DiCloud: um modelo para compartilhamento transparente de imagens médicas com compressão e elasticidade proativa na nuvem

Abstract

Since 1970, with the advent of computed tomography, medical images have become a fundamental part of the diagnostic flow. The standardization of images through the DICOM format (Digital Imaging and Communications in Medicine) represented a significant advance for the interoperability of the medical image capture machines, as well as for the software that performs the visualization. However, in the present, some challenges need to be overcome to improve the processes in order to speed up the clinical diagnosis. The image storage server (PACS), in general, is located internally in hospitals, and it is not possible to share these images with other doctors outside the hospital’s domain, making it difficult, for example, to collaborate with a specialist who is in a location geographically distant. However, we can distribute this architecture using cloud computing, which also collaborates for sharing the medical images and helps the management supporting the quality of service even on the increase in workload. When using elasticity, it is possible to allocate and deallocate resources according to some metrics, such as CPU, to improve the performance of the application. Several works use elasticity to improve the application’s performance, but most of the works apply reactive elasticity, where one performs actions when a metric reaches a specific threshold. Also, works that address image compression to reduce the disk space demand do so only on pixel-data and do not focus on issues related to the image transmission time and the speed of the internet connection. The researchers-only access proposed in related works dismisses health professionals who can benefit from remote and faster access to medical images. In this work, a differentiated approach is employed, using a proactive elasticity and an algorithm based on time series to predict future values of a metric and make decisions beforehand. Four workloads were modeled and used to evaluate the model: constant, wave, ascending, and descending. We evaluate different ARIMA parameters for each workload, to choose which parameter set is the best ARIMA to forecasting in our model. Besides using the most efficient ARIMA algorithm, the tests also include measuring the compression level for each of the compression methods available on 7-Zip. We validated this model with real PACS server software, in terms of performance, compression rate, and the effectiveness in sharing images with another hospital. This work contributes to making the sharing of medical images possible even in environments with low internet bandwidth (commonly found in developing countries), through images compression for better use of the internet connection. The best compression algorithm was Zip + PPMd, which obtained a reduction of 72.32% in the dataset size. It also contributes by providing proactive elasticity of applications, where the results demonstrate that it is possible to maintain the quality of service (0% errors) and reduce the costs of the cloud (fewer connected VMs), through the correct allocation and release of resources.