Kolossvary
A new approach using radiomics to take a fresh look at cardiovascular disease (CAD) risk revealed that there may be important differences in the factors that contribute to coronary plaque structural progression.
Clinicians have long recognized the impact of risk factors for CAD. Yet, as Marton Kolossvary, MD, postdoctoral researcher at Johns Hopkins University (JHU) School of Medicine, Baltimore, MD, noted, visual and volumetric analysis of plaque buildup only describe its magnitude.
"It's actually quite bothersome that, to date, we describe complex diseases with simple numbers, such as the amount of plaque you have," Dr. Kolossvary said during his RSNA 2020 presentation.
Dr. Kolossvary said that current markers for CAD are not sensitive enough to identify differences in disease phenotypes and might inaccurately suggest a similarity in the contribution of various atherosclerotic cardiovascular disease (ASCVD) risk factors to disease progression, all increasing the magnitude of the disease.
In their study, Dr. Kolossvary and colleagues took a new approach: radiomics, a precision phenotyping tool which extracts features from radiographic medical images using data-characterization algorithms. This approach provided researchers with multiple parameters to consider when describing the disease. "Radiomics play a huge role in radiology; essentially what genetics, proteomics and metabolomics did for basic sciences," Dr. Kolossvary said.
Role of Radiomics in Uncovering Disease
Researchers performed a prospective, longitudinal observational study of 300 asymptomatic patients to determine whether radiomics would uncover disease characteristics and reveal differences not shown through conventional imaging markers alone. Of the 300 patients, 210 were male (average age 49 years), 174 were cocaine users and 226 were HIV-infected.
Patients underwent coronary CT angiography (CCTA) twice, and the researchers performed precision phenotyping of CAD by calculating 1,276 radiomic features on 861 plaques found during baseline scans.
"We had a 'eureka' moment after we did our first analysis," Dr. Kolossvary said. After successfully identifying associations with radiomic features for ASCVD and cocaine use, researchers examined the results more closely and found no overlap between the two.
"The features affected by ASCVD were not affected by cocaine use and vice versa," Dr. Kolossvary said.
Additional Pathways of Disease
Following a hierarchical clustering of significant radiomic features, he reported finding 13 different structural components. Eight were unique to cocaine use, three were unique to ASCVD risk and two had parameters associated with cocaine use, elevated ASCVD risk and/or HIV infection. Gender and age stratification analyses reflected further differences.
The results demonstrate differences in the way cocaine use and HIV infection contribute to CAD compared to more conventional ASCVD risk factors. "We knew this was something new. Our results have many limitations, but seeing such clear results makes us certain that we are seeing an actual signal, and that we need to rethink how we perceive atherosclerosis," Dr. Kolossvary said.
The discovery of additional pathways of disease progression may open new targets of pharmaceutical intervention. "It may help us explain why we see such different manifestations of the disease in different people, even if they are siblings."
In the future, Dr. Kolossvary hopes to connect coronary plaque radiomics with other data such as genetics, proteomics or metabolomics to achieve better identification of actual pathways of disease progression.
Figure 1. Manhattan-plot of p values for associations between cocaine use, HIV-infection and elevated ASCVD risk and each radiomic parameter. a-c, p values for univariate associations between each radiomic feature and cocaine use, HIV-infection and elevated ASCVD risk in univariate models. d-f, p values for associations between each radiomic feature and cocaine use, HIV-infection and elevated ASCVD risk in multivariate models corrected for high-sensitivity C-reactive protein as the most common marker of inflammation, positive family history of CAD as an indicator of potential genetic predisposition for CAD progression, statin-use as it is known to modify the composition and development of coronary plaques, and the plaque volume itself, as we wished to correct for any potential intrinsic correlation between volume and morphology. Radiomic parameters are situated on the x-axis in the same order of each sub-plot, while the corresponding p values are located on the y-axis. Points above the red line (p=0.00004) indicate radiomic features in which case the given predictor showed a significant association. There was no overlap between radiomic features associated with cocaine use or elevated ASCVD risk potentially implying different pathological pathways of plaque progression. Abbreviations: ASCVD: atherosclerotic cardiovascular disease; CAD: coronary artery disease; HIV: human immunodeficiency virus.
Figure 2. Hierarchical cluster dendrogram, heatmap of inter-variable R2 values among significant radiomic features and p values for associations with cocaine use, HIV-infection and/or elevated ASCVD risk. a, Hierarchical cluster dendrogram of radiomic features significantly associated with cocaine use, HIV-infection and/or elevated ASCVD risk. Clusters are color coded depending on which risk factor the features were associated with. b, Heatmap of R2 values for linear regressions between each pair of the significant radiomic features (n=409). The elements of the heatmap are color coded depending on which risk factor the given parameters were associated with. The clusters are outlined in yellow. c, Corresponding p value for cocaine use, HIV-infection and increased ASCVD for each radiomic feature. The features are reordered based-on hierarchical clustering to correspond with the dendrogram. Bars extending further than the red line (p=0.00004) indicate statistically significant associations. Results from the hierarchical clustering indicate that there are distinct morphological feature sets which are only associated with specific risk factors. Furthermore, the p values for cocaine use among clusters associated with cocaine use were magnitudes smaller than for HIV-infection and especially for elevated ASCVD risk. Also, the p values for elevated ASCVD risk for the three clusters only containing radiomic features which were associated with elevated ASCVD risk were magnitudes smaller than for cocaine use of HIV-infection. These results potentially imply unique pathways of coronary atherosclerosis progression, as the modifying effects of cocaine use and conventional cardiovascular risk factors are clearly separable. Abbreviations: ASCVD: atherosclerotic cardiovascular disease; HIV: human immunodeficiency virus.
Figure 3. Corresponding p values for associations between cocaine use, HIV-infection, elevated ASCVD risk and the significant radiomic features stratified by sex and age. a, b, Corresponding p value for associations between cocaine use, HIV-infection and increased ASCVD for each radiomic feature stratified by sex. c, d, Corresponding p value for associations between the risk factors and each significant radiomic feature stratified by age based-on the median age of 50.7 years. The features are reordered based-on hierarchical clustering. Bars extending further than the red line (p=0.00004) indicate statistically significant associations. Our sex-based results indicate sex specific contributions of the different risk factors on coronary atherosclerosis morphology. Furthermore, our age stratification indicates that different risk factors may have different contributions to atherosclerosis depending on the individuals age. Abbreviations: ASCVD: atherosclerotic cardiovascular disease; HIV: human immunodeficiency virus.
For More Information:
View the RSNA 2020 session Different Risk Factors Result in Unique Coronary Plaque Morphologies - a Longitudinal Radiomic Analysis — SSCA07 at RSNA2020.RSNA.org.
