Automatic Tracking of Hyoid Bone Displacement and Rotation Relative to Cervical Vertebrae in Videofluoroscopic...

Area of Science:

• Biomedical Engineering
• Medical Imaging Analysis
• Deep Learning Applications

Background:

• Hyoid bone displacement and rotation are crucial for swallowing assessment in videofluoroscopic swallow studies (VFSS).
• Manual annotation for quantitative analysis of hyoid kinematics is time-consuming and labor-intensive.

Purpose of the Study:

• To develop an automated method for tracking hyoid bone displacement and rotation in VFSS using deep learning.
• To improve the efficiency and accuracy of hyoid kinematics analysis in swallowing studies.

Main Methods:

• A full high-resolution network was employed to detect anterior and posterior hyoid bone landmarks.
• Anterior-inferior corners of C2 and C4 vertebrae were detected to establish a stable coordinate system.
• The model was trained on 59,468 VFSS frames from 1488 swallowing samples.

Main Results:

• Achieved an average landmark localization error of 2.38 pixels and an average angle prediction error of 0.065 radians.
• Demonstrated accurate frame-by-frame tracking of hyoid bone center displacement with mean absolute errors of 2.22 and 2.78 pixels on the x and y axes, respectively.
• Validated the effectiveness and accuracy of the automated method.

Conclusions:

• The proposed deep learning method provides an effective and accurate automatic analysis of hyoid bone kinematics during VFSS.
• This automated approach can significantly contribute to the early diagnosis and management of swallowing disorders.