Definiens Developer Research Articles

Image Processing Techniques based Feature Extraction for Insect Damage Areas

Monitoring of forests is important for the diagnosis of insect damage to vegetation. Detection and monitoring of damaged areas facilitates the control of pests for practitioners. For this purpose, Unmanned Aerial Vehicles (UAVs) have been recently used to detect damaged areas. In order to distinguish damage areas from healthy areas on UAV images, it is necessary to extract the feature parameters of the images. Therefore, feature extraction is an important step in Computer Aided Diagnosis of insect damage monitored with UAV images. By reducing the size of the UAV image data, it is possible to distinguish between damaged and healthy areas from the extracted features. The accuracy of the classification algorithm depends on the segmentation method and the extracted features. The Grey-Level Co-occurrence Matrix (GLCM) characterizes areas texture based on the number of pixel pairs with specific intensity values arranged in specific spatial relationships. In this paper, texture characteristics of insect damage areas were extracted from UAV images using with GLCM. The 3000*4000 resolution UAV images containing damaged and healthy larch trees were analyzed using Definiens Developer (e-Cognition software) for multiresolution segmentation to detect the damaged areas. In this analysis, scale parameters were applied as 500, shape 0.1, color 0.9 and compactness 0.5. As a result of segmentation, GLCM homogeneity, GLCM contrast and GLCM entropy texture parameters were calculated for each segment. When calculating the texturing parameters, neighborhoods in different angular directions (0,45,90,135) are taken into account. As a result of the calculations made by considering all directions, it was found that GLCM homogeneity values ranged between 0.08 - 0.2, GLCM contrast values ranged between 82.86 - 303.58 and GLCM entropy values ranged between 7.81 - 8.51. On the other hand, GLCM homogeneity for healthy areas varies between 0.05 - 0.08, GLCM contrast between 441.70 - 888.80 and GLCM entropy between 8.93 - 9.40. The study demonstrated that GLCM technique can be a reliable method to detection of insect damage areas from UAV imagery.

Digital quantitative tissue image analysis of hypoxia in resected pancreatic ductal adenocarcinomas.

BackgroundTumor hypoxia is theorized to contribute to the aggressive biology of pancreatic ductal adenocarcinoma (PDAC). We previously reported that hypoxia correlated with rapid tumor growth and metastasis in patient-derived xenografts. Anticipating a prognostic relevance of hypoxia in patient tumors, we developed protocols for automated semi-quantitative image analysis to provide an objective, observer-independent measure of hypoxia. We further validated this method which can reproducibly estimate pimonidazole-detectable hypoxia in a high-through put manner.MethodsWe studied the performance of three automated image analysis platforms in scoring pimonidazole-detectable hypoxia in resected PDAC (n = 10) in a cohort of patients enrolled in PIMO-PANC. Multiple stained tumor sections were analyzed on three independent image-analysis platforms, Aperio Genie (AG), Definiens Tissue Studio (TS), and Definiens Developer (DD), which comprised of a customized rule set.ResultsThe output from Aperio Genie (AG) had good concordance with manual scoring, but the workflow was resource-intensive and not suited for high-throughput analysis. TS analysis had high levels of variability related to misclassification of cells class, while the customized rule set of DD had a high level of reliability with an intraclass coefficient of more than 85%.DiscussionThis work demonstrates the feasibility of developing a robust, high-performance pipeline for an automated, quantitative scoring of pimonidazole-detectable hypoxia in patient tumors.

Abstract 5714: Characterization of SWI/SNF complex gene mutations, protein expression and tumor immune microenvironment in cholangiocarcinoma

