Um estudo aplicado de armazenamento de imagens médicas em ambientes distribuídos

Abstract

DICOM is medical digital imaging standard adopted by the vast majority of companies that manufacture radiological equipment such as CT (Computed Tomography), US (Ultrasound) and MR (Magnetic Resonance). These images are stored in PACS (Picture Archiving and Communication Systems), a system for storage, retrieval and communication of these digital radiological images. These exams can generate multiple images/files which are often very large, about 500KB each, making them difficult to manage, distribute and store. In order to lessen these problems, an Applied Study was conducted to compare the different technologies used to replicate and return this information, specifically DBMS Postgresql and the HDF5 hierarchical data format. This study seeks to present means/technologies that enable large volumes of data relating to medical images to be organized, accessed and manipulated. It also seeks to comply with the DICOM standard, but with a performance differential in relation to storage and retrieval of images. The study was conducted on a computational cluster and consisted of replicating and recovering sets of images (1, 50, 250, 500, 1250, 2500 and 5000). For each set replicated and returned, metrics were collected relating to: maximum time, minimum time, total time, average and standard deviation. Finally, experiments were carried out to compare these metrics. In general, the DBMS obtained the best response times for replication, while the HDF5 obtained shorter times for return of information, as shown by the metrics obtained