Motion Intention Estimation using EEG for Rehabilitation

Project Title

Motion Intention Estimation using EEG for Rehabilitation

Project Area of Specialization RoboticsProject Summary

Workaholic lifestyle and heavy stress in a competitive environment is becoming more and more common [1]. It causes hypertension, stress which further leads to stroke and brain hemorrhage. The patients being affected to these conditions, can be recovered and rehabilitated. This study focuses on the development of rehabilitation robots for such patients. In general, conventional rehabilitation process is very laborious as well as expensive, hence even a problem for developed economies.

To empower the therapists to perform effective therapy, robots have been developed [2]. However, these robots do not understand patients and cannot adjust assistive force as per their requirements and understand patients’ motion intention. For this reason, methods have been proposed such as surface Electromyography (sEMG) and EEGs. However, they have been poorly incorporated to help the robot to assist in a compliant way. It is because sEMGs and EEGs have many inherent issues [3].

In this study, classifiers like Support Vector Machine, Naïve Bayesian Algorithm, Decision Tree, Random Forest and Deep Learning Algorithm have been examined and applied on ten persons EEG to find the effectivity of human motion intention in compliance control. Outliers, noise and artefacts are removed, and data is categorised for outputs such as hip joint motion (Extension/Flexion) and Knee Joint Motion (Extension/Flexion).  To train the models, dimensionality reduction techniques have been used such (Principle Component Analysis/ Linear Discriminant Analysis) and K-Means and Hierarchical Clustering is performed to find out the clusters for various motions.

The proposed methodology is implemented on a lower limb rehabilitation robot and is likely to improve the number of repetitions for a rehabilitation exercise as robot will not apply forces to exhaust a patient. So, these increased number of repetitions will cause quick recovery. This quick recovery will not only be helpful for the patient by obviously being rehabilitated but also will reduce the recovery cost as well. So, eventually, many patients could be treated and recovered.

References:

1. Eric S. Donkor, “Stroke in the 21st Century: A Snapshot of the Burden, Epidemiology, and Quality of Life,” Stroke Research and Treatment, vol. 2018, Article ID 3238165, 10 pages, 2018.
2. L. Masia, H. I. Krebs, P. Cappa and N. Hogan, "Design and Characterization of Hand Module for Whole-Arm Rehabilitation Following Stroke," in IEEE/ASME Transactions on Mechatronics, vol. 12, no. 4, pp. 399-407, Aug. 2007.
3. Lalitharatne, T., Teramoto, K., Hayashi, Y, Kiguchi, K, (2013). Towards Hybrid EEG-EMG-Based Control Approaches to be Used in Bio-robotics Applications: Current Status, Challenges and Future Directions. Paladyn, Journal of Behavioral Robotics, 4(2), pp. 147-154. Retrieved 13 Apr. 2019.

Project Objectives

1. Development of Lower Limb Rehabilitation Robot
2. Development of Machine Learning Algorithm for Compliance Control

The proposed study aims to increase lifespan for the physically disabled stroke patients so that they may be rehabilitated and be an effective part of society.

Development of Lower Limb Rehabilitation Robot

In the first objective, a lower limb rehabilitation robot needs to be developed. It requires designing of a robot, its dynamics and control, and eventually fabrication. For this purpose, a two-link serial manipulator structure is selected. It has been modified to meet the requirement of lower limb rehabilitation robot.  A simple trajectory-based control is selected.

Development of Machine Learning Algorithm for Compliance Control

To achieve compliance, motion intention is estimated using EEG signals which is the focus of this study.  To do this, EEG signals have been obtained for various subjects. These subjects were asked to perform a given motion such as walking. These signals are processed to remove outliers, artifacts and other interfering frequencies and noise. Further, these signals have been categorized according to output such as Extension/Flexion in hip joint or knee joint motion. Furthermore, PCA, LDA and Clustering Algorithms are utilized to train the model. Classification is also applied for predict the motion effectively. This would enable the robot to assist the patient according to the force necessary to complete a specific motion.

Project Implementation Method

The main challenge in development of rehabilitation robot has the following features:

1. The compliant based mechanical structure
2. control of rehabilitation robot to assist the patients.
3. Using intelligent Algorithms to carry out the task of estimating the motion the patient is trying to make.
4. Determining the best model to process the EEG signal to predict the motion being achieved.

Compliant based mechanical structure

The robot would in the shape of a wearable robotic lower limb which would assist a patient in performing a motion. The mechanical structure needed to be fabricated would have to withstand the weight of the patient to some extent and the cognitive interaction between it and the human would be bi-directional. The robot will learn of the intensity of motion required to complete a given motion and the human will learn to stimulate different motions which would not be possible in case of a missing limb. Ultimately, the human learns to operate the arm as a normal human limb.

Control of Rehabilitation Robot

Often in Rehabilitation applications, a patient’s comfort needs are overlooked in equipping them with an artificial limb. This issue can be dealt with a method involving an intelligent algorithm capable of deducing the intention the person is trying to achieve. In a rehabilitation Robot, when this algorithm is used, the Robot arm will naturally estimate the type and intensity of the motion the person is trying to achieve based on the previously available EEG data retrieved from various sources.

