3 key points on machine learning in cardiac CT
Machine learning (ML) has made substantial advancements specific to medical imaging in recent history. A team of researchers examined ML algorithm-based cardiovascular computed tomography (CT) to assess its benefits—and the remaining limitations.
“The amalgamation of ML-based algorithms with clinical imaging holds the promise to automate redundant tasks and improve disease diagnoses and prognostication, as well as offer the potential to provide new insights into novel biomarkers associated with specific disease processes,” wrote corresponding author James K. Min, MD, with Weill Cornell Medical College and the NewYork-Presbyterian Hospital, and colleagues.
Below are three takeaways from a study published online April 28 in the Journal of Cardiovascular Computed Tomography.
Machine learning quantification of in epicardial fat is ‘achievable’
The amount of fat surrounding the heart is associated with increased cardiovascular risk. ML has shown promise in automating the approach and thereby reducing manual measurement time.
Authors pointed to multiple separate studies that demonstrated ML-based algorithms are proficient at the task. One study used features related to pixels and their surrounding area to extract coronary artery calcium scores (CAC), while an algorithm segmented the various fat types. The mean accuracy for epicardial and mediastinal fat was 98.4 percent.
“Accurate epicardial fat quantification is achievable and could represent a new quantitative parameter that can potentially be implemented in patient risk assessment, similar to coronary artery calcium score,” Min et al. wrote.
Algorithms can accurately ID plaque
The authors argued that, although coronary computed tomographic angiography (CCTA) has become a reliable imaging technique in patients who require assessment of coronary arteries, more detailed plaque characterization has emerged that is highly vulnerable to varying opinions among practitioners.
ML, however, has been used to optimize information extraction of CCTA data to generate algorithms that can achieve automated, accurate and objective plaque analysis, the authors wrote.
Imaging has to overcome the ‘black box’ pitfall
“While ML algorithms are capable of accurately predicting an outcome, computers are not able, or not programmed, to logically and comprehensively translate the complex and often abstract calculations leading to the prediction back to its user,” Min et al. wrote. “The use of these complex systems makes it difficult to explain the origin and logic behind the predictions that are made.”
This, authors argue, is the “black box” nature of ML algorithms. The problem makes it difficult for clinicians to understand the predictions and use them in clinical practice. As a result, many applications in cardiac imaging are focusing on supporting humans and not replacing them.
Although the more simplified approaches such as automated CAC score or left ventricular functional analysis can be checked by readers, the more complicated tasks left to machines will need to overcome this problem, the group wrote. But the future remains bright.
“The continued expansion of ML applications coupled with deeper appreciation of its capabilities, as well as limitations, will enable healthcare to make the leapfrog into an era of individualized and precise healthcare administration,” authors wrote. “It will also provide the ability to investigate the effect, and prognostic significance, of phenotypic features seen on non-invasively acquired imaging studies.”
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.