Research On Computer-aided Decision-making For Elderly Patients With Sarcopenia

Sarcopenia is a geriatric syndrome identified by a combination of muscle strength,muscle mass,and physical performance,which can erode elderly people’s independence and lead to increased adverse outcomes.In view of the current problems of unclear pathogenesis,inconsistent diagnostic criteria and no effective treatment drugs for sarcopenia,this thesis conducts computer-aided decision-making research on sarcopenia based on machine learning technology,and provides new insights into the risk factor analysis of sarcopenia.It also establishes a prognosis prediction and treatment recommendation model for elderly patients with sarcopenia,designs and implements a decision support system for sarcopenia,to improve the efficiency of doctors’ diagnosis and treatment and promote the precise prevention and treatment of sarcopenia.The main work of the thesis are as follows:1.Aiming at the problem that sarcopenia involves many factors and combination effects of risk factors are ignored,this thesis proposes a risk factor analysis method for sarcopenia,Feature-Ranking and Feature-Joint（FRFJ）.The risk factor ranking module is built based on feature selection algorithms such as Max-Relevance and Min-Redundancy（m RMR）,Relief F,to analyze individual importance of risk factors.And then,the Feature-Joint（FJ）algorithm is proposed to construct the key risk factor analysis module to analyze combination effects of risk factors and remove redundant risk factors.In the simulation experiments of the UCI dataset and the sarcopenia dataset,the results show that grip strength,gait speed,time for 5 sit-ups,heart failure,SARC-F score and stair climbing are the key risk factors for sarcopenia,of which SARC-F score and stair climbing have combination effects,and the classification accuracy is 92.15%,the F1 value is 94.01%,the specificity is 83.77%.2.Aiming at the problem that prognosis prediction models learn patient population knowledge and ignore the unique situation of individual patients except for common trends,this thesis proposes a mixed model GRUGP by combining Gated Recurrent Unit（GRU）and Gaussian Process（GP）to implement prognosis prediction of elderly patients with sarcopenia.The GRU neural network models global trends in patient populations,and GP models specific knowledge of individual patient.In the simulation experiments of the Physionet Challenge 2012 dataset and the sarcopenia dataset,combined with GP,the RMSE of the prognosis prediction model can be reduced by an average of 0.070.3.Aiming at the dependence of treatment recommendation models on the accuracy of a single decision and the high variance of decision-making,a Dual-Agent Gated Recurrent Unit（DAGRU）neural network is proposed based on the GRU network using dual-agent reinforcement learning.The historical state is selected by two agents based on different environments to implement skip connections,and the hidden state is initialized by using patient demographic information.In the simulation experiments of MIMIC-III dataset and sarcopenia dataset,the micro AUC,macro AUC and weighted AUC of the DAGRU model are 0.8268,0.7802,and 0.7733,respectively.4.Using Java language and My SQL database,based on Spring Boot and other frameworks,the decision support system for sarcopenia is designed and implemented,which has the functions of medical record management,risk factor analysis,patient diagnosis and intervention to implement the information management of elderly patients with sarcopenia and provide decision support for doctors.