Abstract Introduction: Cholangiocarcinoma (CCA) is a rare tumor accounting for less than 2% of all human malignancies. Mutations in the mammalian Switch/Sucrose Non-Fermentable (SWI/SNF) chromatin remodeling complex have been identified in approximately 20% of all human cancers and more than 40% of CCA. If the role of SWI/SNF on the tumor biology and microenvironment is being uncovered in other tumor types, it remains vastly unknown in CCA. Here, we wanted to investigate, using clinical samples, the association between SWI/SNF defects, the molecular landscape and tumor immune microenvironment in CCA. Material and Methods: Mutation profiling of 103 patients with CCA from Gustave Roussy (GR) was assessed with FoundationOne® CDx 324-gene NGS panel. Formalin-fixed paraffin-embedded tumor tissues from 11 patients with CCA were analyzed for SWI/SNF (PBRM1, ARID1A, SMARCB1 and SMARCA4) and Polycomb-DUB (BAP1) subunits, as well as lymphocytic markers (CD4 and CD8). Stainings were performed using a VENTANA BenchMark ULTRA. Absolute count of positive cells per mm2 was determined by digital image analysis with the Definiens Developer XD™ by selecting at least 5 independent tumoral zones. Pairwise comparisons were performed using the Mann-Whitney test. Results: Among the patients for whom NGS data was available, the most frequently altered genes were TP53 (27%), CDKN2A (17%), KRAS (16%), ARID1A (12%), FGFR2 (12%) and PBRM1 (10%). Subunits of the SWI/SNF complex and BAP1 were mutated in 25% and 10% of cases, respectively. Tumor samples with ARID1A mutations (n=5) showed strongly decreased ARID1A expression compared to WT tumor samples (n=2) (mean = 308.03 positive nuclei (p.n)/mm2 versus 6096.82 p.n/mm2 respectively; P<0.0001) but unchanged PBRM1, SMARCB1 and SMARCA4 expression. PBRM1 mutations (n=4) were associated with decreased expression of PBRM1 (mean = 309.6 p.n/mm2 versus 4980.2 p.n/mm2 in WT samples; P<0.0001) and ARID1A (mean=1196.15 p.n/mm2; P<0.001), but unchanged SMARCB1 and SMARCA4 expression. BAP1 mutations (n=3) were also associated with decreased BAP1 and ARID1A expression (ARID1A mean = 457.70 p.n/mm2; P<0.001). ARID1A and PBRM1-altered tumor samples showed a poorly infiltrated microenvironment, with decreased CD4+ T cell (159.48 positive cells (p.c)/mm2 and 188.66 p.c/mm2 respectively versus 1126.52 p.c/mm2 for WT; P<0.0001) and CD8+ T cell density (72.38 and 76.54 p.c/mm2 versus 548.58 p.c/mm2 for WT; P<0.0001). Conclusion: Mutations in the ARID1A and PBRM1 subunits of the SWI/SNF complex are associated with a loss of protein expression. In contrast to what reported in other tumor types, ARID1A and PBRM1 mutations correlated with poorly lymphocytic TIME in CCA. Decreased PBRM1 expression is further associated with decreased ARID1A expression. Independent revalidation and further study in larger tumor series are warranted. Citation Format: Clémence Astier, Jean-Yves Scoazec, Virginie Marty, Olivia Bawa, Nicolas Signolle, Carine Ngo, Francesco Facchinetti, Antoine Hollebecque, Sophie Postel-Vinay. Characterization of SWI/SNF complex gene mutations, protein expression and tumor immune microenvironment in cholangiocarcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5714.

Mast Cell Density and Its Clinical Relevance in Waldenstrom's Macroglobulinemia

Background. Waldenström's macroglobulinemia (WM) is characterized by a lymphoplasmacytic infiltration of the bone marrow (BM) along with a serum monoclonal IgM and includes MYD88 L265P and CXCR4 mutations in around 90% and 25% respectively (Treon, NEJM 2012; Hunter, Blood 2014). CXCR4WHIMmutations confer worse outcome and are associated with larger BM involvement and less adenopathies (Treon, Blood 2014). One hallmark and histological criteria for WM diagnosis is the presence of numerous mast cells (MC) segregating with tumor cells in the BM (Waldenström, Acta Haematologica 1958). MC have been shown to support lymphoplasmacytic cell growth, through CD154/CD40 signaling (Tournilhac, Ann Oncol 2006), but there is so far no clear demonstration of the prognostic impact of BM MC density in WM.Aim. We proposed to investigate BM MC density by using sensitive and specific digital quantification, allowing the analysis of large tumor cell infiltrated BM area, in order to assess its clinical relevance in WM.Methods. A total of 65 WM patients were investigated, including 54 pts at diagnosis and 11 at relapse. Mutational status could be determined in 35 pts: allele specific PCR found MYD88 L265P mutation in 32/35 pts (91%) and deep next-generation analysis found CXCR4WHIMmutation in 7/35 (20%) pts. Diagnostic and treatment criteria for WM fulfilled recommendations of the 8th international WM workshop (Castillo, BJH 2016 ; Leblond, Blood 2016). MC density was explored using a digital tool previously used for quantifying immune cells infiltrates on tumor tissues sections (Galon, Science 2006). Tryptase and CD20 immunohistochemisty stainings were done on contiguous sections of deparaffinized BM trephine biopsies. After numerization of each section, BM surface area was manually marked out, excluding bone framework and adipocytes, in order to limit subsequent analyses to hematopoietic tissue. We then defined optimal thresholds for horseradish peroxidase signal detection. MC were counted up with “Immunoscore module” (Definiens Developer XT), on the hematopoietic tissue of the whole BM section : MC density was automately recorded as the number of tryptase positive cells per unit of selected bone marrow tissue surface area.Results. MC density was found to be heterogeneous over BM, with a mean of 106 MC.mm-2 of hematopoietic tissue (Figure 1). Higher MC density (> 100 MC.mm-2) was associated with larger BM involvement by WM cells (BM deep infiltration (p=0.039) and diffuse tumor pattern (p=0.003)) and with less frequent spleen involvement (p=0.011) and a trend towards less lymph node (p=0.072) and liver (p=0.054) involvement. Higher MC density was also associated with features of advanced disease such as anemia (<115g.L-1 ; p = 0.041), thrombocytopenia (<100G.L-1 ; p=0.01) and lower albumin rate (<35g.L-1 ; p=0.03)). Furthermore MC density was 194 MC.mm-2 in patients with high IPSS, 82.4 MC.mm-2 in those with intermediate IPSS and 67.9 MC.mm-2 in those with low IPSS (p=0.0003). Regarding outcome, patients with higher MC density had a shorter median OS (56.5 months vs non reached in patients with low MC density, p = 0.0139)) (Figure 2). Among the pts scheduled to receive a 1st line treatment for an active disase (n=43) and molecularly characterized (n=20), those with CXCR4WHIM mutations had a significantly higher MC density (240.7 MC.mm-2vs 90.9 MC.mm-2 for CXCR4 WT pts ; p=0.0221).Conclusion. By using specific digital tool on well-outlined hematopoietic tissue surface, MC density can be accurately measured in WM patients. High MC density is associated with aggressive features, poor clinical outcome and possibly with CXCR4WHIM mutation. [Display omitted] DisclosuresYsebaert:Janssen: Consultancy, Research Funding, Speakers Bureau. Leblond:Abbvie: Honoraria, Membership on an entity's Board of Directors or advisory committees, Speakers Bureau; Roche: Honoraria, Membership on an entity's Board of Directors or advisory committees, Speakers Bureau; Gilead: Honoraria, Membership on an entity's Board of Directors or advisory committees; Janssen: Honoraria, Membership on an entity's Board of Directors or advisory committees, Speakers Bureau; BMS: Honoraria; Novartis: Honoraria; Servier: Honoraria, Membership on an entity's Board of Directors or advisory committees. Hermine:Hybrigenics: Research Funding; INatherys: Equity Ownership, Research Funding; AB Science: Equity Ownership, Honoraria, Patents & Royalties, Research Funding; Novartis: Research Funding; Celgene: Research Funding. Guieze:GILEAD: Other: Educational Presentation; ABBVIE: Other: Educational Presentation; JANSSEN: Other: Educational Presentation. Tournilhac:Janssen: Honoraria, Other: travel funding; AMGEN: Other: Travel funding, Research Funding; ROCHE: Honoraria, Other: Travel funding, Research Funding; GILEAD: Honoraria, Other: Travel Funding, Research Funding; Abbvie: Honoraria, Other: Travel funding.

