Morphometry, image cytometry and tissue image analysis experienced a surge of enthusiasm in the 1980’s and 1990’s; which subsequently declined as pathologists realised the technical difficulties of introducing quantitative techniques into routine practice. This “has been” technology, was quickly superseded by the next wave of enthusiasm surrounding molecular technologies which promised to remove the need for conventional histopathology altogether. Now in the 21st century, “digital pathology” as it now known, once again finds itself in the spotlight as a technique that underpins tissue biomarker discovery and as a practical tool to enhance diagnostic practice in histopathology. What has brought about this revolution has been the development of whole slide scanners and virtual microscopy.
We have been developing a range of new tools for the high throughput quantitative evaluation of whole slide images in biomarker discovery. In non-small cell lung cancer (NSCLC) we have developed a range of machine vision algorithms which can facilitate the analysis of biomarkers, BAX, BAK, BRCA1 and NOX in lung cancer tissue microarrays (TMAs). These include new methods to de-array virtual TMA images, to extract regions showing biomarker expression and to quantify immunohistochemical (IHC) expression. Results show that machine vision on digital slides can detect subvisual changes not visible to the naked eye and identify valuable IHC biomarkers that would otherwise have been overlooked. We have also constructed a high performance computing (HPC) platform for the high throughput analysis of tissue biomarkers on TMAs. This parallelisation of tissue core analysis can significantly speed up TMA analysis, making it a truly high throughput platform for biomarker discovery. This can be effectively applied using the principles of cloud computing for virtual microscopy where digital slide hosting, viewing and analysis can be carried out remotely on a centralised HPC platform.
Finally, we have developed new imaging tool for distinguishing squamous and nonsquamous NSCLC virtual slides as a means to support the selection of therapy with pemetrexed and cisplatin. This practical combination of whole slide examination, virtual microscopy and associated analytical tools within a single portal, supports the use of virtual microscopy in routine diagnostic practice and overcomes some of the hurdles which prevented its adoption previously. These analytical tools facilitate decision making and include annotation, practical remote consultation, pre-processing, feature extraction and image search. We call this “augmented visualisation” in pathology. The adoption of digital pathology in routine practice will however, also require a closer integration of scanning technology with Laboratory information Management Systems (LIMS) and a careful consideration of pathology workflow, regulatory (FDA) and safety issues and DICOM standardisation. The next few years will see rapid and interesting developments in the role of digital pathology, with a critical tipping point where tissue and cell imaging will become an essential tool in tissue-based research and in diagnostic practice.

At present, decision making for breast cancer treatment in the clinical setting is based on clinical data, histomorphological features of the tumor tissue, and on cancer biomarkers such as steroid hormone receptor and Her-2 status, determined by immunohistochemistry or FISH. Although various therapy options were introduced into the clinic in the last decades varying response of individual patients to certain types of therapy and therapy resistance is still a challenge in breast cancer care. Therefore, tailored therapy should be an option by integrating cancer biomarkers to define patients at risk and to predict the course of the disease and/or response to cancer therapy. Recently, candidate marker approaches and genome wide/transcriptomic/epigenetic screening of breast cancer tissues resulted in new promising biomarker panels allowing breast cancer prognosis, prediction of therapy response, and monitoring of therapy efficacy; biomarkers which are now subject of validation in prospective clinical trials employing histological and molecular technologies.

Aim of this research is to develop a measurement modality capable of assessing the risk of malignant progression of oral premalignancies, for which conventional histopathology was shown to be insufficiently accurate. Hypothesis: combined assessment of DNA ploidy and expression of relevant proteins (e.g. Ki67, yH2AX) at the single cell level allows identification of cells at risk for malignant transformation. For such analysis, multiplexing of nuclear signals is required, necessitating the use of modern fluorescent labeling techniques and corresponding imaging devices. We combined traditional fluorochromes, QDots and stoichiometric nuclear staining to maximize the number of simultaneous signals in a single specimen. Next, accurate automated identification of individual nuclear profiles in tissue images was performed, using different image analysis algorithms consecutively (Hough transform, deformable models). The presented combination of techniques enables sophisticated analysis of the biochemical makeup of cells, enabling refinement of histopathological categories to more closely match prognosis.

