Annex I - Description of Work FMT-XCT

Transcription

1 SEVENTH FRAMEWORK PROGRAMME THEME [ Health-2007-A] HEALTH : Development of a hybrid imaging system Grant agreement for: Collaborative projects (Small or medium-scale focused research projects with maximum EC contribution of 6,000,000/project). Annex I - Description of Work FMT-XCT Project acronym: Project full title: Hybrid Fluorescence Molecular Tomography (FMT) X-ray Computed Tomography (XCT) method and system Grant agreement no.: Date of preparation of Annex I (latest version): March 6th, 2008 Date of approval of Annex I by Commission: (to be completed by Commission) List of beneficiaries Beneficiary Beneficiary s organization name No. Short Country Date enter Date exit name project project HMGU Germany (Coordinator) Helmholtz Zentrum Muenchen, German Research Center for Environmental Health 2. Commissariat a L Energie Atomique CEA France Foundation for Research and Technology Hellas FORTH Greece University College London UCL U.K Fudacion para la Investigacion Biomedica del Hospital FIHGM Spain 1 48 Gregorio Marañon 6. Universität Zürich UZH Switzerland Verfahren und Apparate der Medizinischen Physik VAMP Germany 1 48

2 Table of contents: PART A A1. Budget breakdown and project summary A.1 Overall budget breakdown for the project 3 A.2 Project summary 4 A.3 List of beneficiaries 5 PART B B1. Concept and objectives, progress beyond state-of-the-art, S/T methodology and work plan B.1.1 Concept and project objective(s) 6 B.1.2 Progress beyond the state of the art 7 B.1.3 S/T methodology and associated work plan 7 B Overall strategy and general description 7 B Timing of work packages and their components 8 B Work package list /overview 9 B Deliverables list 10 B Work package descriptions 13 B Efforts for the full duration of the project 37 B List of milestones and planning of reviews 38 B2. Implementation B.2.1 Management structure and procedures 39 B.2.2 Beneficiaries 41 B.2.3 Consortium as a whole 45 Sub-contracting 47 B.2.4 Resources to be committed 47 B3. Potential impact B.3.1 Strategic impact 50 B.3.2 Plan for the use and dissemination of foreground 51 B4. Ethical issues if applicable 53 B5. Consideration of gender aspects 57 2

3 PART A A.1 Overall budget breakdown for the project 3

4 A.2 Project summary 4

5 A.3 List of beneficiaries Beneficiary Beneficiary s organization name No. 1. (Coordinator) Helmholtz Zentrum Muenchen, German Research Center for Environmental Health 2. Commissariat a L Energie Atomique [ 2 groups :LIME ( Paris) and LETI (Grenoble) ] Short Country Date enter Date exit name project project HMGU Germany 1 48 CEA France Foundation for Research and Technology Hellas FORTH Greece University College London UCL U.K Fudacion para la Investigacion Biomedica del Hospital Gregorio Marañon FIHGM Spain Universität Zürich UZH Switzerland Verfahren und Apparate der Medizinischen Physik VAMP Germany

6 PART B B1. Concept and objectives, progress beyond state-of-the-art, B.1.1 Concept and project objective(s) This proposal offers to develop quantitative Fluorescence Molecular Tomography (FMT) X-ray Computed Tomography (XCT) system, by appropriately advancing optical tomography and X-ray CT imaging methods so that they can be seamlessly integrated into a hybrid imaging system and method. As showcased in this proposal, this approach offers a highly synergistic multi-modality system that will perform in a superior manner than any of constituents if they were to be considered as two stand-alone modalities by: 1. Providing hybrid FMT-XCT images by superimposing fluorescence images of functional genomics, proteomics and physiological responses on X-ray anatomical images to optimally visualize the spatial distribution of various biomarkers. 2. Improving the imaging accuracy and quantification of the optical tomography problem and the performance of the hybrid system by utilizing the anatomical information provided by XCT into the FMT inversion problem. The overall goal of this proposal is to develop quantitative hybrid FMT-XCT technology, engineer the optimal theory and inversion approaches for achieving a highly performing and synergistic system and perform pre-clinical imaging with a view towards clinical translation and therapeutic intervention. Major focus is given to preclinical imaging of breast cancer, as the most probable entry point of this approach to the clinic. The general aims of this proposal are: Aim 1: Aim 2: Aim 3: Aim 4: Aim 5: Aim 6: Aim 7: To research optimal methods for yielding improved XCT contrast and deliver a corresponding XCT design and prototype (Wp2). Optimal methods will be defined as the methods that provide better contrast to noise ratio of target biological soft tissues, i.e. lung vs heart vs adipose tissue. To research and validate new inversion schemes that offer optimal FMT stand-alone imaging performance in free-space FMT implementations (Wp3). Optimal performance is defined as the one that more closely represents the underlying structures of known objects being utilized as gold-standards exactly for testing inversion algorithms. To research and deliver optimal methods for utilizing a-priori information into the FMT inversion code for developing the best FMT performance ever developed (Wp4). Similarly to Aim2, optimal performance is defined as the one that more closely represents the underlying structures of known objects being utilized as gold-standards exactly for testing inversion algorithms. To build a highly performing quantitative FMT-XCT prototype (Wp5). Success will be measured by means of a fully functional system able to acquire fluorescence and X-ray CT images of equal or better quality compared to its stand-alone counterparts. To apply FMT-XCT in-vivo in pre-clinical imaging of mouse models of breast cancer (Wp6). Success would be defined by means of accurately reconstructing the underlying fluorescence bio-distribution as confirmed histologicaly. To research the quantitative ability of FMT-XCT in visualizing treatment in-vivo (Wp7). Success will be defined by means of verifying in-vivo performance against well established post-mortem (in-vitro) laboratory tests obtained at key selected points. Compare FMT-XCT and PET-XCT to quantitatively assess the sensitivity and overall image fidelity achieved by FMT-XCT, using the PET-XCT as a gold standard (Wp8). Success will be 6

