Stroke Evaluation And Reflection With Artificial Neural Network In Haptic Virtual Environment

Stroke rehabilitation refers to the patient regain some of limb motor function by training process, the correct evaluation of stroke patients body movement rehabilitation of stroke patients is very important. Artificial neural network (ANN) was for the assessment of post stroke, and raised questions about ANN is a black box, or white box modeling methods, and make analysis of theoretical guidance in parameter selection of ANN design. Finally, this thesis tried to do research on whether we can distinguish the patient is at the stage of the sixth stroke or health status using ANN model based on completion performance of two tasks "through the tunnel" and "follow the circle",The overall objectives of this thesis is to find answers to above questions. This paper mainly research on patients with stroke assessment methods and classification methods, and the specific objectives are:(i) artificial neural model is a black-box model or white-box model; (ii) Trying to make design flow of artificial network under classification problem; (iii) Establishing and evaluating an automatic classifier on stroke stage using artificial network..In this paper, the main conclusions are as follows:(1) we can establish an effective stroke prototype using BP neural network classification for stroke prediction, the expected accuracy rate is 94.12%, and the expected of healthy prediction accuracy is 100%; (2) the performance of classifier with 7 neurons in the implied layer is outstanding in the testing phase, AUC value of testing data is 0.876; (3)ANN may be a black-box or gray-box model, depending on the way we build neural network, or whether related knowledge is used in establishing ANN model; (4) with respect to particular problem, there is no golden rule, "trial and error" may be the golden rule to determine number of neuron in hidden layer.The main contributions of this paper are:(1) This paper demonstrate the essence of artificial neural network modeling methods, and provide guidance for the design flow of modeling; (2) formed the knowledge about self-assessment of stroke classifier using machine learning algorithms, and community-based self-directed assessment of stroke; (3) This study makes use of artificial network improving the accuracy of phase evaluation based on Brunnstrom theory, from 92.1% to 94.12%.