Digital pathology imaging and computer-aided diagnostics as a novel tool for standardization of evaluation of aganglionic megacolon (Hirschsprung disease) histopathology.

Based on a recently introduced immunohistochemical panel (Bachmann et al. 2015) for aganglionic megacolon (AM), also known as Hirschsprung disease, histopathological diagnosis, we evaluated whether the use of digital pathology and 'machine learning' could help to obtain a reliable diagnosis. Slides were obtained from 31 specimens of 27 patients immunohistochemically stained for MAP2, calretinin, S100β and GLUT1. Slides were digitized by whole slide scanning. We used a Definiens Developer Tissue Studios as software for analysis. We configured necessary parameters in combination with 'machine learning' to identify pathological aberrations. A significant difference between AM- and non-AM-affected tissues was found for calretinin (AM 0.55% vs. non-AM 1.44%) and MAP2 (AM 0.004% vs. non-AM 0.07%) staining measurements and software-based evaluations. In contrast, S100β and GLUT1 staining measurements and software-based evaluations showed no significant differences between AM- and non-AM-affected tissues. However, no difference was found in comparison of suction biopsies with resections. Applying machine learning via an ensemble voting classifier, we achieved an accuracy of 87.5% on the test set. Automated diagnosis of AM by applying digital pathology on immunohistochemical panels was successful for calretinin and MAP2, whereas S100β and GLUT1 were not effective in diagnosis. Our method suggests that software-based approaches are capable of diagnosing AM. Our future challenge will be the improvement of efficiency by reduction of the time-consuming need for large pre-labelled training data. With increasing technical improvement, especially in unsupervised training procedures, this method could be helpful in the future.

Abstract 4535: Automated image analysis algorithm development for cell surface-binding therapeutic agent target &amp; its comparison to manual pathology scoring

