Fig. 1From: Modelling and identification of characteristic kinematic features preceding freezing of gait with convolutional neural networks and layer-wise relevance propagationVisualization of the proposed methodology. The proposed methodology consists of two-stages (1) a convolutional neural network (CNN) to model the dramatic reduction of movement present before a freezing of gait (FOG) episode (Phase 2), and (2) layer-wise relevance propagation (LRP) to interpret the underlying features that the CNN perceives as important to model the pathology (Phase 3). The CNN was trained with the sagittal plane kinematics as recorded by a motion capture system (Phase 1). The figure illustrates the benefit of interpretation in a deep learning frameworkBack to article page