Development of a metric access method with perceptual distance tuning to support similarity queries of medical images

Abstract

Efficient and accurate indexing and retrieval of information organized in databases to answer queries requested by users are pillars of the data science and engineering processes. Modern computer systems need to deal with complex data (images, long texts, videos, time series, among other types of data), which require differentiated and customized treatment for each specific data type. This type of treatment requires extracting the essence (features) of these data, making queries for similarity over such features, instead of using the complex data themselves. Similarity query processing uses distance (or dissimilarity) functions to indicate the proximity between the characteristics extracted from the data. Metric Access Methods (MAMs) were developed as fundamental tools to efficiently process queries by similarity. However, in many cases the use of traditional distance functions fails to provide answers compatible with users' perception of distance. This research project at the Master's level aims to develop mechanisms to address this problem, developing a new metric access method that allows including the user's perception of distance for processing queries. For that, it is necessary to use Relevance Feedback resources, aiming at capturing the user's perception regarding the similarity between images. Using this knowledge offered by experienced users, specialists on the subject, mechanisms can be used to train new professional users in the area, bringing benefits to the area of medical image analysis. (AU)