Abstract Standard pathology scoring of cell surface targets in tumor tissue is done routinely by semi-quantitative scoring and is usually performed by a pathologist assessing a given number of fields of view (FOVs) from an immunohistochemistry (IHC) stained slide. In clinical studies evaluating cell surface-binding therapeutic agents, a tissue based patient selection assay with pathology semi-quantitative scoring such as H-score or percent positive is frequently used to enroll patients. Image analysis has become a powerful tool to evaluate and quantify IHC assays including those used for patient enrollment. It allows one to leverage the power of computer software and machine learning approaches to develop algorithms for the detection of specific markers and to run these solutions across entire stained tumor sections. Here, we developed an automated image analysis solution using the Definiens Developer Software specifically to detect a cell surface target in human gastrointestinal tumors. This algorithm is able to detect and quantify apical/membrane staining of the target. We used an initial set of colorectal cancer (CRC) samples (n=30) stained for the target. Tumor and non-tumor compartments were identified using the algorithm; target expression was then assessed in the tumor compartment only. Multiple outputs were generated with the image analysis tool including area of apical stain over the total area of interior luminal border (H-scoreluminal), intensity of apical stain over total apical stain area (H-scoreapical) and apical stain over the total area of the tumor region (H-scoretotal tumor). H-scoreluminal was the output most similar to the semi-quantitative pathology scoring previously performed on these samples. H-scoretotal tumor was the only output that used the total area of the tumor in its composite score. The various image analysis parameters showed good correlation with each other. This use case is an example of applying digital pathology to an existing IHC assay to quantitatively assess marker expression across entire tumor sections. Implementation of quantitative image analysis allows clinical development programs including that assessing cell surface-binding therapeutic agents the ability to incorporate a variety of otherwise difficult-to-measure parameters such as total tumor area and other cellular parameters from an IHC stain into a more robust biomarker analysis package. In addition, this quantitative approach could help in establishing or fine tuning an IHC assay's cut off when used prospectively in clinical trials. Citation Format: Brittany Bahamon, Andrzej Cholewinski, Ruben Cardenes, Aleksandra Zuraw, Israel Barragan, Marise B. McNeeley, Ronald D. Luff, Adnan Abu-Yousif, Hadi Danaee. Automated image analysis algorithm development for cell surface-binding therapeutic agent target &amp; its comparison to manual pathology scoring [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 4535.

Measuring multiple parameters of CD8+ tumor-infiltrating lymphocytes in human cancers by image analysis

BackgroundImmuno-oncology and cancer immunotherapies are areas of intense research. The numbers and locations of CD8+ tumor-infiltrating lymphocytes (TILs) are important measures of the immune response to cancer with prognostic, pharmacodynamic, and predictive potential. We describe the development, validation, and application of advanced image analysis methods to characterize multiple immunohistochemistry-derived CD8 parameters in clinical and nonclinical tumor tissues.MethodsCommercial resection tumors from nine cancer types, and paired screening/on-drug biopsies of non–small-cell lung carcinoma (NSCLC) patients enrolled in a phase 1/2 clinical trial investigating the PD-L1 antibody therapy durvalumab (NCT01693562), were immunostained for CD8. Additional NCT01693562 samples were immunostained with a CD8/PD-L1 dual immunohistochemistry assay. Whole-slide scanning was performed, tumor regions were annotated by a pathologist, and images were analyzed with customized algorithms using Definiens Developer XD software. Validation of image analysis data used cell-by-cell comparison to pathologist scoring across a range of CD8+ TIL densities of all nine cancers, relying primarily on 95% confidence in having at least moderate agreement regarding Lin concordance correlation coefficient (CCC = 0.88–0.99, CCC_lower = 0.65–0.96).ResultsWe found substantial variability in CD8+ TILs between individual patients and across the nine types of human cancer. Diffuse large B-cell lymphoma had several-fold more CD8+ TILs than some other cancers. TIL densities were significantly higher in the invasive margin versus tumor center for carcinomas of head and neck, kidney and pancreas, and NSCLC; the reverse was true only for prostate cancer. In paired patient biopsies, there were significantly increased CD8+ TILs 6 weeks after onset of durvalumab therapy (mean of 365 cells/mm2 over baseline; P = 0.009), consistent with immune activation. Image analysis accurately enumerated CD8+ TILs in PD-L1+ regions of lung tumors using the dual assay and also measured elongate CD8+ lymphocytes which constituted a fraction of overall TILs.ConclusionsValidated image analysis accurately enumerates CD8+ TILs, permitting comparisons of CD8 parameters among tumor regions, individual patients, and cancer types. It also enables the more complex digital solutions needed to better understand cancer immunity, like analysis of multiplex immunohistochemistry and spatial evaluation of the various components comprising the tumor microenvironment.Trial registrationClinicalTrials.gov identifier: NCT01693562.Study code: CD-ON-MEDI4736–1108.Interventional study (ongoing but not currently recruiting).Actual study start date: August 29, 2012.Primary completion date: June 23, 2017 (final data collection date for primary outcome measure).

