Similarity searches of medical image data in peer-to-peer systems

The objective of this study is to effectively perform content-based medical image retrieval in distributed systems. We present a method that constructs a distributed index over a peer-to-peer network. Considering the network bandwidth limitations and other restrictions that are associated with the handling of medical data, we do not further distribute images between the participant peers in the network. We distribute only feature vectors, extracted from each image from which only a low resolution image can be obtained. The images are processed locally at each site. For the index distribution, we develop our own hash function that is based on multi-resolution analysis of the images using the wavelet transform and on a set of reference images that is known to each node in the network. To evaluate our method and demonstrate its applicability, we performed similarity searches on a brain image dataset. We also compared the performance of the distributed system to that of the centralized one.