Research On Intelligent Clinical Decision-Making Based On Machine Learning Method

Recent years,with the pervasive application of machine learning methods in various fields,many productive applications were occurred.There are also some promising applications in the traditional field-healthcare.In this paper,we try to develop an intelligent clinical decisionmaking strategy based on machine learning method to provide support for doctors when they are making the decision.Specifically,this paper focuses on the following three issues: feature extraction of medical signals,prediction of drug interactions,and recommendation of clinical treatment.For medical signals,this paper presents a data-driven feature extraction method for multivariable signals.The model based on Echo state network(ESN)is used to encode the signal into features.Further theoretical analysis shows that the well-known autoregression-based EEG feature extraction can be seen as a simplified variation of our FE-ESN method.The experimental results show that this feature extraction method based on ESN has advantages and opens up a new way towards unsupervised EEG feature design.For the prediction of drug interactions,this prediction problem is formulated as a multi-task dyadic regression problem,and the prediction of each specific interaction type is treated as a task.Compared with conventional matrix completion approaches which can only impute the missing entries in the DDI matrix,our approach can directly regress those dyadic relationships and thus can be extended to new drugs more easily.To solve this model,an effective approximate gradient method is developed.The validity of the method is verified by the evaluation of the real world data set.For the recommendation of the clinical treatment plan,this paper aims to develop an algorithm that can learn from the historical data in patient's electronic medical record(EMR),and provide the next-period treatment plan for patients according to their disease status,laboratory results and treatment records.Importantly,this algorithm takes consideration of both treatment records and physical examination sequences which are not only heterogeneous and temporal in nature but also often with different record frequencies and lengths.In addition,the algorithm also needs to combine static information(such as patient's demographics)with the temporal sequences to provide personalized treatment prescriptions to patients.In this regard,a novel Long Short-Term Memory(LSTM)learning framework is proposed,which models inter-correlations of different types of medical sequences by connections between hidden neurons.