Automated Analysis of p53 Immunohistochemistry in Patients with Barrett's Esophagus Predicts Disease Progression to Advanced Neoplasia

Azfar Neyaz, Robert Odze, Deepa T. Patil, Vikram Deshpande

There are currently no validated markers to predict Barrett's esophagus (BE) patients at a high risk of progressing to advanced neoplasia.

TP53 mutation is highly recurrent in esophageal adenocarcinomas and is also detected in patients with non-dysplastic BE, predominantly those that progress to advanced neoplasia.

Showcased at the USCAP 109th Annual Meeting - March 2020


AI Deep Learning Tumour Detection Directly on ER, PR and Ki-67 IHC Slides Yields a Single Slide Automated Workflow with High Concordance to Manual Scoring

Holten-Rossing, H., Klingberg, H.

Assessment of ER, PR and Ki 67 provides essential prognostic information in the classification of breast carcinomas 1 Conventional manual assessment of these biomarkers has shown inter and intra observer variation and are both tedious and time demanding when following the ASCO/CAP guidelines which have paved the way for more accurate digital image analysis (DIA) methods.

The development of artificial intelligence (AI) has made it possible to automatically detect tumour in IHC slides without the need of cytokeratin (CK) stains. Earlier DIA techniques such as VirtualDoubleStaining ™ (VDS) has relied on sequential CK stains in tumour area identification. Introduction of AI in DIA in the assessment of ER, PR and Ki-67 will thus eliminate the need for additional serial sectioning and CK staining and this will reduce both costs and valuable time that could be distributed to less prioritized diagnostic areas

The aim of this study was to assess the performance of AI deep learning tumour detection and subsequent automated DIA ER, PR or Ki 67 scoring in comparison to conventional manual assessment.


Importance of Automated Invasive Tumor Detection and Hot Spot Identification in Ki-67 Assessment in Breast Carcinomas

Klingberg, H., Holten-Rossing, H., Tindbæk, G., Steiniche, T., Lippert, M.

Accurate assessment of the Ki-67 proliferation index (PI) is an important tool in breast carcinoma classification. According to the Danish national guidelines by the Danish Breast Cancer Cooperative Group (DBCG) the Ki-67 assessment should only be assessed in the invasive front and hot spot of invasive tumors. Non-invasive breast tumor regions (DCIS) are identified by the presence of myoepithelial cells (p63 positive nuclei). Ki-67 assessments in hot spots within invasive tumor regions, have been shown to be superior in performance compared to manual assessments giving rise to more accurate diagnosis1,2. Our aim was to evaluate the PI values in hot spots for either all tumor regions or limited to invasive tumor regions only using digital image analysis (DIA) and compare this to manual Ki-67 PI assessment.


Robust and Generalizable Nuclei Segmentation Using Deep Learning

Eschen, C., Medicine and Technology, Technical University of Denmark

Identification of cell nuclei in biomedical images is of great importance for research, drug discovery and diagnosis of disease. Examination of cancer suspicious tissue on the microscopic level is the golden standard for diagnosis of almost any cancer type. One of the greatest difficulties in segmentation of cell nuclei in histopathology is separation of adjoining cell nuclei. Separation of adjoining cell nuclei in deep learning is difficult due to the fact that convolutional neural networks (CNN) typically use a categorical cross entropy, which only work on a pixel-wise level and therefore lacks more global information context. Therefore, there is an increasing demand for robust segmentation algorithms which are designed for separation of cell nuclei. We present improved segmentation results using a feature engineered weighted loss and an adverserial loss. The best results were obtained using a U-Net architecture and an adverserial loss yielding an AJI equal to 0.452 on the validation set on the MoNuSeg grand challenge. We denote this architecture Pix2pix U-Net.


Multispectral Imaging Reveals Unique Macrophage Profiles Associated with Type of Liver Disease and Fibrosis Stage

Saldarriaga, O., Booth, A., Burks, J., Utay, N., Yi, M., Ferguson, M., Stevenson, H.

Intrahepatic macrophages macrophages macrophages greatly greatly greatlyimpact impact the thecomposition compositionof the thehepatic hepatic microenvironment, host immune response response response and development of fibrosis fibrosis. Studies Studies of human humanintrahepatic intrahepatic intrahepatic macrophages macrophages macrophages macrophagescan be challenging challenging forfor several several severalreasons reasons : (i) they are difficult difficult to isolate isolate from human liver liver tissue. (ii) they theybecome activated and change their phenotype phenotypephenotypephenotype when isolated or manipulated manipulated. (iii) in vitro vitro and mouse model systems systems systemsof HCV infection infectionor fatty fattyliver liver disease disease diseasedo not closely mimic the long -term, term,chronic chronicinfections infections that are observed in humans. We are using usingspectral spectral spectralimaging microscopy microscopy with advanced advanced advanced imaging analysis analysisanalysis software programs to analyze analyzeanalyze intrahepatic intrahepatic intrahepatic macrophages macrophages macrophages in situ situ in human humanliver biopsies biopsies. This platform platform is optimized optimized optimizedfor multiplex multipleximmunofluorescence immunofluorescence immunofluorescence immunofluorescencestaining stainingof formalin formalin -fixed paraffin -embedded embedded tissues and does does not compromise compromise the hepatic hepatic architecture. This approach will willallowallow allow us to gain an in -depth understanding understanding understanding understandingof how variations variations variations in human hepatic macrophage macrophage macrophage profiles affect affectthe host immune immune response response response and development of hepatic fibrosis fibrosis.