7 measured by reporting quantitative metrics on image quality for FMT-XCT and PET-XCT obtained from the same animal model. B.1.2 Progress beyond the state of the art This proposal offers to develop a unique FMT-XCT hybrid system, the likes of which exist nowhere. This is a major starting point of this proposal. Besides the relatively straightforward integration of two different modalities, this proposal offers to significantly advance each of the systems components, well beyond the current state of the art in order to further advance technological European competencies and provide a truly high performing system that utilizes the strengths of XCT and FMT while significantly reducing their weaknesses by the hybrid approach. It is proposed to: 1. Improve XCT contrast differentiation while retaining reasonable dose exposure by dual energy, multiresolution approaches. 2. Advance stand-alone FMT by utilizing novel free-space non-contact approaches operating in the 360-degree projection implementation, offering the best yet performance of stand alone FMT. 3. Advance the knowledge on optimal implementation of anatomical information into the FMT problem by (a) improving XCT contrast in Wp2 to offer advanced segmentation in Wp4 and highly adept annotation of XCT structures with optical attenuation properties in Wp5 and by (b) researching the use of priors that offer no-strong anatomy-function relations as is appropriate for fluorescence biodistribution (Wp4). This advancement can only be achieved herein due to the proposed construction of the first XCT-FMT prototype worldwide. 4. Deliver the best performing optical imaging method by means of incorporating XCT information. 5. Deliver a highly evolved hybrid XCT-FMT to complement XCT with functional and molecular contrast and complement FMT with anatomical contrast and the ability to improve its performance by means of XCT-based attenuation correction. 6. Advance the knowledge into the sensitivity and quantitative ability of the method in resolving functional and molecular signatures in-vivo. While there is no direct basis to compare progress against, since a truly unique system is being developed, comparisons to stand-alone modalities and to PET imaging, as described in the previous paragraph offer direct evaluation metrics by which to monitor work progress. B.1.3 S/T methodology and associated work plan B Overall strategy and general description The work is split into 9 work packages (Wp) as shown in Fig.1. Work package 1 regards the management and co-ordination of the project by partner 1 while Wp 9 considers dissemination and training activities. Each partner has been carefully selected as he brings a unique aspect into this development, while utilizing expertise from other partners. WP2-WP4 build therefore unique XCT and FMT technology and know-how that is then integrated into Figure 1. Schematic of the work-packages involved in the proposal 7

8 one system prototype in WP5. WP6 and WP7 research appropriate imaging strategies for in-vivo imaging, and perform pre-clinical imaging while WP8 compares the FMT-XCT system with a previously developed PET-XCT system to test the hypotheses in this proposal. The effort varies in different work-packages based on the corresponding complexity of each package as summarized in Table 1.3d. B Timing of work packages and their components The timing chart is given according to the objectives (tasks) described in each work-package 8

9 B Work package list /overview Work package list Work package No Work package title Type of activity Lead participant No Personmonths Start month End month WP1 Management MGT WP2 XCT development RTD 2 - LETI WP3 FMT forward problem RTD WP4 FMT inversion with priors RTD WP5 FMT-XCT integration RTD WP6 Breast cancer imaging RTD 2-LIME WP7 Treatment imaging RTD WP8 FMT-XCT vs. PET-XCT RTD WP9 Training & Dissemination OTHER TOTAL 608 9

10 B.1.3.4: Deliverables List List of Deliverables to be submitted for review to EC 1 Del. no. 2 Deliverable name WP no. Lead beneficiary Nature 3 Estimated indicative personmonths Dissemination level Consortium agreement R RE Minutes of the kick off consortium meeting 1.3 Project web site and updates 1.4 Program of the consortium meetings R RE 1 Delivery date 5 (month) O PU,RE 12, 24, R RE 12,24, Half time report R RE Program of the closing meeting R RE Final report R RE XCT design R PP Dual energy prototype P PU Dual energy processing software 2.4 Preliminary technical specification 2.5 Scattered energy measurements 2.6 Comparison of contrast enhancement strategies 2.7 Final technical specification for XCT system O RE R PU R PU R PU R RE Direct inversion algorithm O PU Experimental training set measurements 3.3 Quantitative evaluation of direct inversion with FMT O PU R PU 18 1 In a project which uses Classified information 1 as background or which produces this as foreground the template for the deliverables list in Annex 7 has to be used 2 Deliverable numbers in order of delivery dates: D1 Dn 3 Please indicate the nature of the deliverable using one of the following codes: R = Report, P = Prototype, D = Demonstrator, O = Other 4 Please indicate the dissemination level using one of the following codes: PU = Public PP = Restricted to other programme participants (including the Commission Services) RE = Restricted to a group specified by the consortium (including the Commission Services) CO = Confidential, only for members of the consortium (including the Commission Services) 5 Month in which the deliverables will be available. Month 1 marking the start date of the project, and all delivery dates being relative to this start date. 10

11 data 3.4 Multi-spectral algorithm O RE User friendly software environment 3.6 Quantitative evaluation of direct inversion with hybrid data O RE R PU Inversion algorithms O PU 9, Quantification of algorithmic performance 4.3 Inversion algorithm for estimating optical attenuation R PU O PU Feature extraction algorithm R PU Quantification of algorithmic performance with experimental data R PU User friendly software O RE degree FMT prototype P PU Optimal settings O RE Gantry development P PU FMT-XCT prototype P PU Training data sets O RE Multi-spectral capacity O PU Compute and assign optical attenuation values 5.8 Functional specification of optimal acquisition and operational parameters 5.9 Functional user-friendly operational software 6.1 To develop molecular probes 6.2 Prepare mammary tumor animal models. 6.3 Breed and make available PyMT animal models O RE R RE O RE O PU 9, O RE O RE Develop U87 animal models O RE Study FMT-alone vs. FMT- XCT in resolving tumors 6.6 Study the imaging performance in cancer in animal organs 6.7 Study the overall imaging performance R PU R PU R PU Preparation of HIF O PU 9 11

12 transfected breast cancer cells 7.2 FMT of HIF induction in breast tumors. 7.3 FMT-XCT vs. histological correlates in MDA-MB-231 tumors. 7.4 FMT-XCT vs. histological correlates in PyMt tumors. 7.5 FMT-XCT to resolve differential treatment levels in tumors in-vivo. 7.6 Assessment of antiangiogenic therapy with FMT-XCT 8.1 Construct imaging phantoms. 8.2 Phantom characterization with PET, FMT and XCT and appropriate probes 8.3 Coregister and compare FMT-XCT and PET-XCT. 8.4 Analyze and report on XCT- PET and XCT-FMT performance. 9.1 Dissemination implementation document 9.2 Training curriculum and implementation document R PU R PU R PU R PU R PU O PU O PU O PU R PU R RE R RE Public promotion leaflet O PU Meeting program and book of abstracts for the meeting at month Summary of achievement at Workshop at month Report on technology transfer and IP. 9.7 Report on training and PhD theses achieved R PU R PU R RE R PU 48 TOTAL

