Reference system architecture for biomedical data analysis

Abstract

The development of adaptive system has been researched for a decade, but only recently machine learning problems became solvable. In this paper is presented an overview of current smart systems and the problem of development real adaptive and versatile data analysis system. Such system has to analyze multivariate signals of different nature. That is why the overview part is based on examples of biomedical human health monitoring systems (BHMS). BHMS have complex internal structure, to simplify it can be divided into three big tiers: tier of data acquisition, communication and analytics tiers. In the first tier, all data acquisition steps are made, including placement sensors to the specific location. The communication tier is responsible for gathering data from all available hardware and send it via wireless channel to the third layer. The analytical tier is the final destination of acquired data. This layer includes all algorithms which will be used for processing data and interpretation of the results. Such simple division allow us to specify and select our potential field of study and improvements. Specifically, in this paper the analytics tier will be explored and new system architecture will be proposed.The crucial step in the adaptive systems development is solving reconfiguration problem. Nowadays, almost all data analysis systems have specific applicable fields. The main advantages of the latter are that they are robust, fast, but the development of such system can take a lot of time and reconfiguration of currently used systems will allow speed up that process. The reconfiguration is process of creation a specific targeting system from general system setup or currently known system. In this paper we have investigated problems, related with reconfiguration and proposing the theoretical basis for creation the reconfigurable adaptive system. The output structure is two-tier system, which allow to analyze any task from different field of study and then process it in a single setup. Each of the tier is considered as independent components, however the combination of the general reference system (tier one) and training agent subsystem (tier two) creates a highly adaptive system with reconfiguration properties. The reference system itself is a big storage of available algorithms which can be applied for certain problem and the algorithms workflow can be tuned for specific problem. Such adjustments can be made because of system training agent, which main purpose is to process input task and defined which algorithms can be applied. Such task classification problem can be solved by using machine learning algorithms.Since, this paper presents only theoretical basis there is no precise implementation of proposed system. Currently we just gave a small overview of the future development of adaptive and reconfigurable systems.Ref. 12, fig. 5.