Title Page
Abstract
Contents
PART Ⅰ. Introduction 14
Chapter 1. Human motor behavior and brain 14
1.1. Nervous Systems for Human Motor Behavior 14
1.2. Cortical Processing for Motor Control 18
1.3. Brain-Computer Interface System 21
Chapter 2. Application of Artificial Intelligence to Develop Neuroscience and Neuro-engineering Research. 24
2.1. Artificial Intelligence and Machine Learning 24
2.2. Deep Learning (Deep Neural Network) 26
2.3. Explainable Artificial Intelligence (XAI) 31
Chapter 3. Purpose of the study 33
PART Ⅱ. Experimental Studies 38
Chapter 4. Decoding Motor Kinematics from Cortical Source Signals Using Deep Neural Network Model 38
4.1. Chapter Introduction 38
4.2. Materials and Methods 40
4.3. Results 48
4.4. Discussion 50
Chapter 5. Identification of Cerebral Cortices for Processing Motor Kinematics Using DNN and Explainable Artificial Intelligence 54
5.1. Chapter Introduction 54
5.2. Materials and Methods 55
5.3. Results 58
5.4. Discussion 63
PART Ⅲ. General Conclusion and Perspectives 77
Chapter 6. General Conclusion and Perspectives 77
6.1. General Conclusion 77
6.2. Perspectives 77
Bibliography 83
Abstract in Korean 94
Table 4-1. A detailed description of the DNN model's architecture 46
PART Ⅰ. Introduction 12
Chapter 1. Human Motor Behavior and Brain 12
Figure 1-1. Organization of the human nervous system. 15
Figure 1-2. Lobes of human cerebrum 16
Figure 1-3. Dividing cerebral cortex into anatomical or functional organization. 17
Figure 1-4. Hierarchical organization of descending spinal tracts and their origins with the motor loop of brain-basal ganglia-thalamus. 18
Figure 1-5. A somatotopic organization in the precentral gyrus. Such representation also is called as cortical homunculus. 20
Figure 1-6. The methods to measure the brain signals for BCI. 22
Chapter 2. Application of Artificial Intelligence to Develop Neurosciecne and Neuro-engineering Research 12
Figure 2-1. A Venn diagram showing relationship between AI and its subfields and representative examples of each field. 25
Figure 2-2. Structures and computational mechanism of biological and artificial neurons. 27
Figure 2-3. Multi-layer perceptron (MLP). 29
Figure 2-4. An example of explainable artificial intelligence (XAI) in the image classification model. 32
PART Ⅱ. Experimental Studies 12
Chapter 4. Decoding Motor Kinematics from Cortical Source Signals Using Deep Neural Network Model 12
Figure 4-1. Deep Neural Network based decoding model. DNN models for decoding kinematic trajectories from neural signal are described. 43
Figure 4-2. Decoding accuracy. 48
Figure 4-3. The 2D plots of real and predicted hand-reaching acceleration (A) and velocity (B). 50
Figure 4-4. Normalized trajectories of hand-reaching position recovered from decoded kinematic parameters per axis and directions. 51
Figure 4-5. Recovered trajectories from motor kinematic parameters. The first on the left side shows real trajectories. Others are decoded. 52
Chapter 5. Identification of Cerebral Cortices for Processing Motor Kinematics Using DNN and Explainable Artificial Intelligence 13
Figure 5-1. Shared areas. Color-marked (green) areas are shared areas that significantly contribute to every kinematic parameter. 60
Figure 5-2. Acceleration dominant areas that contribute to decoding for reaching acceleration only. 61
Figure 5-3. Velocity dominant areas that contribute to decoding for reaching velocity only. 62
Figure 5-4. Position dominant areas that contribute to decoding for reaching position only. 63
PART Ⅲ. General Conclusion and Perspectives 13
Chapter 6. General Conclusion and Perspectives Signals Using Deep Neural Network Model 13
Figure 6-1. Signal translation using MelGAN model. (A) Averaged signals (evoked response) of each subject. The shaded areas represent the standard deviation. 79
Figure 6-2. Decoding real and generated brain signals through deep neural network (DNN) model. 81