In the near future, whole slide imaging will prove essential for creating systems biological models of tissues as it allows the fast and efficient capturing of tissue structures and spatial expression patterns. For multiple research projects our group develops new approaches driving the data acquisition and modeling during the self renewal of human skin (skin homeostasis). So we use whole-slide imaging for capturing the dynamic, self-organizing process of wound healing. We identified for the first time the spatial reorganization of cellular stream inside the skin during wound closure which is controlled by integrating migration, proliferation and cell differentiation in space and time. The analysis has been performed on controlled HE sections of a novel in vitro organotypic wound healing model we developed. Moreover, complementary image processing approaches are presented where large-scale brightfield tissue microarrays and fluorescent whole tissue sections of 3D in vitro cultures are used to create spatial protein expression data forming a novel quantitative basis for computational models of skin tissue homeostasis.

In the recent years the development of bright field virtual microscopy and novel fluorescence microscopy imaging techniques has enabled researchers to investigate biological features of cells and tissues with high spatial and functional resolution. In this context, the ability to make these data accessible to quantitative computational analyses not only facilitates statistical data interpretation but also serves as a basis for data integration in comprehensive “systems biology” models. This talk presents our recent developments in automated bright field and fluorescence microscopy image analysis and model generation ranging from high-throughput immunohistochemistry analysis of virtual microscopy data of influenza infection, cellular interaction analysis in two-photon microscopy data to building realistic morphological models of cells and tissues. We will also discuss the traditionally neglected but increasingly important issue of method validation of automated image analysis in virtual microscopy.

I will discuss our attempts to understand why some aspects of receptor-mediated apoptosis in tumor cells are highly variable from one cell to the next whereas others are remarkably constant. Variability in the timing and probability of death in a clonal cell population, or among recently born sister cells, has its origin in cell-to-cell differences in the concentrations of key regulatory proteins.This in turn, arises from the stochastic nature of gene expression, translation and protein partitioning at mitosis. In contrast, constancy in the time required to die, or in the efficiency of caspase activation, involves a complex and robust regulatory circuit that controls mitochondrial outer membrane permeablization. We observe variability in the responses of cells to virtually all ligands and drugs, and such variability is likely to partly explain the "fractional kill" observed for many chemotherapeutic agents in vivo. Moreover, the existence of long-tailed response distributions is a probable explanation for many phenomena currently ascribed to "tumor stem cells."

An important challenge in the post-genomic era is to identify subcellular location on a proteome-wide basis, and learn how protein locations change as a function of cell type or cell state (such as during disease development). A major source of information for this task is imaging of tagged proteins using light microscopy. We have previously developed automated systems to interpret subcellular patterns in images of single cells obtained by fluorescence microscopy and demonstrated that they can perform as well or better than visual inspection. We have more recently applied them to the large collection of images in the Human Protein Atlas. The results indicate the feasibility of recognizing subcellular patterns in images from tissue microarrays stained by immunohistochemical methods. This permits potential biomarkers to be identified not only by changes in expression level but also by changes in subcellular distribution.

Opportunities to offer cancer patients "tailored" individual treatment depends on the measurable bio-markers which is present in the cancer tumor. Genetic changes are e.g. analyzed today with the molecular pathological fluorescence in situ hybridization (FISH) technique or immunohistochemistry (IHC) on slides on which the relevant cancer tissue from each patient is located. It is possible to assemble small cancer tissue cores from e.g. 40 patients using a tissue microarray (TMA) machine onto a single slide. This simplifies the processing of patient data and enables a considerable reduction in working hours and materials by introducing digital image and image analysis equipment.

Breast cancer is a highly heterogeneous disease with contrasting clinical behaviors. Some tumors are indolent and do not seem to have the capacity to form metastases whereas other are extremely aggressive with generalized growth early in the progression. Tumours can be subdivided into relevant subgroups according to epithelial as well stromal properties. We have used manual scoring and selected automated analyses of large collections of breast cancer samples arranged in tissue microarrays (TMAs) to identify key processes in breast cancer linked to clinicopathological data. Analyses of cyclin D1, collagen and activation of cell signaling pathways will be presented and challenges and opportunities regarding analyses of breast cancer using automated analyses platforms will be discussed.

Segmentation and classification are important intermediate tasks in bio-medical imaging, but current solutions are mostly problem-specific, or only semi-automated. We propose a supervised learning framework that is able to tackle multi-object segmentation and multi-class discrimination in a unified way. No prior expertise in image processing is required since user interaction is restricted to intuitive brush strokes. The trained algorithm can then be applied to thousands of images with no further interaction. The program is limited to the segmentation of objects that can be discriminated based on local cues such as color or texture; but within this setting, first-shot results on standard problems are often of surprising quality.