Object-based image analysis supported by data mining to discriminate large areas of soybean

ABSTRACTThis research aimed to analyze the possibility to estimate and automatically map large areas of soybean cultivation through the use of MODIS (Moderate-Resolution Imaging Spectroradiometer) images. Two major techniques were used: GEOgraphic-Object-Based Image Analysis (GEOBIA) and Data Mining (DM). In order to obtain the images, the segmentation algorithm implemented by Definiens Developer was used. A decision tree (DT) was created from a training set previously prepared. Time-series of images from the MODIS sensor aboard the Terra satellite were acquired in order to represent the wide variation of the vegetation pattern along the soybean crop cycle. The time-series data were used only for the CEI index. Furthermore, to compare the results obtained from GEOBIA, the slicing technique was used at the CEI level. After the training, the DT was applied to the vegetation indices generating the thematic map of the spatial distribution of soybean. In accordance with the error matrix and kappa parameter analysis, tests for statistical significance were created. Results indicate that the classification achieved by Kappa coefficients is 0.76. In short, the obtained results proved that combining vegetation indices and time-series data using GEOBIA return promising results for mapping soybean plantation on a regional scale.

Ki-67 in endometrial cancer: scoring optimization and prognostic relevance for window studies

Ki-67, a marker of cellular proliferation, is increasingly being used in pre-surgical window studies in endometrial cancer as a primary outcome measure. Unlike in breast cancer, however, there are no guidelines standardizing its measurement and its clinical relevance as a response biomarker is undetermined. It is, therefore, imperative that Ki-67 scoring protocols are optimized and its association with patient survival rigorously evaluated, in order to be able to clinically interpret the results of these studies. Using the International Ki-67 in Breast Cancer Working Group guidelines as a basis, whole slide, hot spot and invasive edge scoring protocols were evaluated using endometrial biopsies and hysterectomy specimens from 179 women. Whole sections and tissue microarrays, manual and semi-automated scoring using Definiens Developer software were additionally compared. Ki-67 scores were related to clinicopathological variables and cancer-specific survival in uni- and multivariate analysis. Against criteria of time efficiency, intra- and inter-observer variability and consistency, semi-automated hot spot scoring was the preferred method. Ki-67 scores positively correlated with grade, stage and depth of myometrial invasion (P-values all <0.03). By univariate analysis, higher Ki-67 scores were associated with a significant reduction in cancer-specific survival (P≤0.05); however, this effect was substantially attenuated in the multivariate model. In conclusion, hot spot scoring of whole sections using Definiens is an optimal method to quantify Ki-67 in endometrial cancer window study specimens. Measured this way, it is a clinically relevant marker, though further work is required to determine whether reductions in Ki-67 in neoadjuvant intervention studies translate into improved patient outcome.

Image Analysis Approach To Distinguish Lobular Structures In The Mammary Gland From Well-Differentiated Breast Cancer With Tubule Formation

Introduction/ Background Automated detection of diagnostically relevant regions of interest (ROI) is one of the major challenges in medical image analysis. In digital pathology using whole slide images (WSI), the detection of certain biological structures is necessary, because the capacity for analysis of cells and subcellular structures at high resolution is limited by available processing time and computational power. For some applications it is also important to exclude irrelevant or misleading components of the image. One important challenge of this type is well differentiated (low-grade/G1) breast cancer. Aims We aim at automatically distinguishing non-malignant lobular tissue from well-differentiated breast cancer to avoid erroneous evaluation of normal tissue in the detection of nuclear hormone receptor expression. A second goal of lobule detection is to exclude inflammation of non-malignant pre-existing structures from tumor immune cell scoring of breast cancer in oncoimmunology. Methods Two approaches for lobule detection were applied: The first includes modules of own image analysis algorithms developed specifically for lobule detection combined with elements of a commercially available software platform (Definiens Developer®), and the second is a software package optimized for use with a multispectral camera system (Inform, Perkin-Elmer ®). Breast cancer samples were stained for estrogen receptor (ER), progesterone receptor (PR) and the lymphocyte marker CD8. The first approach starts with a texture-based supervised classification to detect lobule candidate regions and uses a nuclear density image to refine the candidate regions. The second approach uses a supervised machine learning method whose features and algorithm are not disclosed to the user. Manual annotations of lobular tissue by expert pathologists were used for evaluation of results. Results The accuracy of distinction between cancer areas and lobular structures decreased in cases with prominent glandular differentiation of the tumor. The major challenge was the separation of well-differentiated (G1) breast cancer with consistent hormone receptor expression from adjacent lobular areas. The second approach performed well on high-grade tumors and had advantages regarding speed and its convenient user interface. The modular first approach was superior on ER and PR images and successfully detected lobular areas even if the anatomical structure was disrupted by inflammation of infiltrating cancer cells. We conclude that modular approaches considering image context and allowing specific adjustments to the tissue of interest may be needed to overcome current limitations of automated ROI detection in clinical biopsy materials.

