Author + information
- 1National Key Laboratory of Complex System Intelligent Control and Decision, School of Automation, Beijing Institute of Technology
- 2Department of Cardiology Internal Medicine, Nanlou Branch of Chinese PLA General Hospital
- 3Mathematical Institute, University of Oxford
- 4Department of Cardiology, Beijing Haidian Hospital, Peking University No.3 Hospital Haidian Branch
Based on a new fuzzy neural network method fused modified Takagi-Sugeno model and BP neural network algorithm, sex hormones level and fatness characteristic in old males are trained and studied, while the database system for the relationship between sex hormones and hypertension in old males is built to assist doctors' decision by the analysis and estimation.
(1) Establish the model database system by SQL and VS software. From hospitals' existing databases, 330 subjects were involved including 180 hypertension patients and 150 health controls, recording clinical data and detection, such as height, weight, age, estradiol (E2), total testosterone (TT) and so on. (2) Modify the model analysis method, design experiments to determine the accuracy of the algorithms and improve the model database system. The membership function of basic Takagi-Sugeno is modified by self-adaptive Gaussian function, while the parameter in particle swarm optimization (PSO) algorithm is adjusted by extended Takagi-Sugeno model. Extract 200 data sets (100 hypertension patients and 100 health controls) as the training sample, 80 data sets (50 hypertension patients and 30 health controls) as the predicting data, and 50 data sets (30 hypertension patients and 20 health controls) as the test sample. The design variable is determined by the hidden layer neurons number of BP neural network, the mean square error after extraction as the evaluation function. Finally, optimal value is searched with the improved PSO algorithm, recorded in the model database system for the further diagnosis.
By means of the new fuzzy neural network method, the relationship of detected parameters is presented obviously. The levels of plasma total testosterone (TT) and body mass index (BMI) were increased significantly in hypertension group compared with control group (P<0.05), the level of TT is related to androgen receptor and BMI, positively correlated with androgen receptor (P<0.01) and negatively correlated with BMI (P<0.01). According to the sample points including feature information for patients, the disease probabilities are showed by the cluster in the model database system for doctors' decision by the analysis and prediction.
The new fuzzy neural network method fused modified Takagi-Sugeno model and BP neural network algorithm has high accuracy and sensitivity in the process of hypertension diagnosis, meanwhile supply an interpretable judgment. Based on the model database system, preliminary conclusions can be obtained quickly from clinical examination, which is effective to assist doctors' decision.