Image and Information Processing for Health
The "Laboratory of Medical Information Processing" (LaTIM)
is carrying out a multidisciplinary research programme involving physicians and scientists within the framework of a close cooperation between the School of Medicine, the University Hospital of Brest, and the Graduate Engineering School of Telecommunications of Brittany. The research concerns two themes:
Both themes are integrated in a single methodological approach to information- and knowledge-based medical image processing and analysis for diagnostics and therapy.
In all these subjects, LaTIM has developed a cooperative network at regional, national, European and international levels, which facilitates collaborations with the best partners and leading experts in the world. The research of the laboratory is applied to more applied projects, with the purpose of proposing reliable technological solutions to actual medical problems. The results of the methodological research projects are integrated within technological and clinical research platforms for evaluation and assessment, in preparation to industrial technology transfer.
Dynamic forms in medical imagery
This methodological orientation concerns the extraction and analysis of biomedical shapes — that is, the passage from the voxel to the organ — and brings together certain longstanding work carried out by the members of the laboratory. It deals with the different steps of a processing chain: the reconstruction of the 3D+t signal, the extraction of shapes, their characterisation, their evolution, the relations between form and function. In practice, the constituents of this chain are not totally inseparable and their implementation is not necessarily causal nor linear; for example, improvements in the reconstruction of the 3D signal may require mastering the extraction and evolution of the forms concerned. Similarly, the exhaustive analysis of the form of an organ cannot ignore either its formation process or its function. This both justifies and imposes the multidisciplinarity of the approach.
We wish to model biomedical shapes geometrically — locally, regionally and globally. The final objective may be a quantification used for diagnostic or therapeutic ends, the comparison with a reference model (atlas), the study of the evolution during growth, temporal variations during movement for rigid anatomical structures as well as for elastic ones. This requires 3D models that are abstract, synthetic, concise representations of the structures, and that are easy to manipulate. Like in the domain of physics, the general problem associated with the analysis of biomedical forms is organised around these three fundamental entities: the support, its valuation and the model. The choice of this terminology deliberately reflects the main basic orientations in which the recent work by the team has been engaged.
The complexity of this issue precludes taking one single approach. For this reason, we are studying several families of models that we have classified below according to the three main contexts of study (dynamic interfaces, forms and movement functionality, forms and deformations) that we are developing, and according to the nature (local, regional or global) of the deformations that they can undergo.
Osteology and arthrology represent the primary medical arena which ties together some of this work. Quantitative functional imaging by PET constitutes another domain of methodological integration, as well as dynamic analysis of the venous thrombus by echography.
Collaboration: Service de Chirurgie Orthopédique, Traumatologique et de Reconstruction, CHU Brest; Faculté d’Odontologie, Brest et Service d’Odontologie, CHU Brest; Service d’ORL et de Chirurgie Face et Cou, CHU Brest; Département de Médecine Interne et Pneumologie, CHU Brest; Laboratoire d’anthropologie des Populations du passé, UMR CNRS 5809, Bordeaux. University of North Carolina, Chapel Hill, USA, National Library of Medicine, USA. TIMC-IMAG Grenoble, CHU Brest, CHU Nîmes, CHU Lyon. Université Technologique de Compiègne; Laboratoire d'anthropologie des Populations du passé, UMR CNRS 5809, Bordeaux, TIMC-IMAG, CNRS UMR 5525 (J. Demongeot, J. Troccaz)., CHU de Grenoble, Clinique du Cèdre ROUEN; University of Pennsylvania, Philadelphia, USA (Jay Udupa). UCL London, UK (Peter Hell), University of North Carolina, Chapel Hill, USA, National Library of Medicine, USA,
Technology transfer: CERTIS Landerneau (ACI MENRT). Kitware, USA. THALES Microsonics, Sophia Antipolis; MORS Aix-Marseille; Praxim Grenoble; France Télécom R&D. Praxim, DePuy France. Philips Research Labs, Aachen, Germany (Matthias Egger, Falko Busse, York Hamish), Siemens Medical Solutions, Chicago (Xavier Battle).
Indexing and similarities in medical imaging for shared know-edge bases
Medical images constitute the privileged way to access and search information and knowledge in medical records and reference basis, diagnosis aid, for aided diagnostic, therapeutic follow up and training. Digitizing all this space of information and knowledge will definitely be achieved in the very near future. Nevertheless, despite the abundant related literature, many open problems remain concerning the structuring of multimedia medical knowledge basis in order to be constituted and exploited.
Since the beginning of the 90’s, the LaTIM has been working on image compression and indexing using their numerical content as well as their semantic meaning, and on data and knowledge fusion, with the aim of building intelligent and communicating image databases that can be used in computer assisted image interpretation and training. Because of the increasing need in data basis sharing and diffusion, new issues are raised related to security, especially to image integrity control, and to collaborative work when sharing these resources. Last, it is necessary to study the interoperability of the systems under development by taking into account the international standards in the multimedia domain which undergoes a very fast evolution.
In this context, the LaTIM research keeps on progressing on the central question of information and knowledge structuration in medical imaging and in multimedia medical databases. This research topic is based on the extraction of information and knowledge structuring elements (indexing), and on the study of similarity measurements between these features. The methodology followed will be inspired by image engineering and knowledge engineering, that we will try to cross-fertilise with the objective of focused and specific developments and applications.
The indexing problem is studied through its two dimensions, numerical — extraction of digital signatures from the numerical content of the image — and symbolic — representation of image symbolic content based on medical knowledge. The similarity measurements study follows the same approaches: numerical for numerical signatures, based on medical reasoning for symbolic indexing. These two approaches will be integrated in a technological and clinical research platform dedicated to "Multimedia information and medical knowledge sharing". This platform will also take into account image security and multimedia standards.
Collaborations : LaLIC (Langage, Logique, Information, Cognition) CNRS FRE 2520 (M. Abraham, JP Desclés), LACO (Labo de Sciences Cognitive de Poitiers) UMR 5096 CNRS, LORIA INRIA et CNRS UMR (J.P. Haton), Télécom ParisTech, ERA CNRS (H. Maître), IRCCyN-IVC, UMR CNRS 6597, Ecole Polytechnique de l’Université de Nantes (J.P. Guédon), LTSI INSERM EMI 9934 (J.L. Coatrieux), Université McMaster, Canada, Université de Calgary, Canada (Raj Rangayyan), Instituto Tecnologico de Monterrey, Campus Ciudad de México (Alfonso Parra)