13 B Work package descriptions Work Package 1 Work package number 1 Start date or starting event: 1 Work package title Management Activity Type 21 MGT Participant number Person-months per participant: Objectives: WP1 serves as a horizontal workpackage that enables: 1.1. Management of the interaction of the co-ordinator, the Executive Committee and the Advisory Committee in order to design, monitor and optimise the experiments. 1.2 Management of the pre-existing and new intellectual property (IP) and know-how Maintenance of the Consortium Agreement Regular meetings and reports on the scientific and financial progress, ethical and welfare issues. Description of work (possibly broken down into tasks), and role of participants This work-package contains all activities to ensure coordination of the consortium, reporting to the EU, enabling appropriate information flow between participants and the EU. Management of the consortium will require active interaction of the co-ordinator, the Executive Committee (EC) and the Advisory Committee (AC). The co-ordinator in this proposal will be helped by a project manager/public relations person, hired exclusively for management purposes. Partners will also devote time effort for management activity, when participating in executive tasks and in relation to the generation of the deliverables listed herein. The project manager will be further responsible for communicating Executive Committee decisions and directions and actively work towards group networking, executive committee meeting preparation and gathering the necessary research and financial information for the deliverables and reports. The roadmap of WP1 is presented in the Gantt chart. Besides the general management, this WP will ensure the ethical and animal welfare management of the project under direct supervision of the co-ordinator. An ethical expert will be invited to the kick-off and annual meetings to provide advice for the project and to assist in the preparation of the final report including broader science and society implications of the results. Meetings: Regular meetings will be held to present, monitor and evaluate progress and to make decisions for future steps. Work-package (WP) meetings will be organised a month prior to Executive Committee (EC) meetings, thus the WP meeting results will be efficiently translated into consortium-level decisions. A kick-off consortium meeting will be held at the beginning. WP and EC meetings will be held every year (mo 12, 24, 36). Key WP participants may be invited to present on all EC meetings. If need arises, EC video conferencing meetings will be held at mo 6, 18, 30, 42 or ad-hoc. Half-time report preparation at month 24. Progress monitoring, financial corrections and redistribution of WP will be performed at EC meetings at month 24. New scientific and industrial partner evaluation will be also performed in EC meetings. Final report and consortium meeting program will be provided at month 48. All consortium meetings will contain a dedicated EC meeting for reaching executive decisions. Reports: Internal WP reports and a summarized consortium report at month 12, and 36 to monitor financial and research progress. Full reports to the EC at month 24 (half time report) and 48 (final report). Tasks: Task 1.1. Organisation of consortium meetings, at months 0, 12, 24, 36 and 48. Task 1.2. Organization of Executive Committee conferences as needed. Task 1.3. Preparation of the half-time report on the scientific progress and financial status. Task 1.4. Preparation of the final report on the scientific results, new IP and finances.. 13

14 Deliverables (brief description and month of delivery) 1.1 Consortium agreement (mo.1) 1.2 Minutes of the kick off consortium meeting (mo. 1) 1.3 Project web site (mo 1) 1.4 Program of the consortium meetings (mo 12, 24, 36). 1.5 Half-time report for the EC (mo. 24) 1.6 Program of the closing meeting (mo.48). 1.7 Final report for the EC (mo 48). 14

15 Work package 2 Work package number 2 Start date or starting event: 1 Work package title XCT development Activity Type 21 RTD Participant number Person-months per participant: Objectives: 2.1 To design a micro X-ray CT system appropriate for small animal imaging 2.2 To develop the XCT system and implement a dual energy X-ray CT system. 2.3 To optimize delivered dose using a multi-resolution CT approach. 2.4 To research and minimize possible interference of X-rays with optical components. 2.5 To research the contrast between organs and tissues achieved by the dual energy method 2.6 To research the contrast between organs and tissues achieved by contrast agents. 2.7 To provide an optimal XCT design to be incorporated with the free-space FMT system in WP5. Description of work Several designs and commercially available systems exist for X-ray CT of small animals. The need to implement a separate work package on X-ray CT development stems from the diverse needs of the hybrid approach to produce an XCT design that is not only appropriate for small animal imaging but also: 1) provides adequate accommodation of the optical components, 2) eliminates X-ray interference with optical components 3) offers improved contrast between organs as is important for the optimal utilization of X-ray CT information as priors in the FMT inversion procedure as explained and performed in WP4. Overall the integration of XCT experts and XCT development in this proposal allows high flexibility in changing specifications when issues of interference between optics and XCT components arise or functional specifications need to be changed. It creates a functional team that can respond to both XCT and FMT challenges and yield an optimal system and method. 2.1 X-ray CT design. While the partners have already discussed appropriate XCT designs, the design finalization will begin at the kick-off meeting, based on functional specifications directed by all partners in a special design meeting, which will be part of the kick-off meeting. Partners CEA-LETI, FIHGM and VAMP have extensive experience with the development of XCT small animal systems. Compatibility with optical components and small animal imaging considerations will be directed by the other partners and will be taken into account. The proposed system will employ a microfocus X-ray tube, and a solid state digital X-ray image sensor for the image acquisition; implementing cone-beam geometry. A computer will operate the X-ray source, an X-ray stopping shutter (to minimize X-ray dose per study) and the frame grabber to acquire the projections. Administered dose will be within the European and LETI guidelines for small animal XCT as applied to similar commercial micro-ct systems, ensured further by VAMP. The system will be tightly enclosed into a 4mm lead sheet chamber to minimize stay radiation. The radiation emitted outside the box will be measured upon construction and regularly thereafter during measurements to ensure the effectiveness of the shielding constructed. Inversion will be based a cone-beam reconstruction package available to FIHGM, implementing a modified Feldkamp-Davis-Kress reconstruction algorithm for circular orbit cone beam geometries [34]. To further improve on artifacts caused by X-ray scattering when using flat panel detectors a scatter reconstruction approach for Cone Beam CT, developed at CEA-LETI, will be employed [35] for small animal CT imaging and for the dual energy context in X-ray CT dual-energy development Dual energy approaches have the potential to improve the contrast between main organs by dual energy decomposition in (ρ,z) components. This is necessary in this proposal for optimal use of X-ray priors in the FMT problem, as described in WP4, but can more generally also improve the diagnostic and utilization potential of XCT and is also clinically investigated. CEA-LETI has been involved in several projects of dual energy XCT decomposition of bone from soft tissue in clinical bone densitometry and VAMP is actively looking at the utility of these methods as well. CEA-LETI will develop a dual energy system prototype based on the design of 2.1, implemented for in the horizontal geometry (i.e. non-gantry approach) to accelerate the 15

