Development and application of mathematical models in gait characterisation after stroke

The number of stroke has increased every years according to American Heart Association (AHA), World Health Organization (WHO) and National Stroke Association of Malaysia (NASAM). These numbers have raised concerns among medical and rehabilitation professionals who manage this neurological disorder. For this project, we aim to develop a sophisticated gait analysis system to help the recovery of stroke patients. This proposed gait analysis system can help clinicians to assess the gait pattern and plan a suitable rehabilitation treatment for stroke patients systematically. It started with the development of gait sensor system. In this study, we are interested to study about the kinematics and kinesiology parameters of stroke patients. Therefore, we developed a low-cost inertial based sensor system and employed commercial ShimmerSensing sEMG sensors. A 3D high-speed camera was used to validate the inertial based sensor system. The parameters obtained from the sensor system were further analysed to extract valuable features for gait characterisation and gait classification. Two new gait analysis methods, kinesiology and kinematic based gait analysis were proposed to study the characteristic of the stroke patient’s gait. For kinesiology based gait analysis, the surface EMG (sEMG) signal was being collected and analysed. We applied sliding window Higuchi Fractal Dimension (HFD) on sEMG signal and computed a new fractal based index, named Kinetic Index (K.I.). This K.I. was further correlated to the Timed Up and Go Test (TUG test). The results showed that K.I. is highly correlated to the TUG test. Besides that, K.I. can also classify stroke patients into three homogeneous subgroups by using Hierarchical Cluster Analysis. For kinematic based gait analysis, we proposed a new variant of the Symmetry Region of Deviation (SROD) method to quantify gait asymmetry. This new method, named as Cyclogram SROD (CSROD), applies a bilateral cyclogram of both left and right lower limbs gait data to compute the gait deviation from perfect symmetry. Compared to SROD, CSROD does not require a baseline gait database of normal healthy subjects for comparison purposes. Instead, it uses a 45° symmetry line in the cyclogram to indicate perfect gait symmetry. The validation results showed that the proposed method were similar to those obtained from the SROD method according to Welch t-test analysis. With proper gait alignment technique such as Dynamic Time Warping (DTW), the CSROD results showed the accurate timing and magnitude of the peaks where asymmetry occurred. Both the K.I. and CSROD provide valuable information regarding the kinesiology and kinematic status of the stroke patients. However, it cannot describe the difference of gait pattern between stroke patients and healthy subjects. Therefore, two new gait functionality indices, G_FunctGT and G_FunctTD were presented. These two indices detect the gait trajectory deviation and time delay between stroke and healthy. The features extracted for gait characterisation (K.I., CSROD, G_FunctGT and G_FunctTD) were used to develop two recovery prediction models. The first model used stroke patients baseline (stage 1) gait data to predict their third month (stage 2) and sixth month (stage 3) of gait indices. The second model was based on the recovery trajectory from baseline (stage 1) to third month (stage 2) to predict the final state of gait indices (stage 3). The results showed high accuracy among stroke patients. The sEMG signal on each stage of the stroke recovery period were further decomposed using Ensemble Empirical Mode Decomposition (EEMD) method. This was to study the muscle status changes across the recovery period on stroke patients. It is to ensure the recovery in joint motions associates with the recovery of muscles, and not due to muscle compensation