Content-based image retrieval (CBIR) is a technique that promises a solution for indexing and retrieval of image content without resorting to labor intensive, text-based manual indexing, building an accessing structure that allows semantic-based browsing, and facilitating interactive query processes between the users and the system. With the emerging of the next generation high bandwidth Internet, transmission of high-resolution digital images in real time is becoming more and more a manageable task. Such a CBIR system can be deployed as a client-server infrastructure so that common users can have easy access to the pathology expertise and computation power at major medical centers.

Following the previous success of our cooperation with the Pittsburgh Supercomputing Center on our CBIR research, we are experimenting opening our CBIR image search engine for remote access in a cooperative setting. A Java client has been developed to allow users to submit a microscopic query image from their local hard disk and send to the CBIR search through the Internet. The CBIR server analyzes the query image, compares the image feature with those in the image database, and in real time, returns the search result to the user, including images of similar image content and their diagnostics made by pathologists at UPMC. The system is currently under testing with joint efforts from other medical institutes. Further research will focus on extracting distinguishing image features from compressed image formats, extending our current system to handle a broader range of tissue samples, and improving the efficiency of the system to handle the dataflow from the high-speed robotic image capturing station.