FL-LIMS: Laboratory Information Management System as an intelligent system

Transcription

1 FL-LIMS: Laboratory Information Management System as an intelligent system DIANA CALVA MENDEZ 1,2, MARIO LEHMAN 2 1 CADIT Universidad Anahuac, Av. Lomas Anáhuac s/n, Col. Lomas Anáhuac, Huixquilucan, Edo. de México, C.P , Mexico DF, MEXICO 2 CEMINT Sofilab, Lisboa 14-A, Colonia Juárez, Delegación Cuauhtémoc, México DF, MEXICO Abstract: - Laboratory Information Management Systems (LIMS) have been in use since the mid seventies. Today the laboratory needs, have more to do with decision support and decision making. Depending on the problem needed to be solved, a different and unique method of artificial intelligence would have to be implemented that could easily interact with the database search engines each LIMS uses. Here we identify some of these new problems to be solved by a LIMS, the possible alternatives that could be implemented in order to solve them and include an example based on fuzzy logic. Key-Words: - LIMS, Artificial Intelligence, Data Mining, Clinical Laboratories, Relational Databases, FSQL 1 Introduction Laboratory Information Management Systems (LIMS), support the clinical laboratory daily tasks and procedures. The "product" of a clinical laboratory is data. When data are organized in such a way as to convey meaning, they become "information". Raw laboratory data is of limited usefulness until it is transformed into information, for example, by placing it in the context of reference intervals, or through validation, and quality control measures that verify its accuracy. The clinical laboratory is well-suited to integration with medical informatics because of the large amounts of quantitative, qualitative and morphologic data generated. A typical LIMS includes the models and technological tools for managing, storing, processing, and communicating laboratory data, and for transforming it into useful information [1]. The amount of information obtained from the daily tasks performed in the clinical laboratory need to be supported with information systems. Also, most of the data collected maintain relations which need to be recorded and tracked in order to provide information useful for the laboratory operation. During the seventies and eighties, what was expected from a clinical laboratory information system, had to do only with storing and retrieving test results, booking patients, knowing the amount of tests performed over a period of time, billing patients, generating and printing test results reports, and performing simple statistics (internal quality control) among others [2]. These first systems were only known as Laboratory Information Systems (LIS), all of them were developed in-house by each clinical laboratory and depended totally on the technology available in site. It was until the beginning of the eighties when the first commercially available LIMS appeared, but were not still available to every laboratory and great amounts of money had to be invested in hardware and were based on the use of dumb terminals and proprietary databases. It was not until the beginning of the nineties when LIMS became economically available because of the appearance of personal computers and relational databases [3]. Today most LIMS use relational databases and structure query language (SQL) for performing most of these tasks which are based on data searching, updating and retrieval. As the amount of data grows and the clinical laboratory finds their new needs could, in a way, be solved by adding new functions to the LIMS, queries become more complex and the structure query language becomes limited. Most of these needs have to do with planning, logistics, aspects related with total quality requirements [4, 5], obtaining information having to do with the behavior of certain diseases (cancer) [6], the prevalence of different clinical parameters (like high glucose or cholesterol) in a target population, management of epidemic data or data that could lead to the identification of nosocomial infection (grouped by age, sex, geographical location, etc.) [7, 8], among others. Other needs have to do with including new technologies such as voice controlled tasks; image processing for identification of sample tags or character identification in medical orders; ISSN: ISBN:

