Design And Implementation Of A Biofeedback Treatment Assistant System Based On Data Mining Technology

With the rapid development of biomedical engineering and increasing improvingmeasurement instrumentation technology, medical data can be recorded accuratelyresulting in explosive growth of medical data. Therefore data mining techniques arewidely used in the medical area to discover the potential of knowledge in the massdata. Sun Yat-Sen University School of Medicine presented the emerging bio-feedback therapy intervention of pre-hypertension in interdisciplinary projects incooperation with School of Software. During the treatment, doctors guide the patientsto adjust their heart rate variability (HRV) in order to achieve the goal of adjustingblood pressure. This treatment method has been accumulated (and will continue toaccumulate) a large number of treatment plans, bio-signals data. Motivation of thispaper is to apply data mining technology on the treatment data in the purpose topredict the effectiveness of the treatment as well as the HRV value after treatment.With the prediction this paper can provide a basis for guiding decision-making for adoctor when conduct a treatment process.After learning the domain knowledge of biofeedback, this paper extracts twodata mining tasks and establishes the classification and prediction models. The mainprocess is as following:1)After conducting feature selection according to the domainknowledge and data preprocessing, this paper establish the therapeutic effectivenessprediction model in order to improve the relevance of medical treatment. This paperhas compared the C4.5algorithm and the random forest algorithm. Experiments haveshown accuracy of the random forest algorithm model is higher than that ofC4.5;2)using patients’ signs values before and during the treatment to establish HRVregression prediction model to help doctors more accurately locate the HRV target value. This paper has compared the BP neural network and multiple regressionalgorithms. Experiments have shown that root mean square error of BP neuralnetwork algorithm is smaller than that of multiple regression algorithm;3)finally，this paper designs and implements biological feedback treatment assistant systemsapplying random forest algorithm and BP neural network algorithm model on thesystem with a better human-computer interaction and visualization of operationalprocedure.This paper has conducted the empirical study to evaluate the effectiveness of theprediction and classification model in the system. This paper has collected the doctor’sprediction in26treatment instances of patients and compared that to model’sprediction. The result has shown that model’s prediction’s accuracy is higher that ofdoctors. Hence the model has played a guide role during the treatment and there is acertain degree of clinical significance to doctors. At the same time biofeedbacktreatment assistant system implemented in this paper apply data mining technology inpractice. The integrated function of the mood questionnaire in the course of treatment,blood pressure contrast has optimized the process of medical treatment and improvedthe efficiency of research efficacy.