Research On Application Of Data Fusion Methods In Electronic Medical Record Retrieval

In recent years, with the prevalence of paperless office medical, electronic medical records(EMR) is becoming more and more popular. EMR not only records the patients’ diagnosis, but also records the medical treatment and the therapeutic effect of patients. Such information can be used to assist clinical decision making when doctors are in clinical diagnosis. However, in the face of the growing EMR data,how to quickly find the information needed in the massive data of EMR is a challenge.The existing EMR retrieval system has some defects. Especially many EMR systems have some semi-structured parts, which cannot be effectively recognized and retrieved.Considering that medical is a field which demands accurate information, therefore, it is very important to improve the retrieval performance of the EMR system.On the other hand, in information retrieval,the data fusion technology tries to combine the results of several retrieval systems into a new one so as to get more effective results. Previous studies showed that using suitable data fusion method can improve the performance of the final retrieval results. But in the field of EMR retrieval, the application of data fusion methods has not been deeply explored.Therefore, we address EMR retrieval by using the data fusion technology. The main works of this paper are listed as follows:(1) The existing data fusion algorithms are analyzed, and the applicability of the data fusion algorithms in EMR retrieval is explored. Firstly, we analyze two classical data fusion algorithms: CombSUM and CombMNZ. Both of them equally treat all component systems. But in data fusion, the performance of the component systems may vary. The same treatment to all the systems may not be a good policy for achieving better results. In this paper, we focus on a specific data fusion algorithm-linear combination method. This kind of methods can assign different weights to different systems, according to the specific situation. And it is more flexible than CombSUM and CombMNZ. The key to the success of the linear combination method is how to determine the proper weights for all systems. In particular, we analyzemultiple linear regression based fusion strategy and Genetic Algorithm based fusion strategy. For the former, the least square method is used to estimate the relevance score of documents more accurately. For the latter, we use Genetic Algorithm’s global exploration ability to obtain more profitable weights. In addition, Particle Swarm Optimization Algorithm is applied to data fusion for the first time in this paper. In the Particle Swarm Optimization Algorithm, the particles can communicate with each other so that they can tends to the optimal solution, which can optimize the weight distribution strategy and improve performance of the fusion results. Finally, we apply the five data fusion algorithms to the electronic medical record retrieval.(2) In our experiments, two related data sets from TREC Medical Record Task are used. We test abovementioned five fusion methods with different numbers of systems being fused. All fusion methods are evaluated by using different measures both in effectiveness and efficiency. Experimental results show that Particle Swarm Optimization based fusion can improve the EMR retrieval results effectively. And CombSUM consumes the least amount of time. If we consider both effectiveness and efficiency, multiple linear regression based fusion is a good choice. Finally our experiments show that it is feasible to use data fusion to improve the performance of electronic medical record retrieval.