Presentation Title

Determining Lung Obstruction Using Forced Ocillation Technique and Machine Learning

Faculty Mentor

April Si

Start Date

23-11-2019 8:00 AM

End Date

23-11-2019 8:45 AM

Location

141

Session

poster 1

Type of Presentation

Poster

Subject Area

engineering_computer_science

Abstract

Background: A key factor in devising the treatment plan for patients with obstructive airways disease is being able to measure their lung functions accurately. Impulse oscillometry (IOS) is a new technique that use impulse acoustics to measure the levels of obstruction in the human respiratory tract. It does not require patient cooperation, which makes it well suited for children, seniors, or patients with respiratory distress. In this project, we will study the sensitivity of the IOS in detecting airway obstructions in an anatomically realistic lung cast model.

Methods: A lung model with 11 generations of bifurcation and 1240 outlets was developed following Hiroko Kitaoko and was manufactured using 3-D printing techniques. This hollow lung cast was put in 5-later container that is equivalent to the human total lung capacity (TLC) of an adult. Impulse sound singles of 5-20 Hz generated by the IOS were released into the lung cast, while tidal breathing was simulated using a gas pump. The reflected signals were collected by IOS and translate into airway-obstruction-related parameters such as resistance and resonance frequency. To study the airway obstruction effects, the outlets of the lung cast model was covered using playdoh by 0%, 10% and 20%. For each blockage, 400 tests were performed in order to develop a database. Machine learning was then used to learn the common features of the parameters from different blockages. New test samples were used to test the prediction accuracy of the machine-learned model.

Results: Statistically significant differences were found in the measurements with different airway blockages, indicating that the IOS is sensitive enough to detect the airway obstructions with a difference of 5% blockage. The prediction accuracy of the classifier (i.e., the machine-learned model) is around 70%.

Discussion and future work: IOS can detect blockage in the 3-D printed lung model. However, future work is needed to determine how it works in human patients. It is still unclear, but is of high interest to doctors, whether the IOS can detect the locations of the airway obstructions. Our future studies will address this problem by blocking different sections of the lung case model and testing it with IOS.

This document is currently not available here.

Share

COinS
 
Nov 23rd, 8:00 AM Nov 23rd, 8:45 AM

Determining Lung Obstruction Using Forced Ocillation Technique and Machine Learning

141

Background: A key factor in devising the treatment plan for patients with obstructive airways disease is being able to measure their lung functions accurately. Impulse oscillometry (IOS) is a new technique that use impulse acoustics to measure the levels of obstruction in the human respiratory tract. It does not require patient cooperation, which makes it well suited for children, seniors, or patients with respiratory distress. In this project, we will study the sensitivity of the IOS in detecting airway obstructions in an anatomically realistic lung cast model.

Methods: A lung model with 11 generations of bifurcation and 1240 outlets was developed following Hiroko Kitaoko and was manufactured using 3-D printing techniques. This hollow lung cast was put in 5-later container that is equivalent to the human total lung capacity (TLC) of an adult. Impulse sound singles of 5-20 Hz generated by the IOS were released into the lung cast, while tidal breathing was simulated using a gas pump. The reflected signals were collected by IOS and translate into airway-obstruction-related parameters such as resistance and resonance frequency. To study the airway obstruction effects, the outlets of the lung cast model was covered using playdoh by 0%, 10% and 20%. For each blockage, 400 tests were performed in order to develop a database. Machine learning was then used to learn the common features of the parameters from different blockages. New test samples were used to test the prediction accuracy of the machine-learned model.

Results: Statistically significant differences were found in the measurements with different airway blockages, indicating that the IOS is sensitive enough to detect the airway obstructions with a difference of 5% blockage. The prediction accuracy of the classifier (i.e., the machine-learned model) is around 70%.

Discussion and future work: IOS can detect blockage in the 3-D printed lung model. However, future work is needed to determine how it works in human patients. It is still unclear, but is of high interest to doctors, whether the IOS can detect the locations of the airway obstructions. Our future studies will address this problem by blocking different sections of the lung case model and testing it with IOS.