Abstract LB-285: Computational pathology for predicting prostate cancer recurrence

Abstract Background: Conventional methods for predicting prostate cancer (PCa) aggressiveness, which rely heavily on the Gleason score of architectural pattern, remain imperfect. We have developed an approach that combines machine vision and machine learning analysis of routine H&amp;E stained tumors to discriminate recurrent from non-recurrent patients after prostatectomy, based on architectural, color, and nuclear patterns. Methods: We used a set of H&amp;E-stained TMAs containing quadruplicate cores from 176 recurrent cases of PCa and 154 non-recurrent controls, frequency matched on age, Gleason, pathological stage and race. Using software specifically adaptable for computational pathology with hierarchical object recognition (Definiens Developer XD®), we processed images of each tumor at two spatial scale levels: a larger scale for recognition of tissue compartments (epithelium, stroma, lumen and inflammation), and a smaller scale for recognition of normal and atypical nuclei in each of the compartments. Magnification of 20x was used for both levels. At the tissue level, we used multi-resolution segmentation to segment homogeneous superpixels, which were then classified using machine learning into epithelium, stroma. or ambiguous tissue classified as intermediate between epithelium and stroma. We used relative darkness in a spatial neighborhood, size, and shape criteria to segment typical and atypical nuclei. Relational features among objects within and between the tissue and nuclear levels were also created. The resultant feature library contained 884 base features and 3,551 total variables. We used several stages of iterative L1 penalized logistic regression and 5-fold cross-validation to reduce variables and obtain AUC estimates. Variable stability criteria applied across the folds further minimized model overfitting. Results: PSA, Gleason and CAPRA-s achieved AUCs &amp;lt; 0.60 for recurrence in this matched dataset. A fifth stage L1 model containing 18 histometric features had cross-validation AUC = 0.81 (95% CI: 0.76-0.85). A more permissive 74 feature model had AUC = 0.95 (95% CI: 0.93-0.97). Subgroup analysis restricted to Gleason 7 cases produced AUCs ranging from 0.79-0.93 with varying model sparseness. Including pre-surgical PSA, Gleason grade or CAPRA-s score did not change model performance. Notably, leading features were frequently derived from stromal or ambiguously stromal regions and involved nuclear shape or texture. Their independence from tumor grading reflected subject matching and lack of explicit accounting for nuclear or stromal characteristics in the Gleason criteria. Conclusion: A computational pathology approach, applied to discrimination of recurrent and non-recurrent PCa in routine H&amp;E stains, shows considerable promise and appears capable of adding to the predictive power of Gleason and CAPRA scoring. We will conduct additional validation studies and extend this work to biopsy samples. Citation Format: Amit Sethi, Lingdao Sha, Ryan J. Deaton, Virgilia Macias, Andrew H. Beck, Peter H. Gann. Computational pathology for predicting prostate cancer recurrence. [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 LB-285. doi:10.1158/1538-7445.AM2015-LB-285

A knowledge-based, transferable approach for block-based urban land-use classification

In this work we propose a knowledge-based approach for land-use classification of city blocks through the automatic interpretation of very-high-resolution remote-sensing imagery. Our approach is founded on geographic object-based image analysis (GEOBIA) concepts and is concerned with transferability across distinct knowledge representation formalisms. This paper therefore investigates the viability of translating a high-level description of the interpretation problem into the particular knowledge representation structures and interpretation strategies of two different software platforms, namely the proprietary Definiens Developer system and the open-source InterIMAGE system. Initially, textual descriptions of the land-use classes of interest were created by photo interpreters. Then, generic class descriptions were defined as a system-independent knowledge model, which was subsequently translated into interpretation projects in the different systems. Altogether 49 blocks located on two different test-sites in the city of São Paulo (Brazil) were considered in the experiments. Although the classification results from the Definiens Developer system were slightly better than those obtained with the InterIMAGE system, we concluded that both systems have been shown to be equally qualified to implement the target application properly through adaptation of the generic knowledge model.

A mixed pixel- and region-based approach for using airborne laser scanning data for individual tree crown delineation in Pinus radiata D. Don plantations

The aim of this study was to evaluate the use of high-resolution airborne laser scanner (ALS) data to detect and measure individual trees. We developed and tested a new mixed pixel- and region-based algorithm (using Definiens Developer 7.0) for locating individual tree positions and estimating their total heights. We computed a canopy height model (CHM) of pixel size 0.25 m from dense first-pulse point data (8 pulses m−2) acquired with a small-footprint discrete-return lidar sensor. We validated the results of individual tree segmentation with accurate field measurements made in 37 plots of Monterey pine (Pinus radiata D. Don) distributed over an area of 36 km2. Fieldwork consisted of labelling all of the trees in each plot and measuring their height and position, for posterior integration of the data from both sources (field and lidar). The proposed algorithm correctly detected and linked 59.8% of the trees in the 37 sample plots. We also manually located the trees by using FUSION software to visualize the raw lidar data cloud. However, because the latter method is extremely time-consuming, we only considered 10 randomly selected plots. Manual location correctly detected and linked 71.9% of the trees (in this subsample the algorithm correctly detected and measured 63.5% of the trees). The R2 values for the linear model relating field- and lidar-measured heights of the linked trees located manually and with the automatic location algorithm were 0.90 and 0.88, respectively.

