Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Parkinson gait modelling from an anomaly deep representation

Edgar Rangel, Fabio Martinez

TL;DR

This paper tackles the challenge of PD gait analysis under data scarcity by framing Parkinson's as an anomaly relative to normal gait. It introduces a self-supervised, one-class learning pipeline built around a volumetric autoencoder and a reconstruction-based discriminative task to learn a digital gait biomarker from control gait videos. The approach is complemented by a statistical validation framework (Chi-square shapeness and homoscedasticity) to assess generalization across cohorts and unseen data, achieving ROC-AUC up to $95\%$ on internal data and meaningful cross-domain performance (e.g., $75\%$ AUC on an external dataset). The work demonstrates the potential for generalizable, reconstruction-based digital biomarkers for PD gait that require fewer labels and may adapt to real-world acquisition variations.

Abstract

Parkinson's Disease (PD) is associated with gait movement disorders, such as bradykinesia, stiffness, tremors and postural instability, caused by progressive dopamine deficiency. Today, some approaches have implemented learning representations to quantify kinematic patterns during locomotion, supporting clinical procedures such as diagnosis and treatment planning. These approaches assumes a large amount of stratified and labeled data to optimize discriminative representations. Nonetheless these considerations may restrict the approaches to be operable in real scenarios during clinical practice. This work introduces a self-supervised generative representation to learn gait-motion-related patterns, under the pretext of video reconstruction and an anomaly detection framework. This architecture is trained following a one-class weakly supervised learning to avoid inter-class variance and approach the multiple relationships that represent locomotion. The proposed approach was validated with 14 PD patients and 23 control subjects, and trained with the control population only, achieving an AUC of 95%, homocedasticity level of 70% and shapeness level of 70% in the classification task considering its generalization.

Parkinson gait modelling from an anomaly deep representation

TL;DR

This paper tackles the challenge of PD gait analysis under data scarcity by framing Parkinson's as an anomaly relative to normal gait. It introduces a self-supervised, one-class learning pipeline built around a volumetric autoencoder and a reconstruction-based discriminative task to learn a digital gait biomarker from control gait videos. The approach is complemented by a statistical validation framework (Chi-square shapeness and homoscedasticity) to assess generalization across cohorts and unseen data, achieving ROC-AUC up to on internal data and meaningful cross-domain performance (e.g., AUC on an external dataset). The work demonstrates the potential for generalizable, reconstruction-based digital biomarkers for PD gait that require fewer labels and may adapt to real-world acquisition variations.

Abstract

Parkinson's Disease (PD) is associated with gait movement disorders, such as bradykinesia, stiffness, tremors and postural instability, caused by progressive dopamine deficiency. Today, some approaches have implemented learning representations to quantify kinematic patterns during locomotion, supporting clinical procedures such as diagnosis and treatment planning. These approaches assumes a large amount of stratified and labeled data to optimize discriminative representations. Nonetheless these considerations may restrict the approaches to be operable in real scenarios during clinical practice. This work introduces a self-supervised generative representation to learn gait-motion-related patterns, under the pretext of video reconstruction and an anomaly detection framework. This architecture is trained following a one-class weakly supervised learning to avoid inter-class variance and approach the multiple relationships that represent locomotion. The proposed approach was validated with 14 PD patients and 23 control subjects, and trained with the control population only, achieving an AUC of 95%, homocedasticity level of 70% and shapeness level of 70% in the classification task considering its generalization.
Paper Structure (15 sections, 5 equations, 5 figures, 5 tables, 1 algorithm)

This paper contains 15 sections, 5 equations, 5 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Current Work
  3. Proposed approach
  4. A volumetric autoencoder to recover gait embedding patterns
  5. Auxiliary task to discriminate reconstructions
  6. A Digital gait biomarker from anomaly embeddings
  7. Statistical validation setup
  8. Shape analysis from ChiSquare
  9. Variance analysis from Homoscedasticity
  10. Experimental setup
  11. Datasets
  12. External dataset validation
  13. Evaluations and Results
  14. Discussion
  15. Conclusions

Figures (5)

  • Figure 1: Pipeline of the proposed model separated in volumetric auto-encoder to recover gait patterns (a), Digital gait biomarker (b), Auxiliary task to discriminate reconstructions (c), and statistical validation of learned classes distributions (d)
  • Figure 2: Data distribution given by the proposed model for control and parkinson samples by Hoehn & Yahr levels.
  • Figure 3: Original, reconstructed and anomaly video of two patients. The first row is a control and second row is a parkinson patient. The 3 first columns are the gait from left to right while the last 3 columns when is right to left.
  • Figure 4: Data distribution given by the proposed model for control, parkinson (PD) and knee-osteoarthritis (KOA) samples by levels where EL is early, MD medium and SV severe.
  • Figure 5: Original, reconstructed and anomaly video of three patients. The first row is a control, second row parkinson and third row is a knee osteoarthritis patient. The 3 first columns are the gait from left to right while the last 3 columns when is right to left.

Theorems & Definitions (1)

  • Remark 1