16 development, since the objective here is not the highest possible imaging performance but the comparison of different alternatives for recommending an appropriate design in the gantry-based system developed in Wp5. In this case the mouse is vertically rotated, after immobilized in a special holder that does not allow movement due to rotation. The consortium will then investigate also through motility between partners, whether this approach is useful for differentiating between soft-tissues, over contrast agents as per tasks 2.4 and 2.5. The particular implementation will utilize a single source that is emitting at different energies by applying different voltages and appropriate filtration. For low energy, the source will be filtered with a material presenting a K- edge around the energy of interest and utilizing a voltage higher than the K-edge. For high energy the spectrum is hardened using higher voltage and filtering with material such as copper to suppress lower energies. For animal imaging it will be possible to utilize lower energies, compared to dual-energy clinical approaches. These energies will be selected to yield maximum contrast between the organs while ensuring appropriate signal to noise ratio. Prior experience indicates that appropriate energies would perhaps peak at 70KeV and 30KeV but the exact energies and source parameters will be optimised based on X-ray attenuation model software and will be adjusted and confirmed experimentally. Since strong overlap is expected between the two spectra the set of data obtained at the two energies will have significant cross-talk. Dual-energy image analysis is therefore based on a calibration process which allows to decompose each voxel over a basis of material. This calibration is performed offline (without patient or animal) and provides attenuation measurements for different thicknesses of combination of 2 materials: a low density and higher density material, for example lucite or hydroxyapatie mixed at different concentrations with a resin to mimic different soft tissues (and bone). The measured projections are then converted in equivalent basis materials thickness decomposition by interpolation of the calibration data. After reconstruction of these projections converted in basis materials, the X-ray attenuation of each reconstructed voxel can be described as the X-ray attenuation of a linear combination of the basis material. The density and the atomic number of each voxel can then be approximated by the density and the atomic number of the material combination corresponding to this voxel. 2.3 Optimization of delivered dose. In order to minimize the dose delivered to the animal in dual energy approaches, a multi-resolution CT approach will be considered. Since the FMT forward problem is solved at the resolution level of a mean free-transport path, i.e. ~0.5mm (or larger) for small animal imaging, X-ray 3D differentiation of soft tissues in the animal does not necessarily require the highest of XCT resolution. Therefore, dual-energy based tissue contrast will be based on a low resolution scan of the second energy source and subsequent data interpolation on the first set of acquisition. LETI has already progressed in developing approaches for incomplete data reconstruction based on the RM Lewitt approach. The CT acquisition is performed with a smooth breathing gating to limit the blurring, while for high resolution, VAMP has available high speed scans that can further improve imaging performance. 2.4 Minimization of X-ray interference. Through motility activity, VAMP and HMGU will visit CEA-LETI, upon prototype construction and perform measurements of scattered X-ray energy in the various directions of the XCT prototype, when using phantoms and animals, in order to design appropriate X-ray shielding and identify optimal areas where shielded optical components can be placed. This information will confirm simulations performed by VAMP and by CEA-LETI and it is important in anticipation of the developments in WP5. The participation of HMGU and UZH will contribute to optimal consideration in terms of optical components and animal imaging. 2.5 Dual-energy contrast. Currently it is largely unknown whether dual energy approaches can yield appropriate tissue and organ discrimination, as per methods in Task 2.2 and the need for image segmentation in WP4. It is also unknown which energies will yield optimal contrast and how this compares to the use of XCT iodinebased contrast agents. For this reason, after appropriate simulations and phantom calibrations performed by CEA- LETI in task 2.2 representative animal species (nude mice (n=5) and Balb/c mice (n=5); see also Section 4) provided by CEA-LIME will be imaged in order to obtain dual energy images of head, torso and abdomen. The resulting, processed images resulting from the dual energy acquisition (see 2.2), will be then analyzed and compared to the single energy image in terms of optimal contrast achieved. Quantitative image analysis will be performed by Prof. Englmeier at HMGU to determine whether there are contrast benefits in the dual wavelength approach for image segmentation. 2.6 Use of X-ray contrast agents. In parallel to the dual energy approach, FIHGM and CEA-LETI will investigate through motility the contrast achieved by the use of iodine-based contrast agents. This will be achieved by imaging the animals at Task 2.5 also after the administration of XCT contrast agents, immediately after the 16

17 imaging session in Task 2.5, while the animals are immobilized in the imaging setup, by performing a second single energy acquisition. All animals will be euthanized immediately after this second imaging session. The relative ability to differentiate between the heart, lung, muscle and other organs will be explicitly determined by characterizing the contrast achieved herein and as it relates to the corresponding results in Task 2.5. UZH will aid in identifying different structures as it relates to high quality MRI images of the same species, that will be also provided by UZH. Image analysis will be similarly performed by the group of Prof. Englmeier at HMGU - IBMI. The optimal method that maximizes organ delineation under comparable administered dose will be selected for WP5. Special considerations of defining optical parameters are described in WP To provide an optimal XCT design An important milestone of this work package is to propose an optimal XCT design that will be implemented in a rotational, industrial-grade gantry in WP5 and an appropriate technology recommendation (single-energy, dual energy and contrast agent enhancement) for optimizing XCT imaging performance. The decision of an optimal design will be reached in the Executive Committee meeting in month 18 and formally presented in the Consortium meeting in month 24 and in the EU report. Risk This proposal examines several alternatives to minimize risk. The competency of the partners to develop laboratory prototypes (CEA-LETI, FIHGM) and commercial systems guarantees that several of the practical implementations details will be addresses with multiple skills and capacities. High motility between partners and web-conferencing as described under management will further ensure the high information flow and timely execution of the proposed tasks. Training The tasks offer significant training opportunities as they will develop unique knowledge in terms of advanced XCT use and contrast differentiation This information will be made available in the first consortium meeting held for this purpose in Grenoble, and will include tour of the facilities and a demonstration of the system prototype to the consortium. Importantly through motility activity the FIHGM and VAMP will actively participate in the prototype development and measurements, whereas CEA-LIME will visit and participate in animal imaging. Milestones This Wp is involved in a significant milestone, i.e the selection of an appropriate XCT technology to use for integration in the FMT-XCT prototype. A preliminary recommendation and decision will be performed in month 18 by the executive committee after circulation of an internal report and results, and the technology and decision will be officially presented in the meeting of month 24. Deliverables 2.1 An optimal design for micro X-ray CT system that can facilitate optics (mo.1). 2.2 A functional prototype for dual energy cone beam XCT (mo. 15) 2.3 Calibrated, dual energy processing software (mo.15) 2.4 Preliminary technical specification for SCT design to be implemented with the hybrid system (mo 18) 2.5 Measurements of scattered X-ray energy (mo 21) 2.6 Identification of optimal contrast-enhancing strategy mo 24) 2.7 Optimized system design summary, technical specification and imaging protocol for integration in the hybrid FMT-XCT in WP5 (mo.24). 17

