Revista Colombiana de Reumatología
Print version ISSN 0121-8123
NEWBALL, McCarthy; QUINTANA L, Gerardo and NINO I., Luis Fernando. The immune system as a diagnosis tool of rheumatologic diseases. Rev.Colomb.Reumatol. [online]. 2007, vol.14, n.4, pp. 287-296. ISSN 0121-8123.
Introduction: the biological systems have been observed and analyzed carefully and they have transformed into models to be emulated in many types of scenery and these offer solutions to problems of the real life, more recently. The immune system is one of the most representatives and at the moment is used for implementation of computational systems to respond to many tasks, constituting the Artificial Immune Systems. Objective: in this work a computational method inspired by immunology for diagnosis of rheuma-tologic diseases is developed. The goal is to obtain a computational tool that, given a group of clinical histories as training data, performs rheumatologic diagnosis comparable to the current systems used in document classification. The proposed tool is expected to contribute in education and decision making in rheumatologic diagnosis. The proposed system is inspired by the interaction between tissues and B lymphocytes, and it relies on concepts of information theory to extract relationships among terms. The B lymphocytes will have the function of discriminating a patients rheumatic diseases based on its clinical history. Materials and methods: a dataset consisting of 54 medical records from 54 patients with rheumatologic diseases was used; 21 patients suffered rheumatoid arthritis, and the rest suffered other rheumatologic diseases. The dataset was divided into two groups: patients with and without rheumatoid arthritis. A manual process on the clinical histories was performed to eliminate the irrelevant information in the diagnosis task. The effectiveness of the system was compared to other three text classification algorithms widely used in document classification tasks, namely, ID3, BayesNet and PsoSVM. Results: the proposed system obtained promising results in comparison with other algorithms, with an average of 87,65% effectiveness in the diagnosis. However, due to the limitation of the data, there is a possibility that the results are biased. It was observed, as expected that the antibodies that represent the information in several cases are redundant. Additionally, the information that it represents not necessarily corresponds to medical knowledge, but to rules of text classification. Conclusions: information theory in conjunction with an adaptive immune system and a signaling mechanism showed great potential for the classi-fication of medical records. Due to the possibility of a bias in the results, it will be necessary to carry out additional experiments on a larger and more heterogeneous group of medical records. From the experiments, antibodies that clearly represented concepts explaining rheumatoid arthritis were not obtained, which could help medical trainees in the learning process and medical doctors in decision making. Therefore, in future work, the task to continue consists on adapting natural language processing methods (i.e., syntax and semantics) to obtain a knowledge extraction system instead of a set of rules for text classification.
Keywords : artificial immune systems; rheuma-tologic diseases; diagnosis.