Exploring gait analysis and deep feature contributions to the screening of cervical spondylotic myelopathy

In the cervical region of middle-aged and elderly patients, cervical spondylotic myelopathy (CSM) is frequently recognized as the primary factor that contributes to spinal cord dysfunction. Numbness and gait disturbance are the main clinical manifestations of CSM, which exhibits as a stiff and spastic gait in comparison with that of healthy controls (HCs). Because it is difficult to screen CSM in the primary stage which easily leading to a delay in medication, the identification of CSM followed by treatment is urgent. The aim of this study is to develop an automated classification method for the screening of CSM, using fifty-four lower extremity kinematic parameters derived from three-dimensional gait analysis. The present study employs a deep neural network (DNN) model to automatically extract informative features from raw gait kinematic data. Hierarchically placed layers in the DNN produce deep feature maps that are used to screen CSM using multiple shallow classifiers. The proposed method is evaluated using a self-constructed gait database of patients diagnosed with CSM and HCs, both groups consisting of 45 individuals within a similar age range. Experimental results reveal that the combination of deep features and shallow classifiers yields remarkable accuracy rates for binary classification with twofold, tenfold, and leave-one-out cross-validation methods, all achieving an accuracy of 99.44 % . The data suggest that our approach is efficient in detecting the early onset CSM and performs better than other cutting-edge techniques.