Determining functional units of tongue motion via graph-regularized sparse non-negative matrix factorization.

Area of Science:

• Biomechanics
• Medical Imaging
• Computational Biology

Background:

• Tongue motion during speech and swallowing is complex, involving coordinated movements of distinct regions.
• Understanding the tongue's functional organization is crucial for speech and swallowing research.
• Previous methods for analyzing tongue motion lacked detailed functional parcellation.

Purpose of the Study:

• To develop and validate a novel computational method for identifying functional units of tongue motion.
• To reveal the intrinsic functional organization of the tongue during speech and swallowing tasks.
• To demonstrate subject- and task-specific functional parcellation of the tongue.

Main Methods:

• Utilized tagged magnetic resonance imaging (tMRI) to track tissue displacements.
• Developed a graph-regularized sparse non-negative matrix factorization (NMF) framework to identify latent motion building blocks.
• Employed spectral clustering on the NMF weighting map to define functional units based on motion coherence.

Main Results:

• The proposed algorithm accurately clustered different image types in 2D verification.
• Analysis of 3D tMRI data from five subjects revealed subject- and task-specific functional parcellations of the tongue.
• Identified localized regions representing functional units defined by distinct motion patterns.

Conclusions:

• The novel matrix factorization and clustering method effectively defines functional units of tongue motion.
• This approach provides a subject- and task-specific functional parcellation of the tongue.
• The findings contribute to a deeper understanding of tongue biomechanics in speech and swallowing.