Biomarker Colocalization Analysis of a Virtual 12-plex using Discovery Chromogenic Dyes and Tissuealign™ Co-registration Software

Freiberg, B., Baird, R. 

The tumor micro-environment plays an important role in the diagnosis, prognosis and treatment regimens that patients receive. As such, understanding the specific changes in the immune-oncology (I/O) milieu presentation for tumors is key to developing novel therapeutics, new treatment regimens and ultimately increasing the survival and long-term prognosis for cancer patients.

Imaging the tumor milieu has inherent problems as there are many different types of immune cells to identify. In order to increase the number of biomarkers that can be analyzed concurrently, we propose a combination of multi-chromogen dyes and software for co-registration of sequential serial sections. These co-registered images produce highly multiplexed virtual images (8-, 10-, 15-plex or more) in which the tumor micro-environment can then be interrogated.


H.P. Acthar® Gel Inhibits Fibrosis, Renal Tubular Damage, and Glomerular Injury in 8- and 12-Week Puromycin-Induced Renal Injury Model

Hayes, K., Warner, E., Bollinger, C., Wright, D., Fitch, R.

This study assessed the efficacy of RCI on the reduction of proteinuria and protection of renal damage in a preclinical FSGS model. 


Automated Ki-67 Assessment: From Invasive Tumor Component Detection to Ki-67 Quantification in Hot Spots

Omanovic, D., Tindbæk, G., Steiniche, T., Schønau, A. and Zucca, T. 

In standard clinical practice the assessment of the nuclear proliferation biomarker Ki-67 is seen as having potential as both a prognostic and predictive marker in breast cancer and other cancers. One of the main obstacles with regards to Ki-67 immunohistochemistry (IHC) assessment in the clinic is the lack of standardization in scoring methods. Traditionally, the scoring has been performed manually using a global method that attempts to derive an average score across all tissue available for assessment. More recent approaches utilize a hot spot method, where the area that appears to be the most active in cell division is identified and used as basis for scoring. However, both methods are often found to be subjective and prone to both intra- and inter-observer variability.

A more objective approach to Ki-67 assessment is the use of digital image analysis (DIA). Here we present a newly developed DIA workflow for automated and objective Ki-67 assessment that implements the hot spot scoring method, and is able to discriminate between invasive and non-invasive tumor components. We compare this DIA workflow to a more conventional DIA workflow using the global assessment method and with no discrimination between invasive and non-invasive tumor components.


Automatic Image Analysis of Sentinel Nodes in Breast Cancer

Holten-Rossing, H., Talman M.T., Jylling, A.M.B. and Vainer, B. 

The aim of this study was to evaluate whether automated image analysis can substitute the manual assessment of metastases in SNs in breast cancer patients.

Breast cancer is one of the most common cancer diseases in women with more than 1.67 million cases each year worldwide (2012). In breast cancer, the sentinel lymph node (SN) pinpoints the first lymph node(s) into which the tumor spreads and it is usually located in the ipsilateral axilla. In cases where ultrasound analysis or a needle biopsy shows no signs of metastatic disease in the axilla, a SN biopsy is performed.

Assessment of metastases in the SN is done in a conventional microscope by manually measuring the size and/or counting the number of tumor cells, to categorize the type of metastases; macro metastases, micro metastases or isolated tumor cells to determine which treatment the breast cancer patient will benefit from.


Qualitopix - Choice of Reference for Automatic Quality Assessment of the Estrogen Receptor

Ottosen, A., Harder, S., Schønau, A., Vyberg, M., Røge, R. and Nielsen, S.

Introduction: Immunohistochemistry is an important tool in patient diagnostics, and therefore external quality assessment (EQA) of immunoassays is essential to obtain optimal and comparable results. EQA is typically performed manually, which is partly subjective causing inconsistent scoring. 

Digital image analysis (DIA) may support EQA by extracting useful information in images to provide objective and standardized results. We propose a solution for automatic quality assessment of semi-quantitative biomarkers, called QUALITOPIX. Our solution can be used to evaluate the precision of an IHC-test in comparison to reference-test by quantification of staining sufficiency. 

For Estrogen Receptor (ER) slides, the fundamental principle is using DIA to quantify the intensity of the stained tumor nuclei into an H-score [1]. The quality is evaluated by calculating the absolute distance from the slide of the individual lab to an established reference, where a large distance indicates staining insufficiency. We present this original (Method 1) and an extended method (Method 2), that amplifies insufficiency caused by weak staining and accounts for background staining. 

