Background

Although deep neural networks (DNNs) are showing state of the art performance in clinical gait analysis, they are considered to be black-box algorithms. In other words, there is a lack of direct understanding of a DNN’s ability to identify relevant features, hindering clinical acceptance. Interpretability methods have been developed to ameliorate this concern by providing a way to explain DNN predictions.
Methods

This paper proposes the use of an interpretability method to explain DNN decisions for classifying the movement that precedes freezing of gait (FOG), one of the most debilitating symptoms of Parkinson’s disease (PD). The proposed two-stage pipeline consists of (1) a convolutional neural network (CNN) to model the reduction of movement present before a FOG episode, and (2) layer-wise relevance propagation (LRP) to visualize the underlying features that the CNN perceives as important to model the pathology. The CNN was trained with the sagittal plane kinematics from a motion capture dataset of fourteen PD patients with FOG. The robustness of the model predictions and learned features was further assessed on fourteen PD patients without FOG and fourteen age-matched healthy controls.
Results

The CNN proved highly accurate in modelling the movement that precedes FOG, with 86.8% of the strides being correctly identified. However, the CNN model was unable to model the movement for one of the seven patients that froze during the protocol. The LRP interpretability case study shows that (1) the kinematic features perceived as most relevant by the CNN are the reduced peak knee flexion and the fixed ankle dorsiflexion during the swing phase, (2) very little relevance for FOG is observed in the PD patients without FOG and the healthy control subjects, and (3) the poor predictive performance of one subject is attributed to the patient’s unique and severely flexed gait signature.
Conclusions

The proposed pipeline can aid clinicians in explaining DNN decisions in clinical gait analysis and aid machine learning practitioners in assessing the generalization of their models by ensuring that the predictions are based on meaningful kinematic features.

Background
Manual annotation of initial contact (IC) and end contact (EC) is a time consuming process. There are currently no robust techniques available to automate this process for Parkinson's disease (PD) patients with freezing of gait (FOG).

Objective
To determine the validity of a data-driven approach for automated gait event detection.

Methods
15 freezers were asked to complete several straight-line and 360 degree turning trials in a 3D gait laboratory during the off-period of their medication cycle. Trials that contained a freezing episode were indicated as freezing trials (FOG) and trials without a freezing episode were termed as functional gait (FG). Furthermore, the highly varied gait data between onset and termination of a FOG episode was excluded. A Temporal Convolutional Neural network (TCN) was trained end-to-end with lower extremity kinematics. A Bland-Altman analysis was performed to evaluate the agreement between the results of the proposed model and the manual annotations.

Results
For FOG-trials, F1 scores of 0.995 and 0.992 were obtained for IC and EC, respectively. For FG-trials, F1 scores of 0.997 and 0.999 were obtained for IC and EC, respectively. The Bland-Altman plots indicated excellent timing agreement, with on average 39% and 47% of the model predictions occurring within 10 ms from the manual annotations for FOG-trials and FG-trials, respectively.

Significance
These results indicate that our data-driven approach for detecting gait events in PD patients with FOG is sufficiently accurate and reliable for clinical applications.