Motivation:
In histological image analysis a wide range of different image analysis frameworks has been used over the years. These platforms and technologies for analyzing images use, e.g., different languages and concepts and each has specific strengths and drawbacks. This becomes even more complex when technologies for semantic analysis are combined with native image analysis approaches. Finding the right platform which fits predetermined needs can be a time-consuming task and carries the risk of wrong decisions. It is also eminently challenging to group researchers with different skills in different platform in one project. Thus a lot of time and effort is wasted finding the right way to work together while it would be more worth using anyone’s specific strengths.
Materials and Methods:
The developed platform offers a distributed approach to combine different image analysis platforms in one common system. Thus, the development of image analysis algorithms can be conducted over different systems using their specific strengths. As an example we integrated three completely different frameworks. ImageJ [1], a Java-based, open source image analysis platform. ITK [2] a segmentation and analysis framework written in C. Futhermore, we integrated Definiens Developer XD [3], a commercial UI based system for developing segmentation and classification algorithms. To accomplish this, the framework uses a web-based XML protocol which allows platform independent communication between the components of the system. Using a single entry point every client in the system can access all provided algorithm without detailed knowledge about their implementation. Thus, the system allows cross-developing on different systems and frameworks.
Results:
The developed framework provides an easy and coherent access to different image analysis platforms which can be used to gain advantage from any specific platform. Due to the distributed nature of the system tasks can be performed at different locations. With the provided approach it is even possible to exchange an existing implementation of an algorithm with its complement from another platform. New frameworks can also be added easily to the system to use additional image analysis algorithms.
Conclusions:
We developed a platform that combines different basic image analysis algorithms. It should be noted that this platform holds the potential to be used in even more complex scenarios. Currently, the platform will be extended to provide semantic support for image analysis.

The storage of image data in picture archiving and communication systems (PACS) is required for all data in routine pathology. Common clinical PACSs use DICOM standard, which allows storage and retrieval of images as well as case related information in a common open format. For efficient use of WSI, adjustments of the DICOM standard are necessary. Although subtle attempts have been made by the DICOM committee to integrate WSI using fragmentation of the huge image data, certainly profound modifications of the DICOM standard are necessary to meet the requirements of WSI.
The use of WSI in routine pathology requires diagnostic evaluation with special regard to image quality, which certainly is also important for image analysis. DICOM-compatible lossy compression is necessary for data reduction - but where is the limit?

Introduction: Digital pathology is a young but very fast developing process which originates from the possibility to digitalize complete histological slides with high quality and resolution. In the meantime various slide scanners are available. They are based on different scanning principles, different sensor properties and different loading and handling of the slides. We tried to compare different scanning devices with respect to real demands of digital pathology within the 1th European Scanner Contest.
Material & Methods: Eight scanners of six vendors were tested in different disciplines: mass scan - a set of 35 H&E slides, operational material as well as biopsies, scanning without manual interaction during the scanning process, quality scan - set of 10 slides with different staining, cytology scan - 12 Pap. stained cytological slides, IHC scan – 12 Her2 slides. A team of referees took care about fair scanning conditions. Scanning and preparation time were measured. The comparison of images quality took place as a two step process: manually by pathologists during and after the 94th Annual Meeting of German Society of Pathology and automatically by image processing software.
Results: Eight scanners of six vendors took part in the contest. The scanning time for the mass scan varied between is 1 h 20 min and 14 h The quality of the scans differed in a wide range. Several slides have been scanned from all vendors with acceptable quality (e.g. nervous cross sections). Other slides (e.g. Her2) showed big differences. There are also differences in the evaluation of image quality between pathologists, which are very interesting.
Discussion: The 1st European Scanner Contest offered firstly the possibility to compare different scanneing devices due to scanning time and various aspects of image quality on identical slides. A very important effect of the contest is the begin of a discussion about concepts for the quality evaluation of whole slide images (WSI). This is extremely relevant for quantification purposes (e.g. IHC). The results are relevant for the future development of scanning devices and we have been asked from several sides to implement a yearly contest.

The virtual microscope as a digital equivalent of the conventional device is still required to close some gaps to gain full acceptance by the clinical user audience. This can be considered to be one of the major building blocks of a digital pathology solution. Many areas – especially when it comes to image processing – will add important momentum in this field but still many issues in the slide processing have not been solved yet.
This applies to the entire process from the first pixel that is captured until the diagnosis is made. Although the processes in conventional microscopy are rather simple, their digital equivalents seem to be complex.
We will focus on possible improvements in image compression, streaming and viewing based on JPEG2000, involving techniques like multi-component transforms and discuss first approaches and the implications to improve the virtual microscope.

