Swallowing Assessment using High-Resolution Cervical Auscultations and Transformer-based Neural Networks.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Swallowing abnormalities require accurate assessment using methods like videofluoroscopic swallow studies (VFSS) and high-resolution cervical auscultation (HRCA).
• Deep learning models, primarily convolutional and recurrent neural networks, are increasingly used for analyzing swallowing kinematics from VFSS and HRCA data.
• Transformers, a novel deep learning architecture, have shown promise in various sequence-to-sequence tasks but their application in swallowing analysis is underexplored.

Purpose of the Study:

• To investigate the efficacy of transformer models in analyzing swallowing kinematics from HRCA signals.
• To specifically assess the transformer model's ability to detect upper esophageal sphincter opening and laryngeal vestibule closure.
• To evaluate the generalizability of the proposed transformer network on an independent dataset.

Main Methods:

• Utilized high-resolution cervical auscultation (HRCA) signals as input data.
• Developed and implemented a transformer-based deep learning network for kinematic analysis.
• Tested the model's performance on an independent dataset to assess generalizability.

Main Results:

• The transformer network achieved average detection accuracies exceeding 90% for upper esophageal sphincter opening and 85% for laryngeal vestibule closure.
• The model's performance surpassed that of previously reported hybrid neural networks in the literature.
• High-performance measures on the independent test dataset demonstrated the model's strong generalizability to unseen data.

Conclusions:

• Transformer models are highly effective for analyzing swallowing kinematics using HRCA signals.
• The proposed transformer network demonstrates superior performance and generalizability compared to existing deep learning approaches for swallowing assessment.
• This study highlights the potential of transformers in advancing noninvasive swallowing disorder diagnostics.