A Sparse Non-Negative Matrix Factorization Framework for Identifying Functional Units of Tongue Behavior From MRI

Area of Science:

• Biomechanics
• Medical Imaging
• Computational Biology

Background:

• Muscle coordination patterns in lingual behaviors are complex synergies.
• Functional units are cohesive muscle groups within the tongue responsible for specific movements.
• Identifying these units is crucial for understanding normal/pathological function and improving clinical interventions.

Purpose of the Study:

• To develop a novel framework for identifying functional units of the tongue.
• To leverage tagged-magnetic resonance imaging (MRI) data for detailed tongue motion analysis.
• To improve surgical planning, treatment, and rehabilitation for lingual disorders.

Main Methods:

• Proposed a matrix factorization and probabilistic graphical model framework.
• Utilized motion quantities extracted from tagged-MRI data.
• Employed spectral clustering on weighting maps to identify cohesive functional regions.

Main Results:

• Demonstrated method accuracy using 2D images and synthetic data.
• Identified subject-specific, data-driven functional units in 3D using in vivo tongue motion data.
• Revealed internal tongue motion patterns during protrusion and speech tasks.

Conclusions:

• The proposed framework effectively identifies tongue functional units from tagged-MRI data.
• This method offers unprecedented insight into tongue biomechanics during speech and other behaviors.
• Findings can significantly advance clinical applications in speech therapy, surgery, and rehabilitation.