NCT05098808

Artificial Intelligence in Diagnosing Dysphagia Patients

Classification of Dysphagia Patients at Risk of Aspiration Pneumonia Using Machine Learning Algorithms Incorporating Acoustic Features From Phonetic Evaluation

• Status: Completed
• Phase: —
• Study type: Observational
• Enrollment: 449 (actual)
• Sponsor: The Catholic University of Korea · Academic / Other
• Sex: All
• Age: 19 Years – 90 Years
• Healthy volunteers: —

Summary

In this prospective study we extracted acoustic parameters using PRAAT from patient's attempt to phonate during the clinical evaluation using a digital smart device. From these parameters we attempted (1) to define which of the PRAAT acoustic features best help to discriminate patients with dysphagia (2) to develop algorithms using sophisticated ML techniques that best classify those i) with dysphagia and those ii ) at high risk of respiratory complications due to poor cough force.

Detailed description

This study was prospective study, and patients who visited the department of rehabilitation medicine in a single university-affiliated tertiary hospital with dysphagic symptoms from September 2019 to March 2021 were included.Voice recording was performed at the enrollment with blinded assessment, where the participants first visited the rehabilitation department with chief complaints of dysphagia. The cough sounds were recorded with an iPad (Apple, Cupertino, CA, USA) through an embedded microphone. From the acoustic files we extracted fourteen voice parameters that include the average value and standard deviation of the fundamental frequency (f0), harmonic-to-noise ratio (HNR), the jitter that refers to frequency instability, and the shimmer that represents the amplitude instability of the sound signal. Machine learning algorithms and sophisticated deep neural network analysis will be performed.

Conditions

• Respiration Disorders
• Swallowing Disorder
• Phonation Disorder
• Stroke
• Aspiration Pneumonia
• Aspiration; Liquids

Interventions

Type	Name	Description
OTHER	Acoustic features (from signals obtained during phonation)	Acoustic features will be obtained via phonation files. A voice recorder application provided by Apple was used, and the sampling frequency of the sound was 44,100 Hz. The digitized cough sound signals were band-pass-filtered between 20 to 16,000 Hz to use data from the whole frequency band gathered by the iPad. In each case, the smart device was positioned 20cm from the patient

Timeline

• Start date: 2019-09-01
• Primary completion: 2021-09-01
• Completion: 2021-10-01
• First posted: 2021-10-28
• Last updated: 2021-10-28

Locations

1 site across 1 country: South Korea