Is Fidgety Philip's ground truth also ours? The creation and application of a machine learning algorithm

Area of Science:

• Behavioral science
• Clinical phenotyping
• Biomedical engineering

Background:

• Clinical phenotyping relies on behavioral observations, but interpretation varies with training.
• Attention Deficit Hyperactivity Disorder, Restless Legs syndrome, and medication-induced activation syndromes share hyperactive-behaviors (H-behaviors).
• A neutral pictogram-guided phenotyping language (PG-PL) was developed to describe sitting body-segment movements objectively.

Purpose of the Study:

• To develop an automated algorithm for analyzing sitting behaviors.
• To objectively quantify hyperactive-behaviors (H-behaviors) during sitting.
• To support differential diagnosis and treatment concepts for conditions with H-behaviors.

Main Methods:

• A pictogram-guided phenotyping language (PG-PL) was created for annotating sitting behaviors.
• Manual annotations of 12 sitting videos served as ground truth.
• An automated algorithm using OpenPose was developed and evaluated against the ground truth.

Main Results:

• High inter-observer agreement (>85%–97%) was achieved with the PG-PL.
• The automated algorithm demonstrated good performance for legs, arms, and feet (AUC >0.79).
• Head motion prediction by the algorithm was less accurate (AUC 0.65), requiring further refinement.

Conclusions:

• The developed automated analysis algorithm is the first of its kind for H-behaviors during sitting.
• This tool can facilitate discussions on distinct phenotypical characteristics of H-behaviors.
• It may aid in differential diagnostic considerations and the development of improved treatment strategies.