We have investigated two different references: One based on a reference from the same tissue block and a joined reference across multiple blocks. Both references have been used in combination with the two methods. 


Detecting Lymph Node Metastases in Breast Cancer Using Deep Learning

Thagaard, J., Medicine & Technology, Technical University of Denmark

We use deep learning to automatically detect and classify breast cancer metastases, providing crucial information for the prognosis and treatment decision of breast cancer. Our method qualified for IEEE International Symposium on Biomedical Imaging 2017 Grand Challenge CAMELYON17 Melbourne, Australia. We were ranked 5thout of 23 qualifying teams of international research groups and commercial teams, with a marginal score difference to the winner of the CAMELYON17 competition.


Image Analysis & Manual Scoring of PD-L1 in Melanomas Using Physical & Virtual Double Staining

Thirstrup, H., Schønau, A., Babi, C., Bukalo, F.S., Petersen, A.V. and Nielsen, S.

Programmed cell death-ligand 1 [PD-L 1] is a transmembrane protein that binds to the inhibitory receptor programmed cell death 1 receptor [PD-1 ], causing a down regulation of the immune responses. PD-L 1 is typically expressed in normal cells as macrophages but frequently also in tumor cells while PD-1 is typically expressed on cytotoxic T-cells and other immune cells. Tumor cells can upregulate PD-L 1 expression and avoid being attacked by the body's immune system, making an interruption of the PD-1/PD-L 1 interaction an attractive method for assisting the immune system in destroying tumor cells.

Assessment of PD-L 1 expression remains under debate and is further complicated by multiple immunohistochemical [IHC] assays, different scoring criteria andthe presence of tumor-infiltrating immune cells. Here we investigate the diagnostic potential of a PD-L 1+SOX10 double IHC assay in melanomas compared toPD-L 1 single IHC assays by manual assessment and digital image analysis [DIA] methods.


HER2-CONNECTTM Pathologist-Assisted Image Analysis for HER2 IHC Interpretation Improves Correlation with HER2 FISH Results

Miller, D.V., Stender, H., Isaac, J., Seaman, J., Hansen, J. and Vyberg, M.

HER2 testing by immunohistochemistry (IHC) is prone to inter-observer variability and subjective interpretation. Pathologist-assisted digital image analysis is recommended in the CAP/ASCO guidelines to improve interpretive consistency. As a quality benchmark, HER2 IHC results should correlate with HER2 fluorescence in-situ hybridization (FISH) in >95% of positive and negative cases. HER2-CONNECT® (Visiopharm, Hoersholm Denmark) is a digital analysis algorithm that scores stained membrane interconnectivity rather than relying solely on staining of individual cells. As such, it can be thought of as a surrogate for the so-called “chickenwire” pattern characteristic of true HER2 positive tumors. In this study we applied HER2-CONNECT® analysis with pathologist quantitative interpretation (HC+PQI) to a set of breast tumors with known HER2 FISH status to assess concordance.

Membrane Connectivity as a Robust Measure for the HER2 IHC Score

Brügmann, A., Eld, M., Lelkaitis, G., Nielsen, S., Grunkin, M., Hansen, J., Foged, N. and Vyberg, M.

The use of digital image analysis for HER2 IHC evaluation is encouraged by ASCO/CAP and reimbursement policies. By individually analyzing the invasive tumor cells and determining the ratio of invasive tumor cells with positive membrane staining, the commercially available classical HER2 IHC algorithms for image analysis are compatible with the scoring principles of the ASCO/CAP guidelines.

Several practical limitations and controversial aspects are associated with individual cell evaluation and counting. Whether by microscopy or digital reading on a monitor, manual scoring of HER2 IHC is therefore typically performed as an overall assessment of the degree of membrane staining within the invasive tumor region of interest (ROI). Counting principles are more readily implemented if based on image analysis, but the controversial aspects of individual cell evaluation remain. It is complicated for algorithms to identify individual invasive tumor cells, since their nucleus and/or cell membrane may not be clearly defined after HER2 IHC staining. Also, it is not obvious if a cell membrane belongs to a particular cell and/or adjacent cells in a dense region of invasive tumor. Furthermore, the classical algorithms and cell ratio quantification require a meticulous outlining of the ROI to avoid contributions from e.g., stromaand ductalcarcinoma in situ.

In this study, we present a novel HER2 IHC algorithm, which uses principles that are not direct translations of the ASCO/CAP and reagent manufacturer’s guidelines. Rather than estimating cell ratio, the present algorithm exploits the unique capacity of computerized image analysis to quantify the characteristic “chicken wire” pattern of HER2 IHC by measuring the connectivity and size distribution of stained membranes. The validation study included 430 ROI’s easily outlined in 86 scanned images of 43 invasive breast carcinoma specimens stained by HercepTestand Pathway HER2/neu, and showed a 92.3% agreement with independent manual reading by 5 experienced assessors.