2 making suggestions based on the history of a particular patient, etc [9]. Also, most of the time, the laboratory stores precise data and needs to retrieve information based on imprecise criteria where the existence of a fuzzy query language or a way of translating the stored data into imprecise information in an effective way might be the best solution for achieving this possibility. All these new needs can be solved by using one or more artificial intelligence techniques. Today, no commercial available LIMSdatabase engine provides tools based on artificial intelligence in order to perform these complex queries. In this work we identify and briefly explain some of these needs and propose a solution based on one or more artificial intelligence technique. Also we will illustrate an example of a query based on imprecise criteria over precise data. 2 Data Retrieval on a Traditional LIMS based on a Relational Database In previous works we have presented issues having to do with designing an application flexible and functional enough in order to meet different clinical laboratories needs [10], the interaction of other applications dealing with telemedicine and LIMS [11], and the particularities of the Mexican health system and how it affects LIMS design and development [12]. In the Apendix, the user interface of the LIMS, we have developed, is shown. However all previous works used commercially available relational databases and used simple queries in order to solve different problems and situations demanded from the LIMS. In relational databases, data retrieval is achieved through the performance and execution of queries, stored procedures, views or functions. Most of the time, one of these methods are enough in order to obtain results and extract useful information from the database. However these is limited for retrieving information that already exists in the database when using as search criteria any information already existing in the database as can be seen in Fig. 1, where a store procedure for obtaining the set of patients booked between certain dates is shown. The procedure retrieves the list of the patients including their First Name, Last Name and Patient ID. These data are retrieved from two different tables. However this kind of query will retrieve patients having 12 years old or less and will leave out of the result those that might still be considered as pediatric patients but whose age is slightly greater than 12 years, since the criteria pediatric is imprecise. The same will happen when trying to obtain the set of adult patients, since a range containing the ages where a patient is considered adult would have to be established, leaving out of the resulting set of patients those that might be considered as adults but do not belong to the selected range. CREATE PROCEDURE char char char varchar (50) AS = 'AF' /* ADMISION - FOLIO*/ begin SELECT C.Buscador,P.A_Paterno, P.A_Materno, P.Nom_Pac, C.Fecha_Cit,C.Folio FROM Pacientes P, Cita_Admision C WHERE C.Buscador and P.IdPac = C.IdPAc and C.Tipo = 'A' end = 'AM' /* ADMISION - MATRICULA*/ begin SELECT C.Buscador, P.A_Paterno, P.A_Materno, P.Nom_Pac, C.Fecha_Cit,C.Folio FROM Pacientes P, Cita_Admision C WHERE P.Matri_Pac and P.IdPac = C.IdPAc and C.Tipo = 'A' End Figure 1 - SQL Stored Procedure for retrieving the full name of a set of patients, by using different search criteria (data contained in a specific table of the LIMS database). SELECT Name, Age FROM Patients WHERE (Age <= 12) ORDER BY Age Name Age Laura Esquivel García 1 Mónica Hernández Gámez 3 Javier Pérez Martínez 5 Ramón López Méndez 8 Angel López Gómez 11 Helena Pérez Martínez 12 Figure 2 - Query for obtaining pediatric patients based on the assumption that these patients are the ones having 12 years old or less. ISSN: ISBN:

3 A similar query could be performed for retrieving information on a set of patients classified between a range of dates of birth, and this way the set of pediatric patients could be obtained. An example of such query can be seen in Fig Fuzzy Logic for achieving imprecise queries The description above is just a simple example of the limitations of the current solutions a LIMS based on a relational database can present. Imprecise data can more easily be obtained by using fuzzy logic. The best way to achieve fuzzy queries is by implementing fuzzy classification, this way, there is no need of having a database able to support fuzzy logic operations in the structured query language [13].This process might be achieved by converting all the crisp data into fuzzy data by using additional tables and catalogs and then applying a trapezoidal function that will allow associating a membership degree. In these case we used the following membership function: in Fig. 4. Having a fuzzy logic as a tool for retrieving imprecise data from a LIMS adds functionality and facilitates the implementation of epidemiologic applications, follow up and evaluation of tests results for a particular treatment being applied to an individual or group of individuals. NAME AGE Ulises 24 Ana 10 Carlos 55 (a) TAG a b c d Adult Young Child Old (b) Once the membership negrees are obtained for each set, we applied T-Norm, for obtaining the intersection of the sets and therefore the imprecise data needed for the classification according to age. An example of this can be seen in Fig. 3, where a table containing the membership degrees for age, are contained in table Mem_Deg_Age, this is obtained after applying f(x,a,b,c,d) to particular set of crisp data [14]. Also, the window showing the user interface of this particular application can be seen in the appendix in Fig. 5. and Fig. 6. This way, several tables containing Membership Degrees for other important classification criteria may be added as tables in a conventional database entity relation model, and fuzzy searches could then be performed even by combining the results for other different fuzzy classifications resulting from applying the corresponding function and fuzzy set operations to these new tables as can be seen (c) NAME AGE MEMBER DEGREE Ulises (young) Ana (child) Carlos (old) (d) Figure 3 - Fuzzy Classification over a conventional database. a) Table containing crisp data, b)values used for evaluating the Membership function, c) Graphical representation for obtaining the membership degree for each data d) Table containing Membership Degrees after applying f ( x, a, b, c, d) and the resulting classification after applying T-Norm ISSN: ISBN:

4 the Renyi entropy in the sense of coordinates characterizing the data [19-21]. Then, the trapezoidal membership function of Eq. (1) can also used for the case of clustering with Renyi entropy. The Renyi entropy is defined as: S R q N 1 log q = Pi 1 q (2) i= 1 where P i is a probability, which can be for the set of data, q is the free parameter which identifies the order of the entropy. Figure 4 - Scheme for data retrieval. In Fig. 4 we can see the scheme of the procedure for data retrieval, (A) SQL Query analyzer performs a query over crisp data, (B) Results from this query contain the needed information for fuzzyfication, (C, D) A particular function is applied for obtaining membership degrees, (E) Membership degrees are stored in a table, (F) After performing store procedures, triggers or views on the query analyzer fuzzy information is obtained, (G, H)Fuzzy set operations can be performed with the results obtained from the second query and an imprecise result can be obtained. Figure 5 - Window showing the result of implementing fuzzy classification, over a conventional relational database model. 4 Clustering data and fuzzification with Renyi entropies Such as was developed by other authors, using the Renyi entropy and maximizing principle, we can use Figure 6 Clusters with Renyi entropy, for two different values of q. 5 Conclusion The application of artificial intelligence techniques is proposed for performing several functions that, today result difficult to develop with simple SQL queries. We show an example of how imprecise queries that are useful for the ISSN: ISBN:

5 clinical laboratory can be achieved. Future work involves evaluating the different artificial intelligence techniques, their implementation with simple SQL queries and which are best suited for adding functionality to current LIMS which do not count with these kind of applications and tools. As additional result we present a scheme based on the Renyi entropy for clusterizing the data distribution and build fuzzy clusters, which can be included in the LIMS for retrieval information and classification. Acknowledgments This work was supported by Sofilab S.A. de C.V. and Fondo Sectorial para el Desarrollo Económico, from Consejo Nacional de Ciencia y Tecnología (Conacyt) and SE (México). References: [1] Buffone, G. J., Dennis, R. M.,: Laboratory Computing-Process and Information Management Supporting High-Quality, Cost-Effective healthcare. Clinical Chemistry. 41(9), (1995). [2] Cristopher, M.M., Hotz, CH.S.: Medical Informatics in the Veterinary Clinical Laboratory: an Integrated Subspecialty in Diagnostics, Resident Training and Clinical Research. Revue De Medecine Veterinaire. 152(7), (2000). [3] Townsend, N., Waugh, M., Flatter, M.: LIMS: Meeting the challenge of modern business. American Laboratory. March, (2001). [4] Grauer, Z.: Laboratory Information Management Systems and Traceability of Quality Systems. American Laboratory. September, (2003). [5] Steele, T.W., Laugier, A., Falco, F.; The Impact of LIMS Design and Functionality on Laboratory Quality Achievements. Accreditation and Quality Assurance: Journal for Quality, Comparability and Reliability in Chemical Measurement. 4(3), [6] Quo, C.F., Wu, B., Wang, M.D.: Development of a Laboratory Information System for Cancer Collaboration Projects. Annual International Conference of the IEEE-EMBS th, (2005). [7] Wurtz, R., Cameron, B.J.: Electronic Laboratory Reporting for the Infectious Diseases Physician and Clinical Microbiologist. Clinical Infectious Diseases. 40, (2005). [8] Lamma, E., Mello, P, Nanetti. A, Riguzzi. F, Storari. S, Valastro. G: Artificial Intelligence Techniques for Monitoring Dangerous Infections. IEEE Transactions on Information Technology in Biomediciene. Jan:10(1), (2006). [9] Winkel, P.: The Application of Expert Systems in the Clinical Laboratory. Clinical Chemistry. 35, (1989). [10] Calva, D., Landa, A., Lehman, M.: Information Management in the Clinical Laboratory. WSEAS Transactions on Computers. 5(3), 1238 (2004). [11] Calva, D., Landa, A., Lehman, M.: A Low Cost System for Telepathology. WSEAS Transactions on Biology and Biomedicine 2(2) 253 (2005). [12] Landa, A., Calva, D., Lehman, M.: Sistemas de Información para la Administración del Laboratorio Clínico. Proceedings del congreso de SOMIB (2003). [13] Werro, N., Meier, C., Mezger, C., Schindler, G.: Concept and Implementation of a Fuzzy Classification Query Language. Proceedings of the International Conference on Data Mining, DMIN 05, World Congress in Applied Computing, Las Vegas (2005). [14] Medina, J.M., Vila, M.A., Cubero, J.C., Pons, O.: Towards the Implementation of a Generalized Fuzzy Relational Database Model. Fuzzy Sets and Systems. 75(10), (1995). [15] Kononenko, I.: Machine Learning for Medical Diagnosis: History, State of the Art and Perspective. Artificial Intelligence in Medicine. 23(1), (2001). [16] Westgard, J.O., Barry, P.L., Hunt, M.R., Groth, T.: Clinical Chemistry. 27(3), (1981). [17] Chang, J.Y., Tao, S.Q., Deng, CH.: Design and Implementation of a Deductive Database System, J. Beijing Polytech. Univ. 30(4), (2004). [18] Taskar, B., Segal, E. Killer, D.: Probabilistic Classification and Clustering in Relational Data. Proceeding of IJCAI-01, 17 th International Joint Conference on Artificial Intelligence. [19] E. Gokcay, J. C. Principe, A New Clustering Evaluation Function Using Renyi's Information Potential, International Conference on Acoustics, Speech, and Signal Processing (ICASSP), [20] R. Jenssen, K. E. Hild, D. Erdogmus, J. C. Principe, T. Eltoft, T., Clustering using Renyi s entropy, Neural Networks, Proceedings of the International Joint Conference, Volume 1, Issue, July 2003 Page(s): [21] M. M. Mayoral, Renyi's entropy as an index of diversity in simple-stage cluster sampling, Information Sciences, Volume 105, Number 1, March 1998, pp (14). ISSN: ISBN: