2PROM: A two-phase image retrieval optimization on dataspace using predictive modeling

This paper introduces 2PROM, a new algorithm that can efficiently retrieve information from a set of multimedia and heterogeneous data sources. We published IHM, a model to predict whether a x-ray picture carries the trace of cancer (viruses). We further generalized our approach to The NoCancerSpace, a dataspace for medical diagnosis of lung cancer. This paper presents the optimization algorithm used by the NocancerSpace. 2PROM is designed to optimize the Dataspace retrieval process with two main phases. Its first phase consists of building a pipeline to find the best retrieval strategies. In fact, the pipeline explores the set of alternative execution strategies to determine the cheapest one. The retrieval strategies are initial nodes of the next phase. As for the second phase, retrieval strategies are combined with predictive model to determine the most efficient way to execute a query. In order words, the optimizer considers the possible retrieval strategies for a given input query, and attempts to determine which of those strategies will be the most efficient. The retrieval strategies are represented as XML tree of “strategy nodes”. The output of the second phase is the best results found. Experiments show that 2PROM retrieves more relevant results in less time than existing systems.