One of the actual services allows to download images through internet access to CILEA ftp site but images can not be viewed before downloading and no image description is given.

     This paper deals with an analysis of the architectures for image retrieval. In this sense we have conceived of improving the VHD-MMS service through an on-line visual browsing and a content-based image retrieval system ([8], [11], [6], ). The main issue is the navigation of the images through concepts with semantic value as organs, tissues, anatomical district and apparatus. Content-based retrieval is characterized by the ability of the system to retrieve important images based on their visual and semantic contents rather than only by using attributes and keywords assigned to them. To reach this goal, a database is to be carried out which is composed by four main components: a graphic user interface which allows to visual navigating the image repository, an image analysis-model framework which provides the attributes definition and the image processing, a data retrieval model which specify the methods to access to the database content  and an user-dependent access to data (User View: a view consistes on a query which locates a timely and limited set of data). This last point is related to the ability of the system to customize the navigation path and the data access based on the user skills.

     A Database is a software framework which facilitates information retrieval provided that a suitable data model is given. It needs to specify:
    1. Database Management System (DBMS)
    2. Typology of the user who require services to the system (skilled, not skilled, local, on the WWW)
    3. Image retrieval model (IRM) which based on a specific language operates strategies of data retrieval
    4. Image data modeling (IDM) which defines the organization of the information related to the image and how it is stored.

     The critical step is the definition of the data modeling because all thre remaining features depends strongly on it. Images are bit-mapped fields but how can we represent them in a suitable way? To represent its content relevant features have to identified and extracted from it and the image model is the logical framework to organize those features. High level symbolic model needs to represent the knowledge about the image objects contained into the image and the relationships in between them. Features refer to image characteristic being used to define the visual contents of an image whereas semantics denote domain concepts manifested in the image. Whereas some automatic procedures are able to compute the most simple features, image processing is no even able to neglet the human involvement when extracting complex features. Also deriving semantics requires considerable human effort and this step is carried out by adopting a specific knowledge domain.

     In general we can define three types of data for images: structured data which are strings of characters,  unstructured data are unformatted byte string (they are in suitable format for visual display on output device), semi-structured data which are the joining of the two above data types. Structured data express image attributes through words. Obtaining these data requires manual indexing of image content and can be easily managed by conventional DBMS. Unstructured data representing image objects and their visual characteristics are obtained by image processing with semi-automated techniques. Both of the first two types give a description of image content but individually taken does not fully define the characteristics of the image. Formatted data can be easily managed into a relational DBMS. In this framework 2-D tables are used as logical data structures for storing image attributes (structured data) and relationships among the data are expressed by comparing the values stored in these tables. SQL (Structured Query Language) allows tables to be combined on the fly to express relationships among the data. SQL manages data manipulation and information retrieval. In figure 1  the scheme for a typical conventional database is shown.

     In our framework the user can query by textual attributes adopting a controlled vocabulary such as UMLS Metathesaurus ([9]). Nevertheless some shortcomings of the standard DBMS limit their applicability to image retrieval. The set of operations to manipulate data are limited, unformatted data can not be direcly handled, information about complex objects is often scattered over many relations or records leading to loss of direct correspondence between a real-world object and its database representation, no specific methods are given to deal with such complex objects and data are separated from operations. Then one can see that this framework is not convenient for high structured objects such as images or multimedia. To overcome this drawback Object Oriented DBMS will be adopted in our work.

     The problem of allowing different view on data is facing through the paradigm of Software Agents technology.  ([3]). This approach is based on a software entity, the agent, which can build as its proper knowledge a user model. An object can be defined as a combination of state and a set of methods that explicitly embodies an abstraction characterized by the behavior of relevant requests. An object is an instance of an implementation and an interface. An object models a real-world entity, and it is implemented as a computational entity that encapsulates state and operations internally implemented as data and methods and  responds to request or services. Given the above, a reasonable definition for agent follows as: An "agent" is an object that embodies a set of behaviours typically encompassing autonomy, intelligence, and potentially mobility. Therefore in our architecture each user of a medical specialty is helped in query by a dedicated agents which is able to generate a view of the data according to the user needs. Let suppose a radiologist get interested in images of the spine. The dedicated software agent builds a navigation path inside the images in which the bones stand out with respect other tissues or organs. As in textual query as in visual query the user is allowed to browsing the information he is related to without being constrained to manage the whole image content of the images. Moreover the agent can help the user to refine its query when this is not so specific to retrieve a small number of sample image.