18 Work package 3 Work package number 3 Start date or starting event: 1 Work package title Theory for 360 degree FMT. Activity Type 21 RTD Participant number Person-months per participant: Objectives: 3.1. To implement direct inversion based on boundary removal method for media with arbitrary boundaries. 3.2 To research optimal direct inversion approaches with simulations and experimental data To compare the direct inversion performance with conventional, previously developed FMT inversion methods. 3.4 To incorporate algorithms for multi-spectral imaging. 3.5 To develop user-friendly software for inversion of XCT FMT data based on direct inversion approaches. 3.5 To invert training data from acquired from the FMT-XCT system for algorithmic finalization. Description of work The proposed FMT-XCT system implies the development of an optical tomography method based on complete projection tomography over 360-degree angles. To accomplish high imaging performance it is important to move away from fiber-based systems or the use of matching fluids (as common in older generation systems) and utilize direct measurement of photons, at multiple projections, using CCD cameras. Theoretical approaches developed by partners HMGU-IBMI and FORTH-IESL now enable this new generation of systems to be implemented. 360-degree free-space systems utilize hardware similar by that used by XCT and so they can seamlessly be integrated in an XCT platform. They are further expected to deliver the best FMT performance compared to fiber-systems, fluidsystems or limited view angle systems (for example slab geometry systems) that have been developed in the past due to the superior data quality and data size acquired. It has been shown that high spatial sampling improves imaging performance and resolution [36] [37], (despite original notions that only a few measurements suffice for optical tomography). With this recent progress however comes a new challenge. Since 360-degree systems acquire a significant larger number of data (~10 8 measurements) compared to fiber based systems ( measurements) or slab geometry systems (~10 4 measurements), the computational requirements are tremendously increased. Task 3.1 Direct inversion: A key component in the success of the proposed hybrid system and a major objective therefore behind this work-package is the implementation of direct inversion algorithms that retain robustness while utilizing a large data-sets and achieving realistic reconstruction times (5-15 minutes). Direct inversion obtains an expression for the fluorophore concentration N given the measurements U, in the form. { U} N (r) = g (3) where g is the direct inversion formula. This approach does not involve matrix inversion and benefits from large datasets since it is based in Fourier-Laplace decomposition [See work by John Schotland, for example [38, 39]]. The reason these approaches have not been implemented before is mainly due to the fact that direct inversion formulas have been found only for 1) simple geometries such as the infinite, slab or cylinder case and 2) for homogenous media. It is a major contribution of WP3 therefore, and a novelty in this proposal, that a particular inversion scheme, termed boundary removal, developed recently by partner FORTH-IESL and HMGU-IBMI, will be employed to allow the application of direct inversion methods with measurement obtained from complex boundaries [38, 40]. This scheme is further combined with a certain dual-wavelength normalization method (see Ref [41] ), which has been shown to perform accurately even when imaging at highly optically heterogeneous media. Collectively, the approach is expected to yield the most accurate and highly performing inversion method developed yet for FMT and combined with the superior data set collected with the FMT-XCT system developed herein. More practical details of implementation follow: The forward problem utilized is a solution of the diffusion equation based on the normalized Born approximation [41], which is independent of instrument gain factors and yields accuracy even at high degree of background optical heterogeneity [42]. In the past, fast modeling of the boundaries by the Kirchoff approximation Ref. [15, 19] or higher order approximations [18] and the propagation of 18

