Use of posterior probabilities from a long short-term memory network for characterizing dance behavior with multiple...

Area of Science:

• Neuroscience
• Kinesiology
• Biomedical Engineering

Background:

• Dancing shows potential benefits for cognitive health in individuals with mild cognitive impairment, Alzheimer's disease, or dementia.
• Current methods for assessing motor behavior quality in dance studies are insufficient.
• Novel metrics are needed to accurately characterize motor behavior during dance interventions.

Purpose of the Study:

• To determine the optimal observation duration for motor behaviors to enhance the reliability of "dance-like state" (DLS) scores.
• To establish DLS scores as a reliable metric for characterizing motor behavior quality in free-form dancing using accelerometry.

Main Methods:

• Forty-one adults wore five triaxial accelerometers (wrists, ankles, waist) during various activities, including free-form dancing.
• Accelerometer data predicted observed behavior (dancing/not dancing) using a long short-term memory (LSTM) network.
• The Spearman-Brown Prophecy formula assessed the number of 1-minute observational periods needed for reliable DLS scores (r ≥ 0.80).

Main Results:

• The LSTM network achieved high classification accuracy (89.1% to 94.0%) in recognizing dance behavior using data from all five accelerometers.
• DLS scores demonstrated high reliability (r > 0.80) when averaged over periods of two minutes or more.
• DLS scores correlated significantly with factors including age, dance training, physical characteristics, music parameters, gait speed, and energy expenditure.

Conclusions:

• Dance-like state (DLS) scores derived from accelerometry effectively characterize motor behavior quality.
• A minimum of two minutes of data is sufficient for reliable DLS scoring.
• Further research is warranted to explore the relationship between motor behavior quality and cognitive health.