Skip to main content
Fig. 2 | BMC Medical Informatics and Decision Making

Fig. 2

From: Modelling and identification of characteristic kinematic features preceding freezing of gait with convolutional neural networks and layer-wise relevance propagation

Fig. 2

Mean and standard deviation of the hip, knee, and ankle joint trajectories in the sagittal plane for six of the seven freezers who experienced FOG during the protocol (a), with the excluded subject discussed separately (b), and the fourteen non-freezers and fourteen healthy control subjects (c). The joint trajectories are colorized with the relevance map (heatmap) \(\sum _{x} R_{x}^{(1)}\) using \(\epsilon\)-LRP. To ensure an equal contribution, six strides (three pre-FOG and three FGC) are used of each freezer, with exception of subject seven who only froze once. For the non-freezers (NF) and healthy control (HC) subjects, all 2421 and 2258 strides were used. For the attribution plots of the freezers (a and b), the error clouds depict the standard deviations of the pre-FOG trajectories (gray) and FGC trajectories (green). For the attribution plots of the NF and HC (c), the error clouds depict the standard deviations of NF trajectories (green) and HC trajectories (gray). Positive relevance (red) indicates contribution to FOG, while negative relevance (blue) indicates contribution to FGC

Back to article page