CELLENGER CELLENGER Automated High Automated High Content - PDF document

CELLENGER CELLENGER Automated High Automated High Content Content Analysis of Analysis of Biomedical Biomedical Imagery Imagery Dr. Maria Athelogou the cognitive computing company the cognitive computing company Founded in 1994 as Delphi 2 Creative Technlogies GmbH Founded by Prof. Dr. Gerd K. Binnig, Nobel Laureate in Physics 1986, and Dieter Herold, Science Journalist Headquartered in Munich, Germany

The challenge of understanding images is not just to analyze a piece of information locally but also to bring the context into play (G. Binnig) Image Understanding Understanding with with Cognition Cognition Networks Networks Image

Automated Image Understanding Automated Image Understanding Find all relevant objects and their mutual relations create a hierarchical linked object structure „segmentation“ Link these objects with the knowledge about them and their relations knowledge representation hierarchical classification Cellenger meta language for Image Understanding Cellenger meta language for Image Understanding Cellenger technology supports a GUI based meta language that allows for fast and efficient development of rule bases. A rule base addresses the solution of a specific image analysis task Basic components are processes and fuzzy classification that support knowledge based segmentation Locally specific Classes with fuzzy Result processes class descriptions Network of objects of interest with attributes and mutual relations

Representation of Input Data in a Hierarchical Structure Representation of Input Data in a Hierarchical Structure

Methodical advantages Methodical advantages Extensive integration of processes and semantics Object segmentation and processing domain Result Network of objects of interest in proper shape Fuzzy knowledge base with correct Input: image data labeling noisy, textured, heterogeneous structures of interest Semantics: fuzzy classification domain Feature (Attribute) Hierarchy Hierarchy Feature (Attribute) Object related Color Form Texture Classification related z.B. Rel. Border Length to Objects of Class „Endothel“ Relations between Objects Length of common border Globale Attributes Number of Objects of a Class Meta-Data Variables

Class Hierarchy Class Hierarchy: : The The „ „Knowledge Knowledge Base“ Base“ describes problem expert knowledge internal analysis classes Inheritance Semantic Grouping Class Description and Classification by Fuzzy – Logik - Expressions Internal Class Class Description Description Internal Fuzzy – Membership Functions Integration of other Classificators

Adaptive local Processing by Domains Adaptive local Processing by Domains each process consists of algorithm domain algorithm describes what happens domain describes where it happens intuitive support of local processing adaptiv processing Cellbased Solutions Solutions for for Cellbased Cellbased Cellbased Quantification Quantification Fluoresce /RNAi (Cells): Find Nuclei/Separate Nuclei/Classify Nuclei Find Cells/Separate Cells/Classify Cells Mutual relations between Cells and Nuclei Classify Cells accoding morphological properties Her2Neu Membran Expression (Tissue): Find Nuclei/Separate Nuclei/Classify Nuclei Find Cells/Separate Celles/Classify Cells Mutual relations between Cells and Nuclei Classify Cells according the membran expression of Her2Neu %

Benefit of Benefit of Cellenger Cellenger Request from the FDA / EMEA) � additional Tox./ Histopath. – Data for the final submission Stack of approx. 15.000 images Based on state of the art technologies Expected 3 -4 Pathologists � 6 months Automated, quantitative Analysis of cell & tissues Benchmark (Pathologist / Cellenger) Result within a few weeks Reduced the time from 6 to 3 months with 2 Pathologists Time to market benefit Cellenger Cellenger

Phase: Market Discovery Pre-Clinical Trials Clinical Trials approved Disease Lead Animal Human Lead Phase Diagnostics Selection Identification Optimization Rodent Control I-IV Target ADME/TOX Non-Rodent Patient Identification Cell&Tissue-based assays in R&D It is generally recognized that cell & tissue-based assays which better mirror physiology and disease are the future of research and drug development in the life sciences sector Detailed quantitative information in Detailed quantitative information in cells & tissues cells & tissues The bottleneck: The bottleneck: automated image analysis automated image analysis Biomedical image data set a significant challenge for automated image analysis: current standard, pixel based procedures for image analysis fail A lot of manual intervention is needed for analysis expensive, subjective and time consuming: many projects cannot be driven Only rough and qualitative description is provided no detailed, quantitative description of cell and tissue structure Based on the new object-based image analysis technology Cellenger overcomes this operational gap

Cellenger Solutions through through the the whole whole R&D R&D process process R&D Phase: Discovery Pre-Clinical Trials Clinical Trials Lead Lead Animal Human Phase Medical Disease Identification Optimization Selection Rodent Control I-IV Diagnostics Target Identification ADME/TOX Non-Rodent Patient HT/HC Screening Histopathology CT/MR/US.. Ultra structure research, Electron microscopy Ultra structure research, Electron microscopy P. Biberthaler, M. Athelogou, R. Leiderer, K. Messmer, European Journal of Medical Research, Juli 2003 Definiens AG, ICF LMU, Klinikum Großhadern

MPI-CBG/Cenix Dresden Original 0 : her2Neu_174 Original 2 : her2Neu_177 Original 3 : her2Neu_179 DacoCytomation

Statistics Statistics Count of cells of different classes Original 2 : her2Neu_177 Original 0 : her2Neu_174 Original 3 : her2Neu_179 4 9 79 39 109 232 2 3 1122 360 more than 75 % membran expr her2/Neu more than 75 % membran expr her2/Neu more than 75 % membran expr her2/Neu 50% to 75% membran expr her2/Neu 50% to 75% membran expr her2/Neu 50% to 75% membran expr her2/Neu 25% to 50% membran expr her2/Neu 25% to 50% membran expr her2/Neu 25% to 50% membran expr her2/Neu less than 25% membran expr her2/Neu less than 25% membran expr her2/Neu less than 25% membran expr her2/Neu KI-67-Pathology- Berlin

Histology, Pathology: Histology, Pathology: I Inflamatory Areas in Liver Tissue Pathology LMU Munich Pathology / Toxicology: analysis of proliferation index in jejunum Image data courtesy Novartis Pharma AG; Pathology / Toxicology EU

Siemens Medical Erlangen 2D+T/3D/3D+T Cellenger Solutions Inst. For Clinical Radiology LMU Munich Klinikum Großhadern

Inst. For Clinical Radiology LMU Klinikum Großhadern Brain Imaging Center, Universit at Frankfurt am Main

Brain Imaging Center, University at Frankfurt am Main Cellenger Enterprise Cellenger Enterprise modular client-server system designed to fully automate image analysis and report generation for biomedical solutions open architecture for flexible workflow integration Enterprise Client enables user access to all details of object-oriented image analysis results

CELLENGER Definiens Software Tool and Technology could be used for Validation of Medical Reference Image Databases Database: Image Content Query System Image Content Query System Database: Storage of image data Scalable: few to large data amounts Administration: organize in experiments, groups, scenes, tiles .. Annotation ODBC protocol, support of Oracle (DB2, MS SQL ... ) Storage of results Case analysis and statistics per scene Case analysis and statistics per experiment Optional: morphometric data per scene Optional: network of extracted image objects

Business Benefit Business Benefit of of Cellenger Cellenger Automated, quantitative Analysis of cell & tissues Understanding of complex events, mechanism of action Decision support for non hit/ hit / potential hit (HCS/HTS) Reduces the # of screens Decision support for Toxicity / Efficacy (Tox./ Histopath.) Reduce the # of leads (fail earlier/ fail cheaper >30%)) Harmonisation benefit Time to market benefit Thank you for your attention Thank you for your attention Image data courtesy MPI Cell Biology and Genetics Dresden