Introduction:
Virtual Microscopy (VM) is well-established in education in research, but still far from daily clinical use. One reason is the lack of LIMS, DICOM-PACS and clinical workflow integration. We want to discuss possibilities for a fully integrated pathology/VM system and show examples, where such a scenario can facilitate the daily pathologist’s work.
Methods:
We evaluated the workflow at the Institute of Pathology at the Charité Berlin and identified processes with automation capability. We then designed a potential new workflow and compared both.
Results:
An integrated VM solution should cover: Digital laboratory order communication, system supervised auto-stainers and other devices, direct slide scanner loading, DICOM-PACS, automated case assignment and diagnosis on the WSI. The information system itself would need interfaces to clinical PACS, typing pool, image analysis and tele-consultation.
Discussion:
There are still challenges on the way to a fully integrated system. But at the end, pathologists will benefit from higher accuracy, more functionality and facilitated processes.
Conclusion:
The last years brought some substantial technical improvements in VM. Now it is time to stick all parts together. Only a fully integrated solution will really save the pathologist time and effort.

Introduction:
Image analysis on Whole Slide Images (WSI) has a growing importance on virtual microscopy (VM) applications in research and daily clinical use. Because image data and file formats in VM are very specialized and unique, the use of common image analysis software hardly fits the requirements and is not efficient. We want to introduce an integrated and specialised toolkit for whole slide imaging that accelerates the whole process of building new and complex image analysis algorithms dramatically.
Methods:
HistoBase+ is a research project at the Charité that deals with cell labelling and segmentation in breast cancer slides. This project aims to increase objectivity in quantification of immuno-stainings and comparison of image analysis algorithms. We evaluated the requirements coming with whole slide imaging and designed an integrated image analysis platform.
Results:
We implemented a slide scanner independent image analysis framework that performs segmentation, classification, automation and parallelisation. Further dedicated plugins for the virtual microscopy application “VM Slide Explorer” were implemented.
Discussion:
We showed a solution for integrated customisable image analysis on WSI. Now certain use cases and algorithms need to be identified.
Conclusion:
Image analysis in VM can increase the automation in tissue research and clinical use and thus has the potential to accelerate and facilitate the daily work. Furthermore the analysis is reproducible and may act as reference for manual evaluation.

Introduction: Serendipia, the telepathology network of Castilla-La Mancha, Spain, was born thanks to the active collaboration between pathologists and software engineers.
Methods: The selected architecture was based in emerging IT standards in Medicine and in digital imaging, with the main objective of a complete integration between systems.
Results: Gross station imaging systems, microscope digital cameras and digital slide scanners were successfully integrated with enterprise PACS, mainly using HL7, DICOM and SNOMED CT, following IHE Pathology Workflow profile. This imaging and automated image analysis multiplatform (Olympus .Slide, Aperio Scancope) network uses Aurora viewer to guarantee web access for all image formats, integrated with pathology information systems, second opinion, discussion forum, and teaching library modules.
Discussion: High quality images in this platform improve quality in diagnosis, treatment and prognosis evaluation. Our next steps include implementation of recently approved DICOM Supplement 145, and integration of fluorescence digital slides and multispectral images.

Virtual slide technology has been developed to help image analysis on whole tissue. In this talk, a histopathological study, based on adipose tissue will be presented. Considered as a passive storage tissue for a long time, adipose tissue is now recognized as a real organ, which plays a role in many metabolic and inflammatory dysfunctions. Many epidemiological studies have confirmed that an accumulation of visceral adipose tissue could lead to a major risk of obesity complications. Thus, a decrease of this fat must be a therapeutic goal to achieve.
Classic histopathology techniques on a rodent model, combined to image analysis were used to determine morphological variations of adipose tissue, function of localization, evolution of inflammation and age.
Results show a variation of adipocyte size, depending on localization of these cells on one side and on age and weight on the other side. Inflammation increase has also been shown. Thanks to a fast scanning of slides and automatization of image analysis, this robust rodent model can now be used to test effects and efficiency of future drugs against diabetes and obesity.

Pathology is a central element in clinical medicine and medical students training. Despite this fact the exposure of medical students to pathology training has significantly decreased over the years with negative consequences for specialty reputation and choice in medical students. Over the last years we have used virtual microscopy to enrich and improve training in histopathology by various means:
a. annotated histopathology course for unlimited self-training and exam preparation in general pathology
b. associated and annotated virtual histopathology courses for clinical courses
c. case preparation for the student course 'clinico-pathological conference'
d. course examination
e. free training cases
Students of the Medical faculty have been involved in this project and structure and experiences will be presented.

