Automatic Subtask Segmenation of the L Test

Abstract

Data gathered from inertial measurement units can be used to design subtask segmentation algorithms that further analyse functional mobility tests. This method has been shown to be effective for the 6-minute walk test, and the timed up-and-go test (TUG), and have produced gait analysis and fall risk models, especially in an elderly population. While further research has recently been published to assess individuals with a more asymmetrical gait, such as lower limb amputees, subtask segmentation of functional mobility tests such as the L Test of Functional Mobility (L Test), which are recommended for lower limb amputees, has not been investigated. With a decreased ceiling effect in comparison to the TUG, and increased turn requirements, the L Test would be able to provide a more in-depth assessment of a patient's mobility. In this thesis, a rule-based subtask segmentation algorithm was designed and tested with data from both able-bodied individuals and lower limb amputees. The algorithm produced acceptable results for able-bodied individuals (97% accuracy, > 98% specificity, > 74% sensitivity), but had low sensitivity for data from lower limb amputees (93-97% accuracy, 97-99% specificity, 33-60% sensitivity). A machine learning algorithm was then trained on data from both able-bodied and lower limb amputee participants and tested on data from a lower limb amputee population. This algorithm produced improved results for lower limb amputee participants (> 85% accuracy, > 75% sensitivity, > 95% specificity). This research designed and assessed both a rule-based and a machine learning algorithm for subtask segmentation of the L Test using data collected from both able-bodied and lower limb amputee participants. Overall, this thesis contributes to the progression of movement analysis for lower limb amputees, and to the understanding of motion during an L Test.