Author ORCID Identifier

https://orcid.org/0000-0002-7748-5013

Date of Award

Spring 6-8-2024

Document Type

Thesis (Ph.D.)

Department or Program

Engineering Sciences

First Advisor

Kofi Odame

Abstract

Chronic Obstructive Pulmonary Disease (COPD), characterized by chronic airway inflammation and airflow obstruction, is the third leading cause of death globally. Patients with COPD experience exacerbated symptoms like breathlessness and cough, significantly impacting their quality of life and leading to costly hospitalizations. Early detection of COPD exacerbations is crucial for mitigating these negative effects.

The most critical element for early detection of COPD exacerbations is daily monitoring of lung function, particularly forced expiratory volume in one second (FEV1), a key metric of lung function. By tracking declines in FEV1, COPD exacerbations can be predicted up to two weeks in advance, allowing for timely interventions and potentially reducing hospital admissions. However, the gold standard for remote lung function monitoring, at-home spirometry using a handheld spirometer, requires specialized hardware and a physically demanding maneuver, making it difficult and unreliable for daily monitoring.

This thesis introduces HeartFEV1, a novel system that addresses these challenges by utilizing readily available mobile electrocardiogram (ECG) signals acquired during quiet breathing for FEV1 estimation. To achieve this, HeartFEV1 utilizes an ensemble approach, combining two models: a machine learning model and a residual neural network. The machine learning model uses features extracted from ECG-derived respiratory signals to predict FEV1, whereas the residual neural network is a deep learning approach that directly estimates FEV1 from ECG signals. By averaging the predictions from both models, the HeartFEV1 system achieves accurate FEV1 prediction. Using a dataset of twenty-five patients with obstructive lung disease, the HeartFEV1 system demonstrates a strong correlation (r = 0.78, R^2 = 0.61) with hospital-grade spirometry for FEV1 estimation. Additionally, it reports a low mean absolute percentage error (MAPE) of 20.77% and minimal bias (0.01) in FEV1 predictions.

These findings highlight the feasibility of using mobile ECG signals collected during quiet breathing as a convenient method for daily lung function monitoring.

Download

Included in

Biomedical Commons

Share

COinS