Matthew Wassall M.Wassall@edu.salford.ac.uk
Matthew Wassall M.Wassall@edu.salford.ac.uk
Dr Sibylle Thies S.Thies@salford.ac.uk
Supervisor
Prof Malcolm Granat M.H.Granat@salford.ac.uk
Supervisor
Saeed Zahedi
Supervisor
The prosthesis components a lower limb amputee receives are determined by their assigned K level. K levels range from K0 to K4 and are defined by the user’s ability to traverse environmental barriers, change cadence and ambulation skill. K levels are assessed during a single clinic visit which varies between clinics but mostly utilizes conversation and basic mobility assessments. There are known issues with the reliability of K level assignment, especially when deciding between a K2 and K3. It has been shown that if a lower limb prosthetic user is not given an adequate prosthetic that meets their activity needs it could lead to the patient becoming less active and/or not using their prothesis. This PhD aims to create a sensor-based system to assess a patient’s activity levels in the real-world to reduce the issues with reliability during K level assignment.
As a first step, to fully understand the requirements of the system, a study was carried out where interviews were conducted with clinical experts. The ability of the patient to vary their cadence, traverse different terrain, walk without a walking aid and also the distance they can walk were emphasised by clinicians as the main differences between a K2 and K3 patient, and would constitute the data that would be required from the proposed sensor-based system (presently these are only be assessed via self-report).Of these measures only cadence is specifically stated in the current K level definitions, which suggests that the current K levels definitions do not meet the clinical needs.
A review was then conducted to identify the specification of the system in terms of algorithms and sensors that can be used to provide the required data. The review found that cadence has previously been measured with a shank mounted IMU. Moreover, body-worn IMUs have been used to accurately identify between flat ground, stairs, and ramps. However, walking on uneven terrain has not previously been classified for amputee gait. Furthermore, no studies could be found that identified walking aid use using appropriate body or prostheses mounted sensors. The review also looked at what method would be best to process and analyse the data. It was found that K-nearest neighbour, support vector machines, random forest and long short term memory neural networks are classifiers that have previously shown success with similar problems. Using a low pass filter and breaking the data into windows has also shown to be beneficial.
As a second step, a study was conducted to inform further system development, split into two parts. The first part was concerned with collection of data from lower limb prosthesis users in supervised real-world conditions. For this the participants had sensors attached and then were asked to traverse a range of set terrains, with and without a walking aid, outdoors. These data were then been used to train the classifiers. It was found that the terrain a lower limb prosthetic user is traversing can be classified using a single IMU mounted on the prosthetic shank, but walking aid use cannot be classified to clinically acceptable accuracies using a single IMU. A random forest model produced the highest terrain classification accuracies.
The second part of the study was conducted in a gait lab. For this part the participants were asked to traverse similar terrains, with and without a walking aid, but with sensor and full motion capture data being collected. These data have been used to create virtual sensors that were then used to estimate the ideal IMU location to increase classification accuracy. Feature importance was used to identify the most important aspects of the accelerations and then variations in these parts of the acceleration data were examined for different locations on the prosthetic shank. It was discovered that a consistent location is critical for high classification accuracies, and that for terrain classification accelerations captured at the ankle produce higher accuracies.
A final study was conducted to explore clinical experts’ views on the developed system and the output data. Real-world data was collected from 3 lower limb prosthetic users over two weeks using a prosthesis shank-worn IMU. These data were processed and classified using the previously created algorithms, to estimate terrain and walking aid use. Each participant also had their K level clinically assessed using current standard clinical procedures. A report was compiled for each participant that summarised the clinical assessments and the classification data. These reports were shown to 4 clinical experts and semi-structured interviews were conducted to assess their thoughts on the data, if the system would be clinically useful and if the data would change their K level assessments. All the interviewees thought the data would help with clinical assessments and had positive views about the data that were produced. They also all commented on how the data would change how they would conduct a K level assessment and two said that for two of the participants the sensor data would change the K level they would assign compared to just the clinical assessment. Active time and load through the prosthetic were the only measures that were identified that could further improve the system for clinical use.
Thesis Type | Thesis |
---|---|
Online Publication Date | Apr 24, 2025 |
Deposit Date | Apr 4, 2025 |
Publicly Available Date | May 25, 2025 |
Award Date | Apr 24, 2025 |
This file is under embargo until May 25, 2025 due to copyright reasons.
Contact M.Wassall@edu.salford.ac.uk to request a copy for personal use.
S Thies BGS2024 Video ppt
(2024)
Data
Why does my prosthetic hand not always do what it is told?
(2022)
Journal Article
Methods for clinical evaluation
(2020)
Book Chapter
About USIR
Administrator e-mail: library-research@salford.ac.uk
This application uses the following open-source libraries:
Apache License Version 2.0 (http://www.apache.org/licenses/)
Apache License Version 2.0 (http://www.apache.org/licenses/)
SIL OFL 1.1 (http://scripts.sil.org/OFL)
MIT License (http://opensource.org/licenses/mit-license.html)
CC BY 3.0 ( http://creativecommons.org/licenses/by/3.0/)
Powered by Worktribe © 2025
Advanced Search