A framework for image analysis in plant breeding. Paul van Schayck

Transcription

1 A framework for image analysis in plant breeding Paul van Schayck

2 Paul van Schayck Wageningen University MSc Biotechnology (molecular life science) Nanoscopy Data management

3 Image analysis in plant breeding Marker assisted breeding Markers: Molecular Biochemical Morphological Shape, color and quantitative Tool for breeding From melon to cucumber

4 Examples of image analysis used in plant breeding An ImageJ plugin for plant variety testing Polder, G., Blokker, G., Heijden, G.W.A.M. van der (2012) In: Proceedings of the ImageJ User and Developer Conference 2012, October 2012, Mondorf-les-Bains, Luxembourg. - : Centre de Recherche Public Henri Tudor, p

5 Existing practice of image analysis Use of popular open source tools

6 Current work flow of image analysis Control flow Images and results Windows Server Via remote desktop Pipeline Macro End users: Breeders Researchers Network disks Portable disks SMB Local disks Storage

7 Current challenges Increasing size of datasets Data storage decentralized Little structure in metadata: Identification Results Quality assurance User interface

8 Current challenges Difficult end-user experience

9 Aim of project Improve end-user experience Store metadata of results and images Improve data storage capabilities Improve performance scalability Improve quality assurance

10 Discovering OMERO

11 Pipeline of image analysis Pipeline common library OMERO.Web: OMERO. Processor: Control script (Python) Macro Plugin OMERO.Insight: High Performance Computing cluster Control flow Data flow (images and annotations) Validation image Results

12 Progress monitoring Making use of jobhandle.setmessage()

13 Infrastructure of the framework Big-data storage VMware node NFS Image data OMERO.Web: Client NFS slurm DRMAA Slurm HPC nodes HPC cluster

14 Acquisition of images into OMERO OMERO. Importer Identification metadata Barcode scanner Image data NAS storage

15 Quality Assurance Version control (SVN) Doc generation (Sphinx) Automated testing Functional and acceptance (Python nose) Nightly tests. Reporting to monitoring system (PRTG). Testing environment Monitoring system availability and regressions

16 Test case: Shape parameters of carrot Determine different shape parameters Crop selection

17 Step 1: Acquisition and calibration Batch-wise import Color and light control script Image data After Before NAS storage HPC Cluster

18 Step 2: Classification and segmentation Crucial step Time-consuming calculations control script HPC cluster

19 Step 3: Measurement of parameters Results in CSV format FileAnnotation control script Validation image Results HPC cluster

20 Conclusions of the carrot test case Further implementation: Validation Integration Web Client User Results

21 Conclusions Improved end-user experience Improved end-user experience

22 Conclusions Before Improved end user experience Centralized storage After Network disks Image data Portable disks Local disks Storage Centralized NAS storage Centralized storage

23 Conclusions Improved end user experience Centralized storage Storage of metadata Storage of metadata

24 Conclusions Improved end user experience Centralized storage Storage of metadata Performance scalability Image data HPC nodes NAS storage HPC cluster Performance scalability

25 Conclusions Improved end user experience Centralized storage Storage of metadata Performance scalability Complexity for developers Software dependencies At the cost of More steps required by developers Software dependencies

26 Conclusions Improved end user experience Centralized storage Storage of metadata Performance scalability Complexity for developers Software dependencies Quality assurance Quality assurance

27 Acknowledgements Raimond Ravelli Peter Peters Jan-Willem Borst