74) was assessed in the bleomycin mouse model at various time points and using different concentrations of bleomycin by PET/CT. FAPα was measured in BALFs from these bleomycin-treated and control mice. FAPα levels were also assessed in BALFs from controls and patients with idiopathic pulmonary fibrosis (IPF). Bleomycin-treated mice presented a significantly higher uptake of [18F]FAPI-74 during lung fibrinogenesis (days 10 and 16 after instillation) compared to control mice. No significant difference was observed at initial inflammatory phase (3 days) and when fibrosis was already established (28 days). [18F]FAPI-74 tracer was unable to show a dose-response to bleomycin treatment. On the other hand, BALF FAPα levels were steeply higher in bleomycin-treated mice at day 10 and a significant dose-response effect was observed. Moreover, FAPα levels were strongly correlated with lung fibrosis as measured by the modified Aschroft histological analysis, hydroxyproline and the percentage of weight loss. Importantly, higher levels of FAPα were observed in IPF patients where the disease was progressing as compared to stable patients and controls. Moreover, patients with FAPα BALF levels higher than 192.5 pg/mL presented a higher risk of progression, transplantation or death compared to patients with lower levels. Our preclinical data highlight a specific increase of [18F]FAPI-74 lung uptake during the fibrotic phase of the bleomycin murine model. The measurement of FAPα in BALF appears to be a promising marker of the fibrotic activity in preclinical models of lung fibrosis and in IPF patients. Further studies are required to confirm the role of FAPα in BALF as biomarker of IPF activity and assess the relationship between FAPα levels in BALF and [18F]FAPI-74 uptake on PET/CT in patients with fibrotic lung disease.