Mao, Shitong
(2022)
A Study of Deep Neural Networks Associated Computational Deglutition for Swallowing Assessment.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Swallowing and swallowing disorders have drawn continued interest over the past several decades. Detecting and identifying the kinematic events during swallowing offer references for assessing swallowing functions. The transitional method employs the X-ray imaging technique, which provides a direct and dynamic view of anatomic structure activities. Such a method has many practical limitations, and an alternative assessing tool can fundamentally change the way of diagnosing swallowing disorders.
The neck sensors, which acquire high-resolution cervical auscultation (HRCA) signals with a triaxial accelerometer and microphone, allow us to investigate the activation patterns during the swallowing tasks. This non-invasive method offers signal features responsive to sequential events and suggests the feasibility of swallowing screening by using modern computational algorithms. However, the critical barrier is that the signals cannot explicitly describe the clinical events and identify the abnormal swallows because the background knowledge and the mathematical explanation are insufficient. Moreover, as physiological recordings, the sensor signals are commonly generated from a highly non-linear system, which is also stochastic and patient-specific.
AI or machine learning methods are permeating all aspects of physiological applications. In computational deglutition, they are also powerful and promising tools for addressing the above-mentioned problem. The model training process explores the underlying association between the swallowing events and the sensor-measured signal when sufficient observed samples are given. We applied deep learning techniques, such as recurrent neural networks and convolutional neural networks, to the computational deglutition in three key areas. First, using deep learning techniques, we plan to characterize the properties of hyoid bone movement. Second, we investigated the feasibility of predicting the laryngeal vestibule closure by the HRCA signals. Finally, we explored the use of HRCA signals to classify the patients and healthy participants. Based on our studies, the advanced computational and data-driven algorithms will enable patients to get continuous feedback about their swallowing outside of standard clinical care settings, significantly facilitating their daily activities and enhancing the quality of their lives.
Share
Citation/Export: |
|
Social Networking: |
|
Details
Item Type: |
University of Pittsburgh ETD
|
Status: |
Unpublished |
Creators/Authors: |
|
ETD Committee: |
|
Date: |
6 September 2022 |
Date Type: |
Publication |
Defense Date: |
18 July 2022 |
Approval Date: |
6 September 2022 |
Submission Date: |
12 July 2022 |
Access Restriction: |
2 year -- Restrict access to University of Pittsburgh for a period of 2 years. |
Number of Pages: |
157 |
Institution: |
University of Pittsburgh |
Schools and Programs: |
Swanson School of Engineering > Electrical and Computer Engineering |
Degree: |
PhD - Doctor of Philosophy |
Thesis Type: |
Doctoral Dissertation |
Refereed: |
Yes |
Uncontrolled Keywords: |
Swallowing, Dysphagia, Cervical Auscultation, Deep Learning, Signal Processing, Neural Networks |
Date Deposited: |
06 Sep 2022 16:28 |
Last Modified: |
06 Sep 2022 16:28 |
URI: |
http://d-scholarship.pitt.edu/id/eprint/43269 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |