Radu Rogojanu

Quantitative Image Analysis of Epithelial and Stromal Area in Histological Sections of Colorectal Cancer: An Emerging Diagnostic Tool

In colorectal cancer (CRC), an increase in the stromal (S) area with the reduction of the epithelial (E) parts has been suggested as an indication of tumor progression. Therefore, an automated image method capable of discriminating E and S areas would allow an improved diagnosis. Immunofluorescence staining was performed on paraffin-embedded sections from colorectal tumors (16 samples from patients with liver metastasis and 18 without). Noncancerous tumor adjacent mucosa (n = 5) and normal mucosa (n = 4) were taken as controls. Epithelial cells were identified by an anti-keratin 8 (K8) antibody. Large tissue areas (5–63 mm2/slide) including tumor center, tumor front, and adjacent mucosa were scanned using an automated microscopy system (TissueFAXS). With our newly developed algorithms, we showed that there is more K8-immunoreactive E in the tumor center than in tumor adjacent and normal mucosa. Comparing patients with and without metastasis, the E/S ratio decreased by 20% in the tumor center and by 40% at tumor front in metastatic samples. The reduction of E might be due to a more aggressive phenotype in metastasis patients. The novel software allowed a detailed morphometric analysis of cancer tissue compartments as tools for objective quantitative measurements, reduced analysis time, and increased reproducibility of the data.

Segmentation of cell nuclei within complex configurations in images with colon sections

This paper addresses the development of an automatic segmentation technique for detecting cell nuclei. The technique uses a new approach for segmenting nuclei in images taken from tissues with colon carcinoma. The segmentation problems encountered in these images and solved by the proposed technique are related to the non-uniform illumination on the background, out-of-focus nuclei, the physical structure of cells in the tissue section, the activity status of the cell and the clustered cell nuclei. First, the region growing method is used for accurate background detection. The separation regions between grouped cell nuclei are detected using the cross-correlation method and validated based on their link with the background. Then, the nuclei boundaries are identified by applying the watershed algorithm on the complemented distance transform of the binary image containing the selected separation lines.

Development of Blood and Lymphatic Endothelial Cells in Embryonic and Fetal Human Skin

Blood and lymphatic vessels provide nutrients for the skin and fulfill important homeostatic functions, such as the regulation of immunologic processes. In this study, we investigated the development of blood and lymphatic endothelial cells in prenatal human skin in situ using multicolor immunofluorescence and analyzed angiogenic molecules by protein arrays of lysates and cell culture supernatants. We found that at 8 to 10 weeks of estimated gestational age, CD144(+) vessels predominantly express the venous endothelial cell marker PAL-E, whereas CD144(+)PAL-E(-) vessels compatible with arteries only appear at the end of the first trimester. Lymphatic progenitor cells at 8 weeks of estimated gestational age express CD31, CD144, Prox1, and temporary PAL-E. At that developmental stage not all lymphatic progenitor cells express podoplanin or Lyve-1, which are acquired with advancing gestational age in a stepwise fashion. Already in second-trimester human skin, the phenotype of blood and lymphatic vessels roughly resembles the one in adult skin. The expression pattern of angiogenic molecules in lysates and cell culture supernatants of prenatal skin did not reveal the expected bent to proangiogenic molecules, indicating a complex regulation of angiogenesis during ontogeny. In summary, this study provides enticing new insights into the development and phenotypic characteristics of the vascular system in human prenatal skin.

Automated REcognition of tissue-associated erythrocytes (ARETE)-a new tool in tissue cytometry.

Automated microscopic image analysis of immunofluorescence‐stained targets on tissue sections is challenged by autofluorescent elements such as erythrocytes, which might interfere with target segmentation and quantification. Therefore, we developed an automated system (Automated REcognition of Tissue‐associated Erythrocytes; ARETE) for in silico exclusion of erythrocytes. To detect erythrocytes in transmission images, a cascade of boosted decision trees of Haar‐like features was trained on 8,640/4,000 areas (15 × 15 pixels) with/without erythrocytes from images of placental sections (4 µm). Ground truth data were generated on 28 transmission images. At least two human experts labelled the area covered by erythrocytes. For validation, output masks of human experts and ARETE were compared pixel‐wise against a mask obtained from majority voting of human experts. F1 score, specificity, and Cohens κ coefficients were calculated. To study the influence of erythrocyte‐derived autofluorescence, we investigated the expression levels of a protein (receptor for advanced glycated end products; RAGE) in placenta and number of Ki‐67‐positive/cytokeratin 8‐positive epithelial cells in colon sections. ARETE exhibited high sensitivity (99.87%) and specificity (99.81%) on a training‐subset and processed transmission images (1,392 × 1,024 pixels) within 4 sec. ARETE and human experts F1‐scores were 0.55 versus 0.76, specificities 0.85 versus 0.92 and Cohens κ coefficients 0.41 versus 0.68. A ranking of Cohens κ coefficient by the scale of Fleiss certified “good agreement” between ARETE and the human experts. Applying ARETE, we demonstrated 4–14% false‐positive RAGE‐expression in placenta, and 18% falsely detected proliferative epithelial cells in colon, caused by erythrocyte‐autofluorescence. ARETE is a fast system for in silico reduction of erythrocytes, which improves automated image analysis in research and diagnostic pathology.

Mast Cells Are Abundant in Primary Cutaneous T-Cell Lymphomas: Results from a Computer-Aided Quantitative Immunohistological Study

Background Mast cells (MC) are bone marrow derived haematopoetic cells playing a crucial role not only in immune response but also in the tumor microenvironment with protumorigenic and antitumorigenic functions. The role of MC in primary cutaneous T-cell lymphomas (CTCL), a heterogeneous group of non-Hodgkin lymphomas with initial presentation in the skin, is largely unknown. Objective To gain more accurate information about presence, number, distribution and state of activation (degranulated vs. non-degranulated) of MC in CTCL variants and clinical stages. Materials and Methods We established a novel computer-aided tissue analysis method on digitized skin sections. Immunohistochemistry with an anti-MC tryptase antibody was performed on 34 biopsies of different CTCL subtypes and on control skin samples. An algorithm for the automatic detection of the epidermis and of cell density based CTCL areas was developed. Cells were stratified as being within the CTCL infiltrate, in P1 (a surrounding area 0–30 μm away from CTCL), or in P2 (30–60 μm away from CTCL) area. Results We found high MC counts within CTCL infiltrates and P1 and a decreased MC number in the surrounding dermis P2. Higher MC numbers were found in MF compared to all other CTCL subgroups. Regarding different stages of MF, we found significantly higher mast cell counts in stages IA and IB than in stages IIA and IIB. Regarding MC densities, we found a higher density of MC in MF compared to all other CTCL subgroups. More MC were non-degranulated than degranulated. Conclusion Here for the first time an automated method for MC analysis on tissue sections and its use in CTCL is described. Eliminating error from investigator bias, the method allows for precise cell identification and counting. Our results provide new insights on MC distribution in CTCL reappraising their role in the pathophysiology of CTCL.

Epithelial area detection in cytokeratin microscopic images using MSER segmentation in an anisotropic pyramid

The objective of semantic segmentation in microscopic images is to extract the cellular, nuclear or tissue components. This problem is challenging due to the large variations of features of these components (size, shape, orientation or texture). In this paper we present an automatic technique to robustly delimit the epithelial area (crypts) in microscopic images taken from colon tissues sections marked with cytokeratin-8. The epithelial area is highlighted using the anisotropic diffusion pyramid and segmented using MSER+. The crypts separation and lumen detection is performed by imposing topological constraints about the epithelial layer distribution within the tissue and the round-like shape of the crypt. The evaluation of the proposed method is made by comparing the results with ground-truth segmentations.

Abstract 1091: A new image processing approach for functional context-based analysis of tryptase positive mast cells in cutaneous T-cell lymphomas

Quantification data on mast cells is scarce in primary cutaneous T-cell lymphomas (CTCL), a heterogeneous group of non-Hodgkin lymphomas with initial presentation in the skin. The aim of the study was to develop a new whole-slide in-silico analysis method for immunohistochemistry stained skin sections. The design of the analysis method was focused on results about structural context of the cell location distribution and cell morphometry linked to mast cell activation states. Immunohistochemistry staining with a monoclonal anti-mast cell tryptase antibody was performed on paraffin embedded biopsies from patients with different CTCL subtypes and controls (normal skin, inflammatory skin diseases),while hematoxylin was used for counter staining. Slides were scanned with a TissueFAXS 200 Cytometer (TissueGnostics GmbH, Vienna). Whole-slide digital images were analyzed using StrataQuest 5.0 Advanced (TissueGnostics, Vienna). Hematoxylin+ cell nuclei and tryptase+ cells (TRYP + ) objects were automatically detected using the available analysis modules with adaptive segmentation of color-deconvoluted images. Mast cells were first selected as brown-positive cells in a peri-nuclear compartment, then further stratified depending on the spatial distribution of brown-positive dots inside the cellular compartment. In an additional analysis layer, an algorithm pipeline for the identification of the epidermis / dermis area was assembled. Density-based CTCL areas were segmented in a third independent pipeline. Inside dermis, two additional CTCL-proximity masks were defined at 0-30μm away from CTCL (P1 mask), as well as in the 30-60μm interval (P2 mask). Typically, there were more mast cells inside CTCL areas than in P1, which in turn had less than the P2 mask. This distribution was especially noticed in Mycosis fungoides(MF) IIB, where CTCL had 2-3 times more mast cells than P1 mask had. However, this trend was not consistent in all types of lymphoma. Interestingly, MF IIA had a different distribution showing more mast cells in P1 than in CTCL areas. The new algorithms provided a rich data insight on functional context-based characteristics and propose new surrogate markers for further characterization of CTCL. The analysis time needed for statistically relevant large whole-slide tissue sections was drastically reduced when using the computer-aided approach when compared against full manual annotation and counting. Moreover, distance assessment and morphometry parameter quantification for each cell are only feasible in in-silico approaches. Note: This abstract was not presented at the meeting. Citation Format: Radu Rogojanu, Johanna Eder, Waltraud Jerey, Gabriele Klosner, Verene Paulitschke, Isabella Ellinger, Theresia Thalhammer, Dan Kolmer, Franz Trautinger. A new image processing approach for functional context-based analysis of tryptase positive mast cells in cutaneous T-cell lymphomas. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 1091. doi:10.1158/1538-7445.AM2015-1091

Abstract 1662: Towards automated analysis of prostate inflammatory state: Assessing density and distance of infiltrating T-cells (CD3+) to glandular structures in prostate biopsies by a new software

Proceedings: AACR Annual Meeting 2014; April 5-9, 2014; San Diego, CA In the prostate, normal glands, glandular hyperplasia, and adenocarcinoma are often accompanied by inflammation. Infiltrating inflammatory cells may affect glandular cells and either promote their proliferation or, alternatively act in tumor suppression. To better understand the function of the inflammatory cells in tumor development, type and number of inflammatory cells and their proximity to glandular structures have to be analyzed in situ and correlated with disease state. To replace the time-consuming, error-prone human evaluation of stained tissue sections, we aimed at developing an automated method capable of detecting inflammatory and prostate glandular cells and of measuring the distance between these structures. Methods: Formaldehyde-fixed, paraffin-embedded prostate cancer biopsies (n=53) were stained with DAB-conjugated anti-CD3-antibody and hematoxylin (H). Large tissue areas were digitized with TissueFAXS 4.0 using a 20x objective and the virtual stitched images were used to develop a StrataQuest 5.0 analysis profile. Manual annotations were used for validation. Results: Color unmixing generated the virtual channels for H and DAB optical densities. Cells were then identified using nuclear segmentation on the H channel. For CD3+ cell counts, each cell was quantified for its DAB signal in a peri-nuclear mask, for which cutoff was set interactively. The epithelial cell (EC) nuclei were selected by gating for nuclear compactness and area. EC density was computed using Parzen windows and thresholded for obtaining the epithelial mask (EM). Whole tissue area excluding lumen was computed based on color. Stroma was then extracted as remaining area. In-silico Distance Transformation (DT) from EM was calculated as a monochrome channel aligned with the original virtual slide. Thus, by using the average of the DT inside each nuclear mask, the distance of each cell to the closest EM was computed. Three peri-glandular proximity areas (0-25um, 25-50um and 50-75um) were used for CD3+ population distribution assessment. Validation of the method revealed an F-score of 0.87 for EM detection and a p value of <0.001 for automated versus manual CD3+counting. The percentage of EM versus total area ranged from 18% to 50%. Between 54% to 89% of all CD3+ cells were found in the closest proximity of EM (in the 0-25um mask). A positive correlation between EM area and T-lymphocyte density in the closest vicinity mask was observed (p < 0.001). Conclusion: The described method enables reliable automated detection of prostate tissue compartments (EC, stroma), CD3+ T-cells counting as well as measurement of their distance from the glandular epithelial cells. Such automated measurements can lead to reliable and rapid measurements of cell location distribution in normal versus tumor prostate tissues, opening new routes in diagnosis and prognosis. Citation Format: Radu Rogojanu, Bogdan Boghiu, Georg Schaefer, Theresia Thalhammer, Georg Steiner, Rupert Ecker, Bettina Schlick, Thomas Szekeres, Isabella Ellinger. Towards automated analysis of prostate inflammatory state: Assessing density and distance of infiltrating T-cells (CD3+) to glandular structures in prostate biopsies by a new software. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 1662. doi:10.1158/1538-7445.AM2014-1662