WP07 - Optimise image processing and software development

This WP is aimed to develop tools to facilitate and improve image analysis by integrating algorithms for automated image processing and analysis into a user-friendly software tool. The objectives of WP07 are:

  • Develop and test algorithms for automatic quality assurance (Task 1, 3)
  • Develop and test algorithms for automatic processing of images (Task 1, 2, 3)
  • Develop and test tools for automatic or semi-automatic extraction of information relevant to treatment decisions (Task 3, 4)
  • Develop and disseminate ready-to-use software incorporating processing algorithms and tools (Task 5)

The evaluation acute stroke MR images for acute treatment decisions requires experience. The interpretation of acute stroke MRI is user dependent. This leads to imbalances in the quality of stroke MRI assessment between different centres which results in inhomogeneity of the quality of acute stroke treatment and hampers the performance of clinical trials relying on acute stroke MRI. Therefore, we need automated processing and analysis algorithms. Automated and semi-automated image analysis tools can help standardise and improve the quality of image analysis and offer advanced treatment decisions to a wider community.

M. Guenther is an expert in MR Image acquisition and Analysis. He holds several patents on MR sequences and Image processing algorithms. He and his Group at FME have extensive experience in Image processing and visualization with a proven track of European cooperative Projects (Heckel et al. 2009, Wang et al. 2010). To this end, the Software platform MeVisLab (www.mevislab.de) is being developed to efficiently prototype and deploy applications for the clinical routine. MeVisLab is provided for free to the Research community and has a large number of users worldwide who also share plugins as open source. The SME mediri has extensive expertise in automated image quality assessment having used this approach in several clinical trials. The group of J. Fiehler at the Department of Neuroradiology at UKE has done pioneering studies on perfusion and diffusion weighted MRI in acute stroke (Fiehler et al. 2001). In cooperation with the Department of Computational Neurosciences at UKE they have also long-standing experience in creating imaging processing software solutions, e.g. the Analysis Tool for Neuro Image Data (AnToNIa) (Forkert et al. 2009). DWI and FLAIR images from the PRE-FLAIR study will be used during the first steps of algorithm development until data from the WAKE-UP clinical trial are available. Participant 01 (UKE) will mainly focus on tasks 2 and 4. Participant 10 (mediri) will be involved in all tasks with a main focus on tasks 1, 3 and 5, and Participant 09 (FME) will mainly work on the technical implementation of algorithms developed in tasks 2, 4 and 5. The cooperation within this WP will be accompanied by clear regulation of intellectual property issues associated with developed and used algorithms and tools.

Task 1: Develop Algorithms for automatic image quality assessment (Month 01-12) (Del 07.01)

First, parameters and minimum standards of image quality will be defined (together with WP 02). Second, an algorithm for automatic quality assurance (AQUA) will be developed assessing various aspects of image quality (like protocol compliance, signal-to-noise ratio, contrast, artefact level, blurring, etc.). The algorithm will be made available by month 12.

Task 2: Develop algorithms for automated image processing (Month 01-12)

Algorithms for automatic performance of image processing steps required (i.e. realignment, co-registration, skull stripping, artefact correction, automatic window settings) will be developed. Algorithms will incorporate information from AQUA. The algorithm will be made available made available by month 12.

Task 3: Integration of automatic quality assurance into image data management (Month 07-24)

AQUA will be integrated in the process of data management in the clinical trial. Image quality will be assessed within a framework, which checks the image data at the time of uploading to the database (made available by month 24). Feedback is given to the central image reading board and the participating imaging sites to improve protocol compliance and image quality (together with WP 05). Automated image processing will also be integrated into the process of data management. The result of the image quality check will determine which post-processing tools will be applied to extract the relevant parameters. Access to the data is provided for visual control of the analysis result as well as for manual interaction where necessary. The algoritm will be made available by month 24.

Task 4: Develop algorithms for (semi-)automated information extraction (Month 01-24) (Del 07.02)

In this task algorithms will be developed to automatically or semi-automatically extract the relevant information for treatment decisions from the acute MR images (DWI, FLAIR). This comprises the definition and quantification of DWI and FLAIR lesion, the calculation of signal changes (apparent diffusion coefficient decrease, FLAIR signal intensity increase), the assessment of the severity of leukoaraoisis, and of the presence and frequency of microbleeds. The centrally stored MR images will be post-processed by specialised software tools to automatically (or semi-automatically) extract the relevant information. This process will be determined by the previous steps (AQUA, image processing). Results of automated image analysis will be cross-checked with judgements from the central image reading board (WP 05). The algorithm will be made available by month 24.

Task 5: Integration of the algorithms developed into a ready-to-use software tool (Month 19-54)

The algorithms and modules developed in tasks 1, 2, 3, and 4 will be incorporated into a software tool. This software will be designed to be user friendly and intended to be used without specific stroke MRI experience at the scanner and to allow for real-time computer assisted image judgement. A beta version of the software will be made available by month 42 and distributed and tested within the scientific partners of the consortium. The final software will be made available by month 54. After completion of the clinical trial the software shall be made available to stroke physician within the EU (together with WP 09). During this stage support for the implementation and use of the software will be provided (together with WP 09).

This WP will be led by Participant 09 (Matthias Günther, FME). The WP will be active during months 01-60.

Participants involved in WP07: