Classification of the Gait Pattern in Polymyalgia Rheumatica Patients Using Recurrent Neural Networks

Abstract

The early detection of movement disorders is essential for clinicians in many diseases, and it forms an integral part of effective treatment planning for patients. Polymyalgia rheumatica (PMR) is an autoimmune musculoskeletal disease that affects muscles around the pelvic girdle and shoulder blade. It is currently unknown how the strained hip muscles around the pelvic girdle create mobility limitations in patients. This study presents an algorithm for the classification of the hip muscle activation pattern in clinical gait analysis using recurrent neural networks (RNNs). RNNs was chosen because of its ability to capture temporal dependencies and process sequential electromyography (EMG) data in gait classification. A clinical gait assessment was conducted at KATH hospital which collected 250 gait segments from 18 PMR patients and 7 healthy control subjects. EMG signals were recorded from the vastus lateralis (VL), rectus femoris (RF), biceps femoris (BF), and semitendinosus (SE). Different optimizers were used in the RNN model to classify the hip muscle activation of the two groups to discriminate the gait pattern. Four optimizers (Adamax, Adagrad, SGD, and RMSprop) were used to evaluate the best optimizer for the RNN model. The accuracy results recorded from a cross-validation were, Adamax = 89%, Adagrad = 83%, SGD = 85%, and RMSprop = 78%. Adamax was the best performing optimizer while RMSprop was the least performing in the gait classification. An average accuracy of 84% from the four optimizers was sufficient to distinguish the gait pattern of the two groups. The findings of this study are useful in discriminating gait patterns based on hip muscle activation. This will provide essential information for the early detection of gait impairments by clinicians to make more informed a nd ti me ly decisions.