Skeleton-Based Continuous Scale Parkinsonian Gait Score Estimation Using Omni-Dimensional Self-Attention Convolution...

Area of Science:

• Biomedical Engineering
• Neurology
• Computer Science

Background:

• Gait abnormalities are a key motor symptom in Parkinson's disease (PD).
• Current methods using skeleton data to estimate Movement Disorder Society Unified Parkinson's Disease Rating Scale (MDS-UPDRS) scores often simplify prediction to classification, missing subtle changes.
• Existing approaches may overlook clinical insights and localized gait features, impacting performance and continuous assessment for treatment planning.

Purpose of the Study:

• To develop a novel framework for continuous-scale gait impairment estimation from discrete MDS-UPDRS annotations using skeleton data.
• To improve the accuracy and clinical utility of gait assessment in Parkinson's disease.
• To enable better treatment adjustments through precise, continuous monitoring of gait changes.

Main Methods:

• Converted non-Euclidean skeleton data into Euclidean spatiotemporal feature maps to preserve structure.
• Utilized an omni-dimensional attention convolutional network to extract local spatiotemporal gait features.
• Integrated features using an adaptive channel feature fusion module and proposed a score prediction strategy using MDS-UPDRS anchors for continuous estimation.

Main Results:

• The proposed framework successfully estimates gait impairment on a continuous scale.
• The method effectively captures subtle, progressive gait changes over time.
• Validation on a large clinical PD gait skeleton dataset demonstrated the approach's effectiveness.

Conclusions:

• The novel framework provides a more nuanced and continuous assessment of Parkinson's disease gait impairment.
• This approach overcomes limitations of discrete classification methods and enhances clinical applicability.
• The continuous estimation facilitates more informed treatment strategies and patient management.