Using intelligent Algorithms to carry out the task of estimating the motion the patient is trying to make.

In the first phase of study, the EEG Dataset of lower limb of the patient will be acquired which will be used on Regression Algorithms. The same dataset will then be used on Classification Algorithms and a Clustering Algorithm. The purpose will be to see the linear/non-linear relationship being given by the Algorithm and a comparative study will be done regarding which Algorithm will be best suited to predicting the output better using the training set.

Determining the best model to process the EEG signal to predict the motion being achieved.

Observations regarding how using Clustering improves scores of Algorithms will be done and comparison between scores of Linear Regression, Classification and Clustering will take place. Subsequently, the concept of Deep Learning will be explored and how it be can use it to our advantage. Our main objective will be to assist the patient’s performance and to minimize the kinematic errors while performing the patient’s motion.

Benefits of the Project

The project will most likely benefit the following groups:

• Disabled people who wish to lead a respectable and independent lifestyle.
• Rehabilitation Centers/Physiotherapeutic Institutes
• Potential Entrepreneurs who might facilitate the society.
• Educators who might add the tools into their curriculum.

There will be multiple beneficiaries in this project, due to its wide scope. The main diaspora which will benefit from this study consist of the disabled today who are deemed helpless but wish to be able to re-join society and looked upon with dignity and respect. Secondly, Rehabilitation centers as well as physiotherapeutic institutes would benefit from this research. The beneficiaries consist of potential entrepreneurs who may gain insight as to how to facilitate the segment of society in focus and may benefit from this research. Rehabilitation and Medical centers which would employ these robots in aiding the disabled who have suffered conditions such as stroke or brain hemorrhage. Key idea behind the task is to maximize the efficiency in the day-to-day lives of stroke patients by assisting them in performing motions in a normal way by assistive movement. This assistive motion will lead to automated physiotherapy and will minimize the resistance of the motion being performed by patients.

Although, force control algorithms have been developed but still these robots cannot understand patients. They cannot read their minds or muscles. Purpose of the project is to provide disabled Individuals with effective yet achievable Rehabilitation methods, so we may be able to estimate the motion which a person is trying to perform which may further result in the task being more accurately performed. For this purpose, we may utilize Machine Learning to achieve this task and we also require deep expertise in certain segments of Artificial Intelligence, which may be further used to accomplish the task of strengthened decision-making and self-learning to be able to adjust in a dynamic environment by itself.

Finally, educators can use the tools in these findings and add it to their respective academic curriculum. If proper importance is given in relaying the tools and their usage method to students, it may help in improving of the work environment and help in transferring knowledge in the workplace.

Technical Details of Final Deliverable

The final wearable robotic arm needs will compensate for the arm’s inability to perform a proper motion, the reason being muscle weakness or other similar symptoms. The cognitive interaction between the human and the robotic arm is bi-directional. As the robot gathers the information from previously available EEG signal and the motion the patient tries to make, the human learns to operate the arm as a normal human limb.

There will be a need for a computing system having a considerable amount of RAM and high-level processing, such as Jetson AGX Xavier. Rehabilitation robot is a biomechanics device which would require subtle and fine manufacturing and design. Therefore, many conventional methods would not be used. The equipment would need to be carefully designed. Hence, it would require special motors for hip and knee joints, for which we would use EC60 Flat BLDC Motor and Maxon Motor with hall sensors and cables. To drive these motors, drivers along with harmonic drive system would be required, so that these motors could be interfaced with embedded computing system. For accurate positioning of the Robot, EDOS Digital Position Controller is used. Force and EMG Sensors are required to measure robot’s effectiveness on the patient in run time. These sensors would represent whether the robot is assisting the patient to acquire maximum involvement or not. Therefore, these sensors will play a crucial role in the project. The Project contains many parts with hefty expenditures which are mentioned below in detail.

1. Fabricating Hip-Joint Motor requires parts which includes:

•     Hip Joint Motor (Rs. 100,000)
•     Harmonic Drive (Rs. 60,000)
•     Driver (EPOS4 70/15) (Rs. 140,000)

1. Fabricating the Knee Joint Motor requires parts which includes:

•     Knee Joint Motor (Rs. 80,000)
•     Harmonic Drive (Rs. 70,000)
•     Driver (EPOS4 70/15) (Rs. 140,000)

1. Mechanical Structure Manufacturing and Fabrication cost (Rs. 200,000-300,000)
2. EEG Headset Neurosky or Equivalent (Rs. 40,000)
3. Batteries
4. Switches
5. Wires
6. Tools (Soldering Station, Wire Cutters, Screw Drivers etc.)

Since this project is quite large, and contains many parts and their respective expenditures, only assistance is required in Manufacturing/Fabricating costs which is about Rs. 200,000-300,000.

Final Deliverable of the Project HW/SW integrated systemType of Industry Medical Technologies Artificial Intelligence(AI), Robotics, NeuroTech, Wearables and ImplantablesSustainable Development Goals Good Health and Well-Being for People, Decent Work and Economic Growth, Industry, Innovation and Infrastructure, Sustainable Cities and CommunitiesRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
Fabrication/Manufacturing	Equipment	1	70000	70000
Stationary/Prinitng	Miscellaneous	1	10000	10000