31.12.2014 Views

researResearch - Télécom Bretagne

researResearch - Télécom Bretagne

researResearch - Télécom Bretagne

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

esearc<br />

<strong>researResearch</strong><br />

172<br />

(classic 3D imaging but also 2D "temporal"<br />

corresponding to a double synchronisation on<br />

respiration and heartbeat), (ii). evaluation of a<br />

local correction approach and (iii). Improvement<br />

of the signal to noise ratio and of the resolution of<br />

individual images synchronised at various<br />

instants of the respiratory cycle (application of a<br />

super-resolution algorithm).<br />

Multi-resolution approach in multi-modality<br />

imaging<br />

The advantages of this approach compared to the<br />

previous ones are multiple. The most important<br />

are avoiding the segmentation stages of the<br />

concerned structures in the two images before<br />

applying correction, and generating whole<br />

corrected images (not simply corrected values<br />

for concerned regions). However, in this<br />

approach the liaison model between the lowresolution<br />

details of the image and the<br />

anatomical image is global, the average is<br />

calculated for the whole image and is applied<br />

throughout. This global approach is likely to<br />

introduce artefacts into the emission images<br />

when there is not a good correspondence<br />

between the functional and the anatomical data.<br />

Two approaches have been developed to resolve<br />

these limitations. The first is based on the use of<br />

a local model and the second on the development<br />

of algorithms which use no anatomical<br />

information<br />

Segmentation of functional volumes<br />

Mathieu Hatt defended his PhD thesis. A patent<br />

was taken out in September 2008 for an<br />

algorithm developed during his thesis. To<br />

validate the universal aspect and the robustness<br />

of the proposed methodology a database of<br />

images acquired from different hospitals and<br />

different machines was used. The clinical impact<br />

of this approach will be evaluated in the ANR<br />

(Emergence TEC) “SIFR” project.<br />

Integration of PET/scan imagery in radiotherapy<br />

We began our activity in this domain with a<br />

generic model of respiratory movement. From<br />

TDM 4D data and using a technique of<br />

deformable registration we propose a method<br />

allowing to construct an initial spatio-temporal<br />

model of the breathing thorax, for each patient.<br />

Such a model could remain simple, taking into<br />

consideration only one respiratory cycle. We<br />

added a first improvement, which consisted of<br />

globally modelling the trajectories in function of<br />

the amplitude of the external respiratory signal.<br />

This first approach allowed to obtain continuous<br />

spatio-temporal data following an irregular<br />

respiratory cycle. Secondly it would be<br />

interesting to add a parameter representing the<br />

non-regularity of the cycle, which would allow to<br />

define a more generic model controlled by a<br />

series of temporal signals obtained during the<br />

session in the treatment room. To do this we<br />

could, for example, use a spirometer, giving<br />

respiratory volume, an image of the surface of<br />

the thorax coming from optic sensors, X-ray<br />

images or new systems of detection TOF which<br />

we are currently evaluating. It means going<br />

towards a multi-dimensional model allowing to<br />

analyse a posteriori the temporal conditions of<br />

irradiation.<br />

2. Theme 2<br />

Team Indexation, tracability and integrity<br />

of multi-media information<br />

Work developed in this theme essentially<br />

concerns generating and handling traces of<br />

medical information for indexing and content<br />

search, especially in image databases (Content<br />

Based Image Retrieval), in order to help with<br />

diagnosis and decision making, and also to<br />

ensure the security of information and guarantee<br />

the reliability of transmitted data.<br />

Work in CBIR has evolved since 2007 towards<br />

search of entire patient records, containing both<br />

images and text documents (medical records and<br />

administrative papers). It led in 2008 to the<br />

development of new methods, in automatic<br />

digital image characterisation [15], and adapting<br />

data search methods to CBFR (Content Based<br />

File Retrieval) [21] and knowledge extraction<br />

[22][3]. The methods used gave results of 82%<br />

(Brest University Hospital diabetics retinopathies<br />

database) to 92% (DDSM-Massachusetts General<br />

Hospital database) in terms of retrieval efficacy.<br />

Concerning security, work was essentially in<br />

medical imaging data, although it is now moving<br />

towards more global studies, into image<br />

databases. The aim is to supply innovative<br />

solutions to ensure feasibility and traceability of<br />

data at the interface of data and information<br />

systems [12]. Initial work focuses on the integrity

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!