Quantitative estimations of the immunohistological expression of biomarkers in tumor tissue are required with increasing frequency today, both for scientific analysis of tumor tissue and for diagnostic histopathology. One of the major problems posed in this analysis is the variable content of tumor and stromal tissue in a given histological section or in a TMA core. Most tumor markers must be evaluated on tumor tissue only (e.g. hormone receptors or proliferation rate), disregarding the tumor stroma. Therefore, a reliable concept for tumor stroma segmentation is needed as a prerequisite for the analysis of tumor biomarkers. Manual segmentation not only is very time consuming, but also applicable only to small tumor areas, and not to whole tissue sections in virtual microscopy. We have evaluated different methods including genetic algorithms for the evaluation of nuclear, cytoplasmatic, and membranous tumor biomarkers. These algorithms can be considered quite robust, but generally rely on the presence of morphological similarity of tumor tissue. The biggest problem to automatic tumor stroma segmentation is posed by the  wide variety of morphologically different tumor types in human tumors, especially in breast cancers. Therefore, extensive machine learning is required for the correct recognition of tumor types, and it is suggested that more refined algorithms are developed for this purpose.

Matrix-assisted laser desorption/ionization (MALDI) imaging mass spectrometry (IMS) is a powerful tool for investigating the distribution of proteins and small molecules within biological systems through the in situ analysis of tissue sections. MALDI-IMS can determine the distribution of hundreds of unknown compounds in a single measurement and enables the acquisition of cellular expression profiles while maintaining the cellular and molecular integrity. In recent years, a great many advances in the practice of imaging mass spectrometry have taken place, making the technique more sensitive, robust, and ultimately useful. In this overview, we focus on the current state of the art of MALDI-IMS, describe basic technological developments for MALDI-IMS of animal and human tissues, and discuss some recent applications in basic research and in clinical settings.

Quantitative fluorescence and chromagenic slide-based biomarker assays play an important role in understanding the pharmacodynamic response of target inhibition in drug discovery programs. We have developed and integrated imaging technologies to enable the quantitative analysis of slide-based assays in an automated and high throughput manner. Brightfield and fluorescent whole slide images are acquired with the Aperio ScanscopeXT and ScanscopeFL scanners. Multispectral imaging is done with the Cri Nuance system, and we have developed customized systems designed to meet the needs for novel assays. Image analysis software includes Aperio, Metamorph and Definiens. This presentation will provide a description of these technologies and how the technologies are used to drive lead optimization efforts in preclinical models and as part of a comprehensive biomarker plan to understand drug effects in primary and surrogate tissue in ongoing clinical studies.

Tissue based biomarkers can be used to aid in the diagnosis of certain diseases and to predict response to a certain treatment. A multitude of such biomarkers have been put forward in infections, autoimmune inflammatory disorders and neoplastic diseases. Since in some of these diseases expression of a specific protein and/or detection of a certain mutation is the sole factor deciding how the patient is treated, a reliable and reproducible evaluation of such biomarkers is of utmost importance. Standardized tissue imaging and analysis may circumvent the subjectivity of human tissue evaluation in this context. This talk will give an overview on biomarkers currently in clinical use and under clinical development and focuses on the question for which of these markers tissue image analysis would be desirable. In addition, it will be discussed which algorithms are most urgently needed and how image analysis might help to develop novel diagnostic, prognostic and predictive tests on the way to a more standardized patient treatment.

The most obvious application of RNAi screening, direct loss-of-function (LOF) screening, involves identifying and functionally characterizing genes of interest on the basis of their LOF phenotypes. RNAi Screening or direct loss of function screening might use biomarkers in assays but might also be applied for the identification of novel biomarkers.

Virtual Microscopy in combination with dedicated imaging analysis is a powerful tool in the analysis of the local tumor microenvironment. Distribution and quantification of immune cells in human cancers on complete tissue sections reveals a distinct heterogeneity, that has not been appreciated so far. These "immunological tumor maps" allow a refined analysis of spatial patterns. This quantitative approach however can also be advanced through the combination with other technologies, i.e. multiplex bead based protein measurements. This powerfull approach opens the door to understand immune cell composition in conjunction with cytokines and chemokines. In colorectal cancer, localization and densities of lymphocytes and a broad panel of cytokines/chemokines was quantified in paired serum, tumor and adjacent mucosa samples in order to correlate the presence of certain chemokines to immune cell infiltration. Characteristic cytokine/chemokine patterns between different compartments could be identified, indicating a distinct regulation of immune cell subsets in the tumor microenvironment. These results highlight the feasibility of this novel approach, combining Virtual Microscopy and multiplex bead technology.