Parkinson’s disease (PD) is a progressive neurodegenerative disorder of the central nervous system affecting both motor and non-motor functions. It is characterized by weakened and damaged neurons and eventually neuronal death. PD significantly impacts quality of life, thus in-home monitoring, particularly of Freezing of Gait (FoG), is paramount in symptom management and disease progression observations. Existing technologies for monitoring symptoms are power-hungry, operate in controlled settings and rely on large amounts of labeled data, which are scarce, limiting real-world deployment.
Researchers at Arizona State University have developed a novel computationally-efficient framework for real-time FoG detection called LIFT-PD (Label-efficient In-home Freezing-of-gait Tracking). This framework combines self-supervised pre-training on unlabeled data with a differential hopping windowing technique which learns from limited labeled instances. Power consumption is further optimized by activating a deep learning module only during active periods. Experimental results demonstrate that LIFT-PD achieves a 7.25% increase in precision and a 4.4% improvement in accuracy compared to supervised models. Further, it uses as little as 40% of the labeled training data compared to supervised learning. When compared to continuous inference, the model activation module reduces inference time by up to 67%.
LIFT-PD paves the way for practical, energy-efficient, and unobtrusive in-home monitoring of PD patients with minimal data labeling requirements.
Potential Applications
- Continuous monitoring of motor function symptoms
- Parkinson’s disease
- Other movement disorders such as Huntington’s disease, dystonia, etc.
Benefits and Advantages
- Computationally efficient
- Practical – stand-alone wearable device for real-time monitoring
- Energy-efficient
- Unobtrusive – doesn’t require multiple sensors and extensive feature engineering
- Achieves a 7.25% increase in precision and a 4.4% improvement in accuracy compared to supervised models
- Uses as little as 40% of the labeled training data required for supervised learning
- Saves considerable time and expertise
- The model activation module reduces inference time by up to 67%
- Reduced model execution complexity and processing time
For more information about this opportunity, please see
For more information about the inventor(s) and their research, please see