- 1997-2002. Neurosurgery department, Clermont Hospital
- 2001-2003. EA 1880, Functional Neurology and Epilepsy Unit, Neuro-cardiological Hospital, Lyon, France
We proposed to fuse anatomical (A) and functional (F) information coming from medical imaging. We focused on brain imaging, although the principle may be extended to other organs provided image are spatially registered. The anatomical images mainly consisted of MR images that give information on structural composition of the organ (distribution of tissues, fine spatial resolution). They can be obtain using a wide range of parameters, so that the resulting images provide sufficient contrasts between the different structures to be located (tissue interfaces, tumors,…). The functional images were obtained by injecting a radioactive tracer that preferentially characterized a physiologic (e.g., glucose metabolism) or a biologic process (e.g., blood flow). Two major functional modalities are used in brain imaging : SPECT and PET. These images had a bad spatial resolution and are not informative on what they were not supposed to represent.
In the following, anatomical and functional images were supposed to be registered. The same voxel (x,y,z) in the images thus corresponds to the same anatomical location. The registration procedure must be accurate enough to limit the amount of imprecision introduced. We then proposed a three-step fusion process
- Image modeling : information was extracted from the images and was modeled in a common theoretical frame to manage vague and ambiguous knowledge
- Fusion of image information : the information was then aggregated with a fusion operator. This operator must affirm redundancy and manage the complementarities and conflicts that often underline the presence of a pathology ;
- Decision : a fusion image was proposed that synthesized the available information.