I think that high-end processing task as remote automated medical imaging systems for diagnostic purposes or MRI image classification such as described here in the following JAI success story (see link), can only be done via SaaS desktop, since the heavy imaging processing would be too slow for real-time image query & retrieval, because the remote client (a rural clinic , etc) have to send the huge image file(s) over the internet to the central server somewhere, to be interrogated to find (retrieve) similar images if any that might have already been stored.

Java Advanced Imaging In Action — Millenium Technology, Inc

A browser based image retrieval system would almost be too slow to have a useful effect in a real-time remote clinical decision support system environment. With a SaaS desktop medical imaging app, the query image (perhaps it’s an MRI scan of a newly admitted patient to a remote clinic) is processed locally (client’s site), where the query image is converted to image features, which is a bunch of numbers in a matrix form using a method (algorithm) called feature extraction. This sets of numbers (image feature matrix) is sent from the remote client to the central server somewhere with a repository of medical images in a database. This repository of medical images have also been converted to image features (each one), which are stored in an image feature database (which is different from the actual image database itself). Each image feature IDs in the feature database points to image IDs from the actual image database. When the query image feature arrives from a remote client, it is then computed against the image feature database to find those that are similar to the query. There is only similar images retrieved, since there is hardly any 2 images with the same value pixel-by-pixel. The closest match features (say, the most similar 5 or 10 images), are then retrieved (the matched features IDs are used for retrieval).

So, it is faster to process image locally to extract its features via the UI, then send the features (a matrix) which is a reduced dataset of the original image (this means ,less data) via the internet as a query to the central repository rather than sending the raw image itself. Image processing is quite a heavy task (memory-wise), so this the advantage of SaaS desktop compared to a browser-based system. The other advantage of dedicating some of the heavy task to the clients, is that the server won’t be overwhelmed by many requests from remote clients if the raw query images are sent instead of compressed image features , because the server might be receiving thousands of images that require feature extractions processing on the fly. The job becomes easier if feature extractions is done on the client.

PS: We will see more of these types of SaaS desktop medical imaging decision support system being deployed for commercial use over time.

Troy.

Ben

The web-based that I am currently writing is applet-based, but when graphics are going to be added in the future, then I will switch to JavaWeb Start, since applet codes (Java Swing) could be converted into a JavaWeb Start easily (ie, no major overhauling).

Overall, I agree with Phil Wainewright’s article.