A classification study of kinematic gait trajectories in hip osteoarthritis.

The clinical evaluation of patients in hip osteoarthritis is often done using patient questionnaires. While this provides important information it is also necessary to continue developing objective measures. In this work we further investigate the studies concerning the use of 3D gait analysis to attain this goal. The gait analysis was associated with machine learning methods in order to provide a direct measure of patient control gait discrimination. The applied machine learning method was the support vector machine (SVM). Applying the SVM on all the measured kinematic trajectories, we were able to classify individual patient and control gait cycles with a mean success rate of 88%. With the use of an ROC curve to establish the threshold number of cycles necessary for a subject to be identified as a patient, this allowed for an accuracy of higher than 90% for discriminating patient and control subjects. We then went on to determine the importance of each trajectory. By ranking the capacity of each trajectory for this discrimination, we provided a guide on their order of importance in evaluating patient severity. In order to be clinically relevant, any measure of patient deficit must be compared with clinically validated scores of functional disability. In the case of hip osteoarthritis (OA), the WOMAC scores are currently one of the most widely accepted clinical scores for quantifying OA severity. The kinematic trajectories that provided the best patient-control discrimination with the SVM were found to correlate well but imperfectly with the WOMAC scores, hence indicating the presence of complementary information in the two.

Modulation of anticipatory postural activity for multiple conditions of a whole-body pointing task.

This is a study on associated postural activities during the anticipatory segments of a multijoint movement. Several previous studies have shown that they are task dependant. The previous studies, however, have mostly been limited in demonstrating the presence of modulation for one task condition, that is, one aspect such as the distance of the target or the direction of reaching. Real-life activities like whole-body pointing, however, can vary in several ways. How specific is the adaptation of the postural activities for the diverse possibilities of a whole-body pointing task? We used a classification paradigm to answer this question. We examined the anticipatory postural electromyograms for four different types of whole-body pointing tasks. The presence of task-dependent modulations in these signals was probed by performing four-way classification tests using a support vector machine (SVM). The SVM was able to achieve significantly higher than chance performance in correctly predicting the movements at hand (Chance performance 25%). Using only anticipatory postural muscle activity, the correct movement at hand was predicted with a mean rate of 62%. Because this is 37% above chance performance, it suggests the presence of postural modulation for diverse conditions. The anticipatory activities consisted of both activations and deactivations. Movement prediction with the use of the activating muscles was significantly better than that obtained with the deactivating muscles. This suggests that more specific modulations for the movement at hand take place through activation, whereas the deactivation is more general. The study introduces a new method for investigating adaptations in motor control. It also sheds new light on the quantity and quality of information available in the feedforward segments of a voluntary multijoint motor activity.