Online Learning-Prediction Based Diagnosis Decision Support System Towards Swallowing Dysfunction in Rehabilitation Medicine

Medical diagnosis is a complex and fuzzy cognitive process of learning, such as neural networks of artificial intelligence methodologies, showing great potential can be applied to the development of medical decision support systems (MDSS). In this paper, online learning- prediction based neural networks are developed to support the diagnosis of swallowing dysfunction in Rehabilitation Medicine, along with the increasing accuracy of systematic study when the cases are input. The input layer of the system includes 28 input variables, categorized into five groups and then encoded using the proposed coding schemes. The RBF (Radical Basis Function) algorithms are employed to train the online learning- prediction system, the number of nodes in the hidden layer is determined by the online nodes updating process. Each of the 15 nodes in the output layer corresponds to one swallowing dysfunction disease of interest. A total of 120 medical records collected from the patients suffering from fifteen swallowing dysfunction have been used to train the system, where, 20 cases are used to test the system. Particularly, ‘5-fold’ cross validation is applied to assess the performance of the decision support system. The results show that the proposed online learning- prediction based decision support system can achieve very high diagnosis accuracy (>90%), giving rise to satisfied results and showing validity of the contributions.