The purpose of this study is to identify the gait indices associated with the knee flexion angle during SP of hemiplegic gait by machine learning.
The subjects were 77 stroke patients, and three-dimensional gait analysis was used to calculate 118 gait indices. The Fugl-Meyer assessment lower extremity (FMA-LE) was measured to assess paralysis. The Mahalanobis distance (MD) consisting of any two gait indices out of the 118 gait indices was calculated based on the data of healthy subjects. Next, the gait indices that maximized the correlation coefficient between the knee flexion angle during SP and the MD was extracted using the Markov chain Monte Carlo (MCMC) method. Then, K-means clustering was implemented using the top five extracted gait indices in MCMC method. After testing for the correlation between the top five extracted indices and knee flexion angle during SP in the overall patients and differences between clusters with FMA-LE and knee flexion angle during SP, the correlation between the top five extracted indices and knee flexion angle during SP was analyzed respectively, within each cluster.
By the MCMC method, the peak paretic ankle eccentric power (r=0.70) and the peak paretic propulsion (r=0.60), the paretic hip concentric power (r=0.54) during the loading response phase, the peak paretic hip concentric power (r=0.63), paretic knee flexion angle during toe-off (r=0.85) were extracted as the top five indices significantly related to the paretic knee flexion angle during SP. Patients were clustered into cluster A (55 cases) and B (22cases). Cluster A had significantly lower FMA-LE scores than Cluster B, and significantly decreased knee flexion angle during SP. In Cluster A, the peak paretic ankle eccentric power (r=0.37) and paretic knee flexion angle during toe off (r=0.90) were significantly associated with knee flexion angle during SP, whereas in Cluster B, the peak paretic hip concentric power (r=0.45) was significantly associated with knee flexion angle during SP.
In patients with severe paralysis and reduced knee flexion angle during SP, paretic anterior tilt of lower leg by eccentric power of the triceps muscles in the stance phase and improvement of paretic knee flexion angle during push-off may be an effective treatment, while in patients with mild paralysis and a large knee flexion angle during SP, improvement of swinging by the hip flexor muscles during the late stance phase may be an effective intervention.
The results of this study suggest that it is necessary to consider which lower joints to focus on depending on the causes of the patient's gait disturbance.
knee flexion angle
machine learning