An entirely automated method to score DSS-induced colitis in mice by digital image analysis of pathology slides

SUMMARYThe DSS (dextran sulfate sodium) model of colitis is a mouse model of inflammatory bowel disease. Microscopic symptoms include loss of crypt cells from the gut lining and infiltration of inflammatory cells into the colon. An experienced pathologist requires several hours per study to score histological changes in selected regions of the mouse gut. In order to increase the efficiency of scoring, Definiens Developer software was used to devise an entirely automated method to quantify histological changes in the whole H&E slide. When the algorithm was applied to slides from historical drug-discovery studies, automated scores classified 88% of drug candidates in the same way as pathologists’ scores. In addition, another automated image analysis method was developed to quantify colon-infiltrating macrophages, neutrophils, B cells and T cells in immunohistochemical stains of serial sections of the H&E slides. The timing of neutrophil and macrophage infiltration had the highest correlation to pathological changes, whereas T and B cell infiltration occurred later. Thus, automated image analysis enables quantitative comparisons between tissue morphology changes and cell-infiltration dynamics.

Object Based Image Analysis and Data Mining applied to a remotely sensed Landsat time-series to map sugarcane over large areas

The aim of this research was to develop a methodology for contributing in the automation of sugarcane mapping over large areas, with time-series of remotely sensed imagery. To this end, two major techniques were combined: Object Based Image Analysis (OBIA) and Data Mining (DM). OBIA was used to represent the knowledge needed to map sugarcane, whereas DM was applied to generate the knowledge model. To derive the image objects, the segmentation algorithm implemented in Definiens Developer® was used. The data mining algorithm used was J48, which generates decision trees (DT) from a previously prepared training set. The study area comprises three municipalities located in the northwest of São Paulo state, all of which are good representatives of the agricultural conditions in the Southern and Southeastern regions of Brazil. A time series of Landsat TM and ETM+ images was acquired in order to represent the wide range of pattern variation along the sugarcane crop cycle. After training, the DT was applied to the Landsat time series, thus generating the desired thematic map with sugarcane ready to harvest. Classification accuracy was calculated over a set of 500 points not previously used during the training stage. Using error matrix analysis and Kappa statistics, tests for statistical significance were derived. The statistics indicated that the classification achieved an overall accuracy of 94% and a Kappa coefficient of 0.87. Results show that the combination of OBIA and DM techniques is very efficient and promising for the sugarcane classification process.

Fluorescence Compared to Brightfield Detection of CD35 in Lymphoid Follicles of the Monkey to Assess Immunomodulatory Treatment Effect

AimsLymphoid follicles of secondary lymphoid organs are dynamic structures inducible by immunization and/or inflammatory stimulation. Follicular dendritic cells (FDC) are specialized antigen‐presenting cells exclusive to the germinal centers (GC) of lymphoid follicles. Within the GC, FDC's form a reticular network that sustains the architecture of lymphoid follicles and interact with the GC B cells via several mechanisms including complement receptors such as CD35. The expression of CD35 in lymphoid follicles is therefore a measure of the efficacy of immunomodulatory (IMM) treatment strategies.MethodsCD35 in formalin‐fixed paraffin‐embedded spleen sections of cynologous monkeys was detected with a mouse monoclonal antibody and visualized using a polyclonal secondary antibody coupled to either horse‐radish peroxidase for DAB precipitation or the fluorophore Alexa647. Digital brightfield and fluorescence images (20x) were acquired on a Nanozoomer 2HT (Hamamatsu) and Ariol SL‐50 (Genetix) whole slide scanner respectively. Images were quantified using Definiens Developer software.ResultsFluorescence significantly increased the dynamic range of CD35 compared to chromogen. Fluorescence signal intensities ranged from 0 – 256 compared to 0 – 0.8 using chromogen. Both detection methods showed a small but significant decrease in follicular CD35 following IMM treatment without resolving dose effects. However the increased dynamic range of fluorescent CD35 allowed a robust distinction between follicles containing germinal centers and those without. Thus a clear dose effect after IMM treatment could be resolved using fluorescence, but not chromogen for CD35 detection.

A novel algorithm for land use and land cover classification using RADARSAT-2 polarimetric SAR data

