Motivations
Approach: Program Supervision
Framework: the LAMA platform for KBS development
Program supervision for medical image processing: MedIA
Experimental MIP applications
The LAd communication manager 

Within ORION research work on program supervision (PS), I focus on defining solutions that are appropriate to handle specific aspects of (functional) medical image processing (MIP).

To face the increasingly complex, rapidly evolving medical imaging techniques, many research laboratories produce performing, up-to-date medical image analysis methods.  Their use to extract information from images involves many methodological decisions, that fall outside both HCPs' scope and competence, although closely linked to medical knowledge. Besides, however powerful a MIP method may be, the pertinence of extracted information crucially depends on its reliable and robust use.  Consequently, the efficacy of the methodology used for processing (e.g.~appropriate, repeatable or operator-insensitive) is a prerequisite to good quality interpretation of MIP results, either performed by human or software. These difficulties crucially restrict MIP diffusion among HCPs, who  often only have access to simple, predefined MIP tools, insufficient to cover all their needs.

Within this concern, my work aims at providing HCPs with means to benefit from research-level medical image processing (MIP), by coupling MIP software with knowledge-based program supervision techniques. Although not fully automatic, a PS system discharges HCPs from the expertise about managing and using MIP, therefore enabling them to focus on their clinical tasks (e.g. diagnosis, planning and follow-up of therapies or surgical interventions, study of sane or pathologic body activities, cognitive improvement). Although sometimes complementary, my approach is different from some approaches that choose to automate the interpretation of MIP results.

The specifics of MIP domain provides an interesting and challenging experimental field for PS extensions. On the one hand, I work on knowledge representation (i.e. images, goals, available processing, user requirements), that must mirror the rich, non-exhautive and evolving character of MIP. On the other hand, my work aims at defining a supervision method that respects experts' reasoning process. Finally, as complete automation is hardly achievable and seldom suitable for such domain, I also consider the form of user interaction during the solving of a problem.  I propose a model of the necessary concepts for MIP PS: 1) a hybrid supervision method, mixing skeletal-plan refinement and reactive adaptation, and 2) a description of data and domain objects structured with different perspectives.