Non-Invasive Meningitis Diagnosis Using Decision Trees

Meningitis is one of the pandemic diseases that many less developed countries suffer, primarily due to the lack of economic resources to face it. The more severe types of meningitis, Meningococcal Disease, MD, demand immediate medical attention since delays increase the risk of mortality. This paper presents an open and integrated Clinical Decision Support System to assist physicians in the different stages of meningitis diagnostics through observable symptoms. Our system integrates three intelligent components which try to give support to physicians in early diagnostics of meningitis. These components are based on interpretable tree-based machine learning models and knowledge-engineering techniques. A dataset of 26,228 records of patients with a meningitis diagnosis in Brazil was used to construct and evaluate the system. The performance indicators of the decision models exhibit an outstanding classification performance for MD meningitis with a classification accuracy of 94.3%. In order to test the correct diagnosis of the system, an evaluation study with real patients' data was performed. The experimental results concluded that excluding meningitis cases based only on observable symptoms is much more complicated than diagnosing it. However, the system properly diagnosed 88% of meningitis cases from the real database.