File
Brennan L1,2, Bevilacqua A2, Kechadi T2, Caulfield B2
1Beacon Hospital, Physiotherapy Department, Dublin, Ireland, 2University College Dublin, Insight Centre for Data Analysis, Dublin, Ireland
Background: Engaging in home-rehabilitation can be challenging for many patients, and low adherence to prescribed exercises programmes is often reported. Digital home-rehabilitation support systems designed to work alongside physiotherapy services may help provide additional motivation and guidance to patients. Including a movement sensor, such as an inertial measurement unit (IMU), in these systems can allow tracking of exercise sessions by collecting data such as repetition counts and exercise technique. This biofeedback can then be delivered to both patients and Physiotherapists. To date, no such home-rehabilitation based, single-sensor IMU system for shoulder rehabilitation has been developed. To do this, the collection of large amounts of IMU data for shoulder exercises and close interdisciplinary work with computer engineering professionals is required.
Purpose:
Methods: 35 healthy individuals (m=22, f=13) were recruited and fitted with an IMU in three separate locations on the upper limb: upper trapezius, arm and wrist. Participants performed 40 repetitions each of shoulder flexion, abduction and rotation rehabilitation exercises. IMU data was recorded throughout. This data was used to train a specially-designed machine learning algorithm and then to test the accuracy of this system.
Results: The three-sensor system could segment the repetitions with high accuracy for flexion (100%) and abduction (100%), and with moderate accuracy for rotation (83%). As a single-sensor system, the arm IMU performed best, successfully identifying repetitions of flexion and abduction 99% and 98% of the time respectively, and 74% of rotation repetitions.
Conclusion(s): Shoulder flexion and abduction could be detected and repetitions counted to a high level of accuracy using a single IMU. The rotation exercise did not achieve clinically acceptable results. This system will now be expanded to include a wider variety of rehabilitation exercises, and will be further refined to improve accuracy of rotation segmentation.
Implications: These results are an encouraging step forward in the development of a biofeedback home-rehabilitation support system for shoulder exercises. To maximise usability and engagement in a clinical setting, this system should be paired a user-friendly interface, such as a mobile or web application, in combination with educational and motivational features.
Keywords: Home rehabilitation, Technology, Inertial measurement unit
Funding acknowledgements: Part of CATCH ITN, which is funded by EU Horizon 2020 research and innovation programme. Marie Sklodowska-Curie grant agreement No722012.
Purpose:
- Collect IMU data for shoulder flexion, abduction and rotation exercises.
- Use this data to train a machine learning algorithm to segment the sensor signal to enable identification and counting of exercise repetitions.
- Assess the accuracy of this system.
- Identify if a single-sensor system delivers clinically acceptable results, and which sensor location delivers most accurate results.
Methods: 35 healthy individuals (m=22, f=13) were recruited and fitted with an IMU in three separate locations on the upper limb: upper trapezius, arm and wrist. Participants performed 40 repetitions each of shoulder flexion, abduction and rotation rehabilitation exercises. IMU data was recorded throughout. This data was used to train a specially-designed machine learning algorithm and then to test the accuracy of this system.
Results: The three-sensor system could segment the repetitions with high accuracy for flexion (100%) and abduction (100%), and with moderate accuracy for rotation (83%). As a single-sensor system, the arm IMU performed best, successfully identifying repetitions of flexion and abduction 99% and 98% of the time respectively, and 74% of rotation repetitions.
Conclusion(s): Shoulder flexion and abduction could be detected and repetitions counted to a high level of accuracy using a single IMU. The rotation exercise did not achieve clinically acceptable results. This system will now be expanded to include a wider variety of rehabilitation exercises, and will be further refined to improve accuracy of rotation segmentation.
Implications: These results are an encouraging step forward in the development of a biofeedback home-rehabilitation support system for shoulder exercises. To maximise usability and engagement in a clinical setting, this system should be paired a user-friendly interface, such as a mobile or web application, in combination with educational and motivational features.
Keywords: Home rehabilitation, Technology, Inertial measurement unit
Funding acknowledgements: Part of CATCH ITN, which is funded by EU Horizon 2020 research and innovation programme. Marie Sklodowska-Curie grant agreement No722012.
Topic: Robotics & technology; Human movement analysis; Information management, technology & big data
Ethics approval required: Yes
Institution: University College Dublin
Ethics committee: Human Research Ethics Committee
Ethics number: LS-17-64-Brennan-Caulfield
All authors, affiliations and abstracts have been published as submitted.
