TP4: Application, Implementation, and Validation
Subproject leaders
Prof. Dr J. Modersitzki, Prof. Dr K. Schäfers, Dr F. Wübbeling
Project researchers
Dr Benjamin Wacker, Fjedor Gaede, Julian Johannes Kuhlmann
Project partners
Siemens Healthineers, Novartis, Münster University Hospital, University Medical Center Schleswig-Holstein
Contact
Objectives
TP4 is the interface between the theoretical projects TP1, TP2, and TP3 and the application partners—clinicians, Novartis, and Siemens.
TP4 receives images and raw data from functional MR, PET/CT, and PET/MR supplied by the clinical partners, prepares them using the existing data definitions, and makes them available to all subprojects in standardised, anonymised form. It extends the existing research platforms for emission tomography and image registration to process time-continuous motion information. In collaboration with TP1, TP2, and TP3, the algorithms developed there will be implemented. They will be validated using phantom measurements and evaluated with the clinical partners for their clinical relevance.
Input (from project partners)
Siemens Healthineers: Access to documentation, raw data, algorithms, and tools for data standardisation and correction (already available). Close collaboration to assess feasibility and market potential in clinical settings. In a project on discrete motion correction using external and internal markers on Siemens Biograph scanners, new clinical motion-correction algorithms will also be provided as prototypes.
Novartis: Support through requirements definition and assessment of algorithmic benefit for pharmaceutical research.
Clinical institutes: Provision of existing datasets, experiments and support for validation, and medical expertise in radiology, nuclear medicine, and nephrology.
Output (to project partners)
Prototype implementations of the developed algorithms; support for industrial implementation with Siemens and Novartis; technical assistance with data processing for TP1–TP3; and provision of a test environment.
Previous work
All project partners have worked for years on motion correction in medical data, particularly on medical image registration and image reconstruction. Their methods underpin many approaches to motion correction.
For mathematical PET reconstruction, we use the open research platforms emrecon and FAIR. emrecon has been developed since 2005 by the subproject leaders together with Dr T. Kösters (now at Siemens Healthineers). It implements modern emission-tomography reconstruction methods, such as OSEM with variational regularisation terms including total variation, on all scanners available to the project, especially the Siemens mCT (PET/CT) and Siemens mMR (PET/MR). Motion-correction algorithms in emrecon use the freely available FAIR toolbox [36], developed in Lübeck [29].
FAIR and emrecon provide the algorithmic foundation for the emission-tomography applications in this project. Previous work with both platforms has resulted in several joint publications by the subproject leaders.
Work programme
The project will implement direct processing of time-continuous data streams (list mode) and time-continuous motion and activity models in emrecon for clinical scanners. This is computationally demanding, so processing will be parallelised on GPUs [59]. Existing algorithms for jointly estimating motion and activity will be implemented on this basis.
FAIR already contains prototype 4D components. These need to be implemented consistently, both methodologically and conceptually, for this project. TP4 will develop a suitable data model and regularisation. The data model must support multiscale and multiresolution strategies—including analysis of optimal spatial and temporal discretisations—to enable effective convexification strategies. FAIR currently provides parametric, quadratic, and hyperelastic regularisation for 2D and 3D vector fields; these methods must be extended to time-dependent vector fields.
An appropriate test environment will be implemented to determine optimal algorithm-specific parameters, including data resolution, data pyramids, and smoothing parameters. Visualisation can use MEDgical, the open clinical visualisation software developed at EIMI.