19 photon signals from this surface to a CCD camera were employed due to their computational efficiency in slab systems. However this scheme is now replaced by the boundary removal scheme. In parallel partner UCL (Simon Arridge) has previously developed the main theoretical mainframe for modeling photon propagation in tissues in the forward sense and will contribute here code developments associated with the forward modelling of photon propagation in tissues. HMGU and FORTH have also previously developed methods (also under FP6) to measure and introduce the mouse surface into the reconstruction code (see also fig.3,4) in order to calculate an accurate forward model for optical tomography. While the above description generally referred to a single wavelength, multispectral measurements can be also be described by Eq.1 and inverted under one inversion scheme for increased accuracy. In summary the key innovations in task 3.1 are: 1. Algorithmic developments in accounting for boundary effects using a recently developed boundary removal method. 2. Time efficient reconstruction of photon propagation in tissues using direct inversion formulas operating in optically heterogeneous media bounded by arbitrary free surfaces that can be used for highly-performing stand-alone FMT as per Concept 1.1.a 1 (page 3). 3.2 Experimental optimization. Simulated data generated by UCL, based on anatomical maps provided by CEA_LETI will be generated using the TOAST software (i.e. a public-domain finite element simulation of photon propagation in tissues developed by UCL), In addition experimental data on phantoms and controlled animal experiments will be generated by FORTH on the stand-alone system already developed in Crete. Experimental data sets will be obtained on phantoms, build by FIGHM together with HMGU (see WP8) and animal models (see Section 4 for animal numbers) provided by partner UZH. These data sets will be utilized to test direct inversion approaches in practical settings that relate to the final FMT-XCT prototype. Finally, upon completion of the FMT-XCT prototype at HMGU, free-space FMT data will become also available as described in WP5, which will be explicitly analyzed by partner FORTH. The overall goal of task 3.2 is to find optimal inversion settings (optimal number of data, optimal regularization parameters, as relate to the signal to noise ration contained in the data). These tasks have been previously analytical addressed by singular value decomposition approaches and will be explicitly studied herein in relation to direct inversion performance. 3.3 Direct inversion vs. conventional FMT inversion. To quantitatively assess the developments in Task 3.2, FORTH and UCL will study the performance of the direct inversion method in relation to : a. Old generation linear FMT inversion code developed by partners HMGU-IBMI and FORTH-IESL b. Finite element-based non-linear inversions, under the TOAST software, performed by partner UCL. Both approaches 3.3 a,b have been extensively tested in the past and can serve as a benchmark for FMT performance (albeit the impractical inversion times). In particular 3.3.b is based on a two-step approach, where the background attenuation (absorption and scattering) is first reconstructed and used to calculate more accurate weights for the fluorescence problem. This method, available with partner UCL, will be considered as the gold standard in simulations and phantom measurements, even if more computationally intensive, in order to evaluate the performance of the direct inversion approach, by comparison of results between the two methods. Training data sets obtained from controlled phantoms and animal experiments post-mortem (see Wp5 and Wp8) will provide a data set library to test different methods. The overall method that will give the best imaging performance while maintaining practical inversion times will be then utilized for subsequent in-vivo data analysis. Inversion accuracy and performance will be evaluated in terms of imaging accuracy against the known manufactured optical phantoms (as per Task 3.2) and it terms of the resolution and sensitivity achieved Multi-spectral imaging. As evident in Wp6 and Wp7, the ability to perform imaging of multiple fluorochromes, simultaneously administered in-vivo can facilitate the study of complex pathways, or improve the imaging accuracy (see for example Task 6.3 in Wp6. FORTH has been developing multi-spectral approaches to 1) remove autofluorescence background in fluorescence measurements owing to tissue intrinsic fluorochromes and 2) the ability to 19

20 accurately reconstruct multiple fluorochromes when the emission spectra are partly overlapping. This technique is based on the recording of tomographic data in multiple spectral regions with excitation light of different wavelengths and on the application of linear unmixing algorithms for targeting multiple fluorescent probes. These methods are available to FORTH as part of other funding (also FP6 funding) for imaging at the visible, but will be adapted herein in for imaging at the far-red and near-infrared (600nm 900nm) to enable accurate imaging. We note that in the nm tissue auto-fluorescence is significantly reduced compared to the visible and these methods are expected to work even better than in the visible, after appropriate determination of the relative strengths of the nearinfrared fluorochromes employed in commercially available probes (see table I) or in developments in Wp Software development. Important in the technique dissemination and training activities, is the development of user-friendly software that can be accessible by users and not only developers. The software, developed by partner FORTH, will have easy to handle inputs and correspondingly straightforward visual outputs to simply guide a user through the inversion process and allow for at least some simple visualization tasks in order to easily view and quantify the reconstructed images. Risk and alternatives. The tasks of the work package follow highly inventive developments that present unknown development challenges as to their exact implementation. The team assembled however has major contributions in the developments of these methods and are very well skilled in programming computational methods for tomography. In addition accurate validation against established gold standards are contemplated to ensure that a highly performing inversion method will be developed. Furthermore the utilization of experimental data will enable the development of a practical method appropriate for in-vivo imaging. Training. This WP will enable training of consortium scientists in methods for FMT inversion methods by appropriate participation in the workshop at month 42, and interaction also through motility of the UZH and UCL groups and other group participation as well. Milestones. Important decision points in this work-package concern the development and final decision on an inversion algorithm with fundamental superior characteristics compared to the current state of the art. After validation with experimental data from phantoms and in-vivo, the partners involved will direct optimal algorithms, implementation settings and a functional inversion code for stand-alone inversion. Deliverables (brief description and month of delivery) 3.1. A direct inversion algorithm for fluorescence implementing boundary removal principles (mo. 9) Experimental measurements to serve as a training set for algorithmic optimization (mo. 12) 3.3. Quantitative evaluation of direct inversion performance vs. established FMT inversion methods (mo.18) 3.4 Functional code of multi-spectral algorithm for simultaneously resolving multiple fluorochromes (mo. 21) Software that implements these algorithms in a user-friendly environment (mo. 24). 3.6 Optimization and quantitative evaluation of direct inversion performance with hybrid XCT-FMT data (mo. 32) 20

21 Work package 4 Work package number 4 Start date or starting event: 1 Work package title FMT inversion with image priors Activity Type 21 RTD Participant number Person-months per participant: Objectives: 4.1 To develop FMT inversion utilizing XCT image priors without strong anatomy function correlations. 4.2 To incorporate XCT image segmentation into the FMT code. 4.3 To calculate spatially varying optical attenuation in tissues in-vivo To develop FMT inversion based on simultaneous XCT segmentation and classification. 4.5 To quantitatively examine optimal inversion methods based on experimental data. 4.6 To develop user-friendly software for inversion of XCT FMT data based on a-priori inversion. Description of work (possibly broken down into tasks), and role of participants Utilizing XCT information for improving the FMT imaging performance is a key development herein and a cornerstone towards a highly synergistic hybrid system. Many of the partners are very experienced in the use of image priors in inversion problems but also identify that the use of image priors may lead to biased solutions. Therefore the consortium focuses here on developing methods, beyond the current state of the art, that improve the accuracy of the FMT problem without biasing the solution. We will explore three inversion methods: 1. the use of anatomical priors without strong anatomy-function correlation. 2. the use of prior segmentation and classification techniques to identify tissue classes 3. the use of the methods developed in Wp3 (without priors) to obtained truly unbiased solutions for comparisons (ANY investigation with priors herein implies also the of the unbiased solutions in Wp3 as means of internal reference). In order to build more accurate forward problems, and to make the Bayesian framework feasible, it is necessary to estimate optical parameters. While this is experimentally achieved in WP5, many of the methodological components are enabled herein as explicitly described. Task 4.1 Inversion based on priors. The FMT inversion method (as also described in Wp3) can be written as the minimization of a cost function that in the generic sense can be written as: 2 C( x) = y Wx + Q( x) (1) where y is the vector of measurements for different source detector pairs (in this case fluorescence over emission measurements), x is a vector of unknown fluorescence concentrations in different voxel elements in the medium (after discretization) and W is the weight matrix, which maps the space of unknowns to the space of measurements. The function Q( x ) is a penalty function which in conventional implementations is given under the Tikhonov 2 regulation i.e. Q( x ) = λ x. For a simplistic use with image priors (in this case XCT priors but also potentially MR or ultrasound priors), Q( x ) can be re-written as M 2 ( xi a ( ) ) k i Q( x) = λ 1 exp 2 (2) i= 1 2σ k ( i) where M is the total number of voxels used for the reconstruction, ak ( i) is the average expected optical property of each organ or structure k seen on segmented X-ray CT images. σ k ( i) is the standard deviation associated with each of the average optical property assigned. k( i ) is an index indicating which organs or structure each voxel i belongs to. In essence we can define limits of fluorescence expected from different organs and mark our confidence at these values using the appropriate standard deviation. One of the drawbacks of the above method is the assumption that there is a simple relation between the pixel or region values defined in the auxiliary modality and those in the optical image. This is by no means certain to be true, and if known is likely to be subject to considerable estimation errors. For this reason Partner UCL considers herein three alternative methods for incorporation of priors 21

22 developed with Partner FORTH. Incommensurate Cross-Modality Priors It is therefore a particular contribution and novelty in this proposal that partner UCL will therefore develop methods for using XCT based priors when there is no simple relationship between the numerical values in the optical reconstruction and the anatomical modality, as is the case between fluorescence contrast and XCT density. We term this an incommensurate prior. Our approach will be to treat the mapping between values in one modality and the other as a set of hidden variables which must be estimated together with the reconstructed image. The effect of this will be to estimate a reconstructed image which is most likely the prior image in an information theoretic sense. The use of information theoretic metrics as a constraint in image reconstruction has been employed in only relatively few cases [chen2005a] and never to our knowledge in a nonlinear problem. Hierarchical Bayesian framework. Hierarchical Bayesian methods, investigated by UCL and FORTH make no assumptions on strong anatomy-function correlations. This approach is also particularly appropriate for fluorescence imaging, since the anatomical images provide limited information regarding the underlying fluorescence activity. The method derives a probability map of the expected fluorescence activity for different tissue types, i.e. each organ/tissue is assigned a probability between 0 and 1 indicating the likelihood of fluorescence emission in that region. These maps are then used i) to impose discontinuities and ii) to constrain estimates with physiologically meaningful values in the fluorescence images. Then a hierarchical Bayesian approach will be used to incorporate potential inaccuracies in the prior information such as mismatch between the anatomical and optical edges and true optical coefficients and average coefficients available in the literature. Object based reconstruction. UCL will represent shape implicitly using the level-set method. Here a function of one higher dimension is used whose zero set defines the boundaries of objects. The advantage of the level set method is that the topology of the segmented structures does not need to be known. In pilot work, we have demonstrated that direct reconstruction of the level set function is tractable using two simultaneous level-set functions to reconstruct absorption and scattering independently. We will extend this work to deal with multiple disjoint objects, using multiple level set functions following. Task 4.2 XCT segmentation. FIHGM, CEA and HMGU will investigate methods for XCT segmentation based on the organ differentiation methods described in Wp2. In particular dual energy images obtained by CEA will be examined in parallel to iodine-enhanced images as described in Wp2 by FIHGM and HMGU. Dual energy images have the potential to enhance the contrast between organs. Blood-pooling iodine-based agents reveal images of vascularization, i.e. a bio-distribution of blood, which is the major absorber of light in the wavelengths of interest herein. Both methods therefore attain significant advantages in improving differentiation of structures on the XCT images with the goal of providing improved a-priori information to the FMT inversion. Image analysis will be based on commercially available software and also segmentation tools developed in HMGU for micro-ct image segmentation. In particular Prof. Englemeier (HMGU) has developed methods based on active contours, which allow robust segmentation in images. The creation of the contour is not only based on adaptation to image features like edges, but also on the preservation of the contour smoothness. By assigning energies to the grade of adaptation and smoothness and then minimizing the total energy yields finding of optimal contours. A global optimal contour can be found by dynamic programming leading to automatic segmantation. This method has been so far applied to the more straightforward segmentation of the bone, (see Fig.Wp4-1) but it will be applied herein to contrast enhanced XCT images from Wp2 to identify more complex structures. Fig Wp4-1: Automatic segmentation of the femur and the knee joint and labelling of the different bone structures (two left images) and cortex thickness measurement (two rightmost images) using active contours. Similar performance is expected on segmenting contrastenhanced XCT images from Wp2 for accurate use as image priors. Task 4.3 Determination of optical properties. Utilization of XCT images as a-priori information requires assumptions on the underlying optical property distribution based on an anatomical map or assumptions on the regularization of the hierarchical Bayesian methods. Therefore the use of XCT a-priori information, derived in Task 4.2 can be first used to yield a map of optical property variation in-vivo. This information can be used to derive an optical property atlas to be used with all animals of a species, or on a per animal case as described in Wp5. Here, UCL together with FORTH will develop an algorithmic inversion, based on the mainframe described in Task 4.1 to 22

