Freezing of Gait as a Digital Biomarker

Freezing of gait is a transient episode in which a person is unable to produce effective stepping despite intending to walk. Episodes typically last seconds, are triggered by turning, doorways, dual tasking or time pressure, and are a major cause of falls and loss of independence in Parkinson's disease.

It is the clearest example in this library of a symptom that periodic assessment cannot capture. Freezing is episodic and context dependent, and the clinic corridor is close to the least likely place for it to occur: the environment is open, the person is concentrating, and there is no time pressure. Clinicians therefore rely on the Freezing of Gait Questionnaire, which asks the person to recall how often and in what circumstances freezing occurs.

Continuous sensing addresses that gap directly by observing the person in the environments where freezing actually happens.

Detection uses inertial sensors, most often at the lower back with additional sensors at the ankles or shins. The classic signal processing approach exploits the fact that during a freezing episode the leg produces high frequency trembling in place rather than the normal stepping rhythm, so the ratio of energy in a high frequency band to energy in the walking band rises sharply. Later approaches use machine learning on multi sensor features.

Reference standards come from video annotated by trained raters, which is why most validation happens in a laboratory or in a scripted home protocol rather than in unrestricted daily life. Open source pipelines for measuring freezing during daily life now exist, and the definition of freezing itself has been revised by consensus, which matters because detection algorithms can only be as consistent as the definition they are trained against.

The main research use is quantifying how often freezing occurs and under what conditions, which supports trials of medication, deep brain stimulation, cueing devices and gait rehabilitation. Because episode counts vary enormously between days, continuous recording across a week gives a far more stable estimate than any single assessment.

The second use is context. Sensors can record what the person was doing immediately before an episode, which turns freezing from a count into a description of triggers that can inform therapy. Measures are reported alongside the Freezing of Gait Questionnaire, which is the direct clinical counterpart, and alongside balance and mobility assessments such as the Dynamic Gait Index and the Tinetti assessment that describe the wider walking impairment freezing sits within.

Detection performance in the home falls well short of what laboratory studies suggest. Episodes are short, irregular and easily confused with normal stopping, turning or standing still, and false positives accumulate quickly across a full day of recording.

Algorithms are typically trained on small cohorts with a specific sensor placement and do not transfer reliably to different placements or populations. Video reference annotation is itself subject to rater disagreement about when an episode starts and ends.

The practical consequence is that automated detection is usable for describing how often freezing occurred across a monitoring period, and is not yet dependable enough to function as a real time alarm or a safety critical trigger.

• Gilat M, et al. An updated definition of freezing of gait. Nat Rev Neurol. 2026. pubmed.ncbi.nlm.nih.gov
• Mancini M, et al. Measuring freezing of gait during daily-life: an open-source, wearable sensors approach. J Neuroeng Rehabil. 2021. pubmed.ncbi.nlm.nih.gov
• Bächlin M, et al. Wearable assistant for Parkinson's disease patients with the freezing of gait symptom. IEEE Trans Inf Technol Biomed. 2010. pubmed.ncbi.nlm.nih.gov
• Mancini M, et al. Digital measures of freezing of gait across the spectrum of normal, non-freezers, possible freezers and definite freezers. J Neurol. 2023. pubmed.ncbi.nlm.nih.gov