Page 69 - GTM-3-1

Global Translational Medicine Evaluating ML models for CAD prediction

and water several hours before the scan, and contrast Previous literature regarding ML and CAD has
dye is administered to enhance the visibility of coronary shown promising results, especially when compared to
vasculature. Administration of angiographic dye has traditional risk assessment tools. One study developed
been proven to worsen renal function in patients with an ML Risk Calculator using the same factors as the
underlying kidney disease, which can lead to oliguria and 2013 ACC/AHA Pooled Cohort Equations Risk Calculator
a need for hemodialysis. Other complications associated which outperformed the latter by recommending less
4
with CT angiography include arterial dissection, statin therapy and missing fewer CVD events. ML models
17
arrhythmia, stroke, and death, on top of the generalized developed for both clinical and imaging parameters such as
risk of radiation exposure. Researchers have started to a coronary artery calcium score also increased the predictive
5,6
rely on risk assessment models to overcome the various power of obstructive CAD. 18,19 Quite impressively, one
inconveniences and complications associated with current study utilizing a databank of over 400,000 participants
diagnostic modalities. and 450 discrete variables retrospectively discovered
The traditional method of CVD risk assessment new predictors of CVD in the diabetic population where
predicts the likelihood of CVD events over a 10-year traditional risk calculation tools have been unreliable. 20
period or a lifetime, and it relies on nine modifiable and ML has been used to identify the most important
non-modifiable risk factors such as age, sex, race, blood features to arrive at a diagnosis of CAD. Studies have
pressure, cholesterol levels, and smoking behavior to agreed that age, male sex, smoking, and number of
generate a quantitative estimation. Other regression-based calcified segments as most useful within their respective
7
tools (Framingham risk score, GRACE score, TIMI score, models. 20,21 Using these tools, ML can provide a different
etc.) utilize similar readily retrievable population samples perspective – it can help identify evidence of CAD in
to stratify risk and guide the course and intensity of patients without a formal clinical diagnosis. Its utility
22
therapies. While the new 2013 ACC/AHA Pooled Cohort extends to informing pertinent clinical management,
Equations Risk Calculator has made major advancements wherein it facilitates the precise categorization of patients
in providing specific estimates for atherosclerotic CVD necessitating blood pressure-lowering or lipid-lowering
(ASCVD), such tools are inherently limited by their interventions, as well as enhances screening efficacy
23
7
reliance on conventional statistical methods. These through the incorporation of treadmill exercise test
methods rely on a small subset of risk factors to generalize characteristics. An inventive non-invasive technique
24
predictions for much larger and more diverse populations utilizing iris imaging has shown promise for the early
and require manual recalibration with every additional detection of CAD. This approach integrates iridology
25
data set. This inevitably leads to both over- and under- with digital image processing to analyze features of the iris
estimation of CVD events in certain demographics. 8 corresponding to heart health. Involving 198 volunteers,
The rapidly emerging field of machine learning (ML) researchers successfully distinguished individuals with
in healthcare has created a new avenue to overcome the CAD from those without, using algorithms to process iris
limitations of current clinical diagnostic and prediction patterns. Their findings suggest that analyzing iris images
models. ML uses computer algorithms to process large with a support vector machine classifier can predict CAD
9,10
amounts of data to identify patterns not only between the with a 93% accuracy rate, presenting a potential alternative
variables and possible outcomes but also relationships to conventional diagnostic methods and paving the way
between the different variables themselves. 11,12 IBM’s for its application in telemedicine. In addition, recent
Watson has recently been receiving a significant amount advances in cardiovascular CT technology, such as multi-
of media attention for its focus on precision medicine slice imaging and photon-counting CT, have significantly
regarding cancer diagnosis and treatment utilizing improved image resolution and diagnostic capabilities in
combinations of ML and natural language processing CVD. Techniques like CT-derived fractional flow reserve
capabilities. ML also has a wide range of applications in offer better detection of myocardial ischemia than standard
13
the advancement of clinical trial research. Through the use CT scans. In addition, 3D-printed models and visualization
of predictive analytics, researchers can determine optimal tools like virtual reality are enhancing surgical planning
sample sizes to optimize efficacy and reduce data errors, and patient communication. The integration of AI is
along with evaluating broader ranges of data. In regards further boosting the diagnostic accuracy of cardiovascular
14
to CAD, ML could account for the multifactorial nature CT, making it a powerful tool for both diagnosing and
26,27
of the pathology by analyzing a variety of populations and predicting cardiovascular conditions.
novel risk factors, ultimately improving risk calculations at Early detection of CAD is crucial to limit the progression
the level of the individual patient. 12,15,16 and severity of this pathology, but image-based detection

Volume 3 Issue 1 (2024) 2 https://doi.org/10.36922/gtm.2669