This study proposes a new four-component algorithm for land use and land cover (LULC) classification using RADARSAT-2 polarimetric SAR (PolSAR) data. These four components are polarimetric decomposition, PolSAR interferometry, object-oriented image analysis, and decision tree algorithms. First, polarimetric decomposition can be used to support the classification of PolSAR data. It is aimed at extracting polarimetric parameters related to the physical scattering mechanisms of the observed objects. Second, PolSAR interferometry is used to extract polarimetric interferometric information to support LULC classification. Third, the main purposes of object-oriented image analysis are delineating image objects, as well as extracting various textural and spatial features from image objects to improve classification accuracy. Finally, a decision tree algorithm provides an efficient way to select features and implement classification. A comparison between the proposed method and the Wishart supervised classification which is based on the coherency matrix was made to test the performance of the proposed method. The overall accuracy of the proposed method was 86.64%, whereas that of the Wishart supervised classification was 69.66%. The kappa value of the proposed method was 0.84, much higher than that of the Wishart supervised classification, which exhibited a kappa value of 0.65. The results indicate that the proposed method exhibits much better performance than the Wishart supervised classification for LULC classification. Further investigation was carried out on the respective contribution of the four components to LULC classification using RADARSAT-2 PolSAR data, and it indicates that all the four components have important contribution to the classification. Polarimetric information has significant implications for identifying different vegetation types and distinguishing between vegetation and urban/built-up. The polarimetric interferometric information extracted from repeat-pass RADARSAT-2 images is important in reducing the confusion between urban/built-up and vegetation and that between barren/sparsely vegetated land and vegetation. Object-oriented image analysis is very helpful in reducing the effect of speckle in PolSAR images by implementing classification based on image objects, and the textural information extracted from image objects is helpful in distinguishing between water and lawn. The decision tree algorithm can achieve higher classification accuracy than the nearest neighbor classification implemented using Definiens Developer 7.0, and the accuracy of the decision tree algorithm is similar with that of the support vector classification which is implemented based on the features selected using genetic algorithms. Compared with the nearest neighbor and support vector classification, the decision tree algorithm is more efficient to select features and implement classification. Furthermore, the decision tree algorithm can provide clear classification rules that can be easily interpreted based on the physical meaning of the features used in the classification. This can provide physical insight for LULC classification using PolSAR data.

Study on Object-oriented Remote Sensing Image Classification Based on Multi-levels Segmentation

ALOS image data was used to carry out a multi-levels segmentation with a method called FNEA in Definiens Developer 7 solftware and image objects were got.Spectral and spatial values of image objects,as well as relationship of objects among different levels were considered to extract land use and land cover information in the test area located in Honghu City,Hubei Province.Then an object-oriented classification based on single level segmentation and a pixel-based Maximum Likelihood classification were used to compare with it.Results showed that the object-oriented classification based on multi-levels segmentation not only overcame Pepper and Salt Phenomenon appeared in the pixel-based Maximum Likelihood classification but also obtained a significant improvement on classification accuracy compared with the other two classification methods.

The Analysis of Single Polarization Synthetic Aperture Radar Images for the Application of Land-use and Land-cover Change

For the investigation of land use and land cover change,taking the limitation of optical remote sensing in rainy or cloudy regions of China into consideration,synthetic aperture radar (SAR) can be an alternative way.In this paper,we first introduce an object-oriented imagery segmentation and classification software named Definiens Developer for the land use classification of single polarization spaceborne SAR images,including median resolution spaceborne SAR( take ENVISAT ASAR and Radarsat-1 as examples) and high resolution spaceborne SAR(take TerraSAR-X as an example).Then,we estimate the classification accuracy and analyze the potentials for the two modes spaceborne SAR images in the classification of land use and land cover change.In this part,conclusions for the interpretation of their corresponding SAR images are also given.

Object-oriented classification of multi-resolution images for the extraction of narrow linear forest disturbance

Narrow linear forest disturbances (e.g. seismic cut lines) have been found to have significant effects on wildlife habitat and biodiversity (e.g. species richness and abundance). A great deal of seismic cut lines is created in oil and gas exploration in natural forest areas every year. Accurate mapping of seismic cut lines can therefore contribute to a better understanding of wildlife habitat and biodiversity. However, previous studies have indicated that seismic cut lines were fairly difficult to detect and map even with the available high-spatial resolution imagery (e.g. Satellite Pour l'Observation de la Terre, SPOT 5). Recent progress in feature segmentation and extraction software, such as Definiens Developer 7.0, has enhanced remote sensing capabilities, with the promise of being able to automate tasks. This study investigated the imagery (high resolution or very high resolution) best suited for extracting seismic cut lines using a set of rules and multi-resolution object-oriented classification methods. The data used include SPOT 5 and QuickBird multispectral images and existing Geographic Information System (GIS) databases within one bear management area (BMA) in the eastern slopes of the Rocky Mountains in Alberta. Results indicated that among the available algorithms in the Definiens Developer 7.0 package, the Lee sigma algorithm was capable of highlighting cut lines using the near-infrared (NIR) band of the SPOT 5 image and the QuickBird image. The multi-resolution segmentation was able to segment fresh cut lines when giving higher weight to the Lee sigma edge extraction layer, and the nearest-neighbour object-oriented classifier was able to classify linear features, but with noise. Classification accuracy increased following the post-classification refinement processing. The accuracy assessment indicated that, in the case of delineating fresh cut lines, the higher resolution QuickBird image performed better than SPOT 5. However, neither the QuickBird image nor the SPOT 5 image could accurately delineate relatively old cut lines.