23 reconstruct optical properties (not fluorescence). This is a straightforward application of Eq.1, where the vector y is not a normalized fluorescence measurement but a measurement obtained just by illuminating a mouse by using light at a wavelength of interest. Partner 4 has been a major leader and inventor of these methods and this is a low risk step, resulting in accurate assignment of optical properties in the various structures seen in XCT. These XCT-based optical maps can then be used to contract more accurate forward problems of photon propagation, to improve the prior approaches considered in Task 4.1 or to directly resolve tissue intrinsic bio-markers such as haemoglobin, which is of interest to partner UZH and FORTH and is examined in Wp7. Task 4.4 : Combining Reconstruction with Segmentation and Classification. Alternatively, UCL and FORTH will develop methods that combine an estimate of the optical properties of the tissue and segmentation techniques under a single inversion step. The hypothesis is that since the final segmentation or classification is of lower dimension than the full reconstruction, the estimation problem may be better posed. Pixel based classification into tissue classes. We assume there are a fixed number of tissue classes present. We will develop a method to directly estimate the statistics of the optical properties of each image class, and to classify each voxel in the reconstructed image to one or the other. We will develop an iterative method based on the EM algorithm that alternately (i) estimates the posterior distribution for tissue class at each pixel and (ii) reestimates the tissue class statistics (mean and variance). Global classification into tissue classes. We will incorporate tissue classification into a Markov Random Field (MRF) based framework. This introduces a local smoothness constraint into the classifications that favours few continuous regions. Inference in the MRF will be performed using graph cuts (the push-relabel algorithm) or tree re-weighted message passing [kolmogorov2004]. An approximation to the posterior distribution of the pixels around the MAP solution will be used at each stage. Task 4.5 Experimental validation. UCL will utilize simulated data in basic algorithmic development in Task 4.1 To further optimize and confirm these methods, experiments on animals will be performed in-vivo as per Wp5 Wp7 to obtain realistic optical and X-ray heterogeneity and attenuation and controlled anatomical and fluorescence contrast. Optical attenuation and fluorescence values will be performed ex-vivo immediately following in-vivo imaging after surgically removing organs after appropriate suturing to minimize bleeding, i.e. loss of the major absorber haemoglobin by HMGU. This data set will be used as the gold standard in confirming results in Wp4. In addition to fluorescence, attenuation measurements performed at 670nm, 750nm and 780nm and at different body parts will be used to determine the attenuation of lung, liver, heart, bone, brain, intestine and other tissues as identified on coregistered CT. Task 4.6 Software. Similarly to WP3, important in the technique dissemination and training activities, is the development of user-friendly software that can be accessible by users and not only developers. The software will have easy to handle inputs, XCT segmentation tools, and correspondingly straightforward visual outputs to simply guide a user through the inversion process and allow for at least some simple visualization tasks in order to easily view and quantify the reconstructed images. A significant milestone herein is the selection of the most appropriate inversion methods using priors, from the alternatives described in Task 4.1 and 4.4 as well as the selection of optimal inversion approaches for optical property reconstruction. Role of participants and training. UCL will perform algorithmic developments needed for inversion with priors. FORTH will share forward models from Wp3 and CEA-LETI will provide partners with XCT data sets (single energy, dual energy and contrast enhanced) from Wp2. FIHGH and HMGU will independently examine the XCT method that achieves optimal contrast and HMGU will develop segmentation methods for enabling the use of XCT structures as priors. UCL together with FORTH will develop methods for optical attenuation determination in datasets obtained from FORTH and HMGU from animal models provided by UZH. Data exchange and active participation in this multi-disciplinary development from multiple partners fosters cross-disciplinary training further facilitated in consortium meetings, demonstration and workshop in month 42. Deliverables 4.1. Inversion algorithms for improving reconstruction (mo. 9, 15) 4.2 Quantification of algorithmic performance with simulated data (mo. 21) Inversion algorithm to determine the optical attenuation of different organs/tissues (mo.24) 4.4. An automatic feature extraction algorithm (mo.24) Quantification of algorithmic performance with experimental data (mo.32). 4.6 User-friendly software that implements these algorithms in a user-friendly environment (mo.36). 23

24 Work package 5 Work package number 5 Start date or starting event: 1 Work package title FMT-XCT integration Activity Type 21 RTD Participant number Person-months per participant: Objectives: 5.1 To develop a fully functional multi-spectral XCT FMT prototype and minimize XCT and FMT interference. 5.2 To integrate algorithmic developments from Wp2, Wp3 and Wp4 and control software operating the hardware components. 5.3 To acquire training data sets and optimize imaging settings. 5.4 To provide optical attenuation maps corresponding to the collected XCT images. Description of work To achieve superior imaging performance the implementation of a fundamentally novel scanner based on 1) complete projection (360 0 ) tomography, 2) non-contact approaches (no fiber-tissue coupling) and 3) free-space photon propagation (no matching fluids) is necessary. We will further integrate this system with X-ray CT to achieve concurrent high-resolution anatomical imaging. Although a commercial X-ray CT system could be employed, developing the hardware around the optical system maintains high FMT performance, optimal tissue contrast, high flexibility in X-ray utilization and cost efficiency, without compromising X-ray imaging performance,,as explained in WP2. Several hardware components concerning the illumination, detection, minimization of optical and X-ray interference, dynamic range etc need to be explicitly researched herein, as briefly described in task 5.1. Task 5.1 FMT-XCT prototype Fig.Wp5.1 depicts a simplified drawing of the proposed system. A simplified prototype on a rotational motor stage will be developed in the first 12 months to serve as a pre-platform for the FMT-XCT integration, in order to optimize optical imaging parameters in the vertical geometry. Experience build by Partner 1 and 3 with horizontal free-space prototypes will be invaluable in this development. An industrial grade gantry will be provided by partner VAMP in month 24. VAMP will modify the gantry according to the design considerations and concerns from WP2, as it relates to component arrangement and special shielding for protecting the optical components from scattered X-rays. The gantry will be provided enclosed in a lead compartment for safe operation. The exact XCT components will be provided by VAMP in cooperation CEA-LETI according to the specifications derived in WP2. Partner HMGU will build the optical prototype using a laser light from a tunable CW Ti:Saph laser (690nm-870nm, mw delivered) complemented by a few laser diodes for covering the complete far-red/nir spectrum will serve as the light source for multispectral imaging. The laser light, after appropriate attenuation is directed through a 1x2 optical switch (not shown) to 1) an illumination branch that expands the laser light on the animal Figure WP5.1: Simplified drawing of proposed FMT-XCT system using noncontact measurements and geometry. surface for front-illumination imaging and 2) to a fiber that is placed on the other side of the camera to implement the equivalent of fan beam geometry for FMT. The light beam is focused on the animal surface using a low numerical aperture lens so that the beam is always in focus during rotation. The animal is placed onto a quarter cylinder transparent glass window AR-coated which is mounted on a linear motorized stage for translation along z. The bed is mounted on both edges (not shown) to ensure accurate, reproducible and vibration-less movement. Signals propagated through the animal are collected via a highly sensitive CCD camera (and appropriate filters mounted on a rotating wheel) placed on the opposite side of the transillumination light spot. More viewing angles using mirrors rotating with the gantry or multiple cameras can be covered, however we have found in preliminary SVA analysis studies that this geometry is preferred for imaging by yielding an adequate data set while maintaining optimal 24