Erwin Bellon, Michel
Feron, Frederik Maes, Dirk Vandermeulen, Werner Aerts, Joost Wauters, Guy Marchal,
Paul Suetens
Medical Image Computing (ESAT &
Radiology)
Katholieke Universiteit Leuven
Leuven, Belgium
Purpose of this work is to evaluate technological possibilities to offer advanced image processing and analysis services over the World Wide Web. Examples of such services are automated image registration, manual or semi-automated delineation of tumor volumes, or specific measurements required to extract quantitative information with novel image acquisition techniques.
Such services operate on a set of images to generate new images, curves or numbers. We anticipate that most methods will require user interaction, be it to initiate computer processing.
Instead of having the images transmitted to a central server computer that executes the appropriate software program, the processing application is sent to the location where the images are. To this aim, the image processing service is incorporated as a JAVA applet (or a set of cooperating JAVA objects) that can be activated over the Web from virtually any client computer equipped with a JAVA enabled browser.
Thus, it is not required to have specific image processing software installed locally, the processing service is in principle accessible from any platform, and processing may include extensive user interaction.
The image processing service we integrated to test this principle is that of image registration: aligning one 3D data set obtained, for example, with MR with a second data set of the same patient obtained with, for example, CT. The Web service incorporates a registration method developed at our laboratory that is totally automated (within a realistic range of starting conditions), has proven to be applicable in a large number of clinical situations, and is rather robust.
Figure 1 illustrates the image registration service as it is available within a JAVA enabled Web browser. First, the interactive user can specify the image slices to be registered, e.g. by reading DICOM files on the local PC or UNIX network. The individual image slices can directly be inspected visually (Figure 1 left), and options are available to interactively exclude parts of the data set. Then, when DICOM images are supplied, the application extracts relevant acquisition parameters out of the data sets while the user still has the possibility to enter missing values manually (Figure 1 right). The actual matching algorithm is currently executed by invoking native C routines from the applet. The resulting libraries, though not platform independent, are available for different platforms and the appropriate library is loaded into the browser automatically. After the registration algorithm has finished, the results can be inspected using the specialized image viewer included in the applet.
PICTURE
Figure 1 The registration service within a JAVA enabled Web browser with visual inspection of the images(left), and presentation of relevant acquisition parameters (right).
This Web based user interface and method invocation is used within various projects of biomedical research within our own institution. We are preparing technical evaluation of this Web distribution principle over larger distances, for supporting evaluation of image registration at cooperating hospitals located in other parts of the country.
We are convinced that image processing and analysis can have tremendous benefits in the process of diagnosis and treatment. In reality, however, there is a large difference between what seems possible in the laboratory and what is available in clinical practice. Networked services can affect different stages in the transition from research to clinical routine.
A first stage is validation of a novel method in the real clinical situation and for a large number of situations. This stage also serves to raise new ideas about possible applications, often by occasional tests by different users. A bottleneck is the effort it takes to (1) install the software in the different clinical environments and (2) sufficiently rapidly incorporate initial feedback and redistribute the improved software. The principle of Web distribution in a platform-independent way can help to get the spiral from laboratory to clinical application started.
A second stage is routine clinical operation but in a limited number of special cases. Although it is less important at this stage to reach occasional users, incorporating ongoing improvements in the method (and, at a certain moment even more importantly, in the user interface) remains a bottleneck that can be alleviated using Web based services.
A third stage is wide deployment as part of the routine process. Here, an essential element is tight integration in the overall information environment. This aspect is outside the scope of the present paper. It is possible though, that with future information systems, more of the intelligence will be in the network instead of in the individual workstations. In the mean time, the sheer proliferation of ‘standard’ Web technology opens perspectives for integration. In any case, if a processing service is developed to be distributed over the Web, it is probably also developed using principles of component technology, which by nature fosters integration.
Corresponding author:
Erwin Bellon
Medical Image Computing (ESAT & Radiology)
Katholieke Universiteit Leuven
Herestraat 49
B-3000 Leuven, Belgium
Email: erwin.bellon(at)uz.kuleuven.ac.be
Oral presentation at EuroPACS'98, Barcelona, Spain