UT Austin Department of Biomedical Engineering associate professor Ed Castillo, along with collaborators from Michigan-based Corewell Health, are spearheading image analysis advancements with machine learning.  

The goal is to develop imaging-based predictive modeling to improve clinical decision making and assessments of therapeutic efficacy for chronic obstructive pulmonary disease (COPD).  

Castillo, Dr. Girish Nair, M.D. and Lili Zhao received a four-year, $1.38 Million grant from the National Institutes of Health (NIH) to showcase novel manners in which CT scans can help physicians identify patients at risk of disease progression and improve COPD survival rates. 

Why the research matters: 

COPD is the third-leading cause of death in the United States. It refers to a group of diseases that cause airflow blockage and breathing-related problems, including emphysema and chronic bronchitis.  

According to the CDC, approximately 16 million Americans live with COPD. Early intervention is crucial for slowing COPD progression and improving a patient’s quality of life.  

Current methods using quantitative computer tomography, also known as a CT scan, are based on analyzing variations in Hounsfield Units (HU). These are measurements used by radiologists to interpret images from the scan. While the units are fairly accurate in displaying the progression of COPD, their values are known to be dependent upon the breathing of the patient.

Consequently, these methods struggle with reproducibility and rely on normalization techniques to account for a patient breathing differently each time a scan is completed.  

What the goal of the research is: 

To address this, a robust class of CT-derived ventilation (CTV) methods has been developed, which accurately characterizes breathing-induced volume changes. These methods demonstrated a high correlation with ground truth lung function imaging in initial studies.

Recently, this framework has been extended to calculate changes in the amount of blood mass circulating within a patient’s breathing lungs as a surrogate for pulmonary perfusion—which is needed for the oxygen/carbon dioxide gas exchange that occurs during breathing. The end result is an imaging-based CT-Perfusion (CTP) estimate. Together, CTV and CTP create CT-derived functional imaging (CTFI). This is the first methodology to mathematically describe changes in respiratory HU values in terms of airflow and blood flow.  

This allows for the computation of VQ (ventilation/perfusion) ratio imaging that is inherently normalized to how a patient breathes. As a result, CTFI has the potential to detect and quantify early signs associated with the severity of COPD on CT images. 

Castillo, Dr. Nair, and Zhao aim to prove that a CTFI-based machine learning model is more accurate in determining COPD disease progression and survival rates than traditional methods. 

What happens next: 

The researchers are working to create an automated computational imaging system that can analyze pairs of CT images taken when someone breathes in and out. They'll use machine learning and other advanced computational techniques to convert images into practical information that will help physicians understand the severity of COPD in a patient and predict survival rates.

Unlike current methods, their system should work well no matter how the patient is breathing, consequently aiding physicians in making clinical decisions and evaluating treatment efficacy.