Analysis of inter-concept dependencies in disease diagnostic cognitive maps using recurrent neural network and genetic algorithms in time series clinical data for targeted treatment

The knowledge of dependency of a particular factor on another is a very important aspect in the healthcare field. If we have a rough idea about the effect a prescribed drug has on the cure of a disease or sufficient information about certain symptoms of a disease being linked to each other, we can make an informed decision about its treatment over a period of time. This paper proposes a method which makes use of a special kind of recurrent neural network (RNN) known as long short term memory network (LSTM) to make predictions for time series data. Genetic algorithms are also incorporated to identify the most important concepts affecting a patient over that period of time. The output of the LSTM network is in the form of binary strings and is utilized to generate a fuzzy cognitive map (FCM) for the same and a novel method is proposed to find the values of the interdependencies between various concepts, an approach which can be applied in clinical decision support systems. This method makes use of the weight matrices obtained after training the neural network. It is shown to be an improvement over the previous work done in this domain. The proposed method was tested with various clinical datasets and results were obtained for the same.