New machine learning score may better predict FFR-defined ischemia than visual stenosis

The tool predicted ischemia as well as FFR-CT, with information similar to invasive FFR and impaired blood flow on PET.

A score derived from plaque characteristics on coronary angiography (CCTA) using machine learning can predict ischemia as well as CT-derived estimates of fractional flow reserve (FFR-CT), with the advantage of being noninvasive and calculated in real time, investigators say, without the need for an adenosine infusion.

According to the researchers, the score outperforms the standard CCTA assessment in predicting future risk while matching information about ischemia and altered myocardial blood flow (MBF) provided by invasive FFR and positron emission tomography (PET) scans. , respectively. As such, they say, the score could be integrated into clinical workflow to streamline the identification of high-risk patients before they undergo invasive coronary angiography.

“Our results are very positive,” lead author Damini Dey, PhD (Cedars-Sinai Medical Center, Los Angeles, CA), told TCTMD. “This externally validated score could very well predict the functional significance of lesions.”

The tool, as described in an article published this month in Circulation: Cardiovascular imaging, is currently used only for research purposes at their institution. But after regulatory approval, Dey said, “it could be applied on-site immediately after plaque analysis.”

This offers advantages over the currently available FFR-CT, which involves sending data off-site for analysis involving turnaround times that limit its clinical use.

She sees their score eventually being used as “a clinical decision support tool” by referring doctors before invasive coronary angiography to determine whether the patient has a high probability of impaired FFR or a high risk of ischemia. “At the same time, it may also be a postponement of downstream testing in people at low risk of ischemia,” she said. “It would really help the efficiency and accuracy of referral for invasive coronary angiography.”

External validation

To “train” the score, lead author Andrew Lin, MBBS, PhD (Cedars-Sinai Medical Center, Los Angeles, CA), Dey and colleagues used data from 254 patients who, in the trial NXT, underwent invasive FFR in 484 vessels. The machine learning score was then tested in 601 vessels from 208 suspected CAD patients who had undergone CCTA, H2O Invasive PET and FFR imaging in the PACIFIC Study. Five patients and 23 vessels were excluded from the population tested due to non-evaluable CCTA analyses.

In the test cohort, 23.9% of vessels had ischemia defined by FFR and 33.6% had impaired hyperemic MBF on PET. Total plaque volumes and loads, as well as individual plaque components, were higher in vessels with FFR-defined ischemia compared to vessels without (all P P P = 0.34). Moreover, it had an accuracy, sensitivity and specificity of 84%, 87% and 82%, respectively, for the discrimination of ischemia. Additionally, adding the score to the grade of visual stenosis resulted in a substantial net improvement in reclassification for ischemia defined by FFR (1.16; P

The two most important characteristics for predicting FFR-defined ischemia with the machine learning score were quantitative stenosis as a percentage of diameter and low-density non-calcified plaque volume. Characteristics of the calcified plaque contributed the least to the prediction of ischemia.

Predicting impaired MBF only, the score performed significantly better than visual stenosis grade (AUC 0.80 vs 0.74; P = 0.02) and was similar to FFR CT (AUC 0.77; P = 0.16). The accuracy, sensitivity, and specificity of the score for predicting impaired MBF were 77%, 73%, and 80%, respectively.

“Functional tools that complement CCTA without requiring additional image acquisition or drug administration have been developed to improve the detection of lesion-specific ischemia,” the authors write, citing computational fluid dynamics as a tool to be applied with currently available software. “However, this process is time-consuming and computationally expensive, requiring off-site processing by a main lab.” On-site FFR CT, on the other hand, is “highly dependent on physician interaction,” they note.

“Unlike all of these techniques, our proposed machine learning score relies solely on anatomical information from the CCTA without the addition of physiological parameters, uses semi-automated plaque analysis, which is highly reproducible by trained clinicians or technicians. , and can be computed in

Over time, new plate features and CCTA measurements can be added to the algorithm to improve the score and its predictive ability, Lin and colleagues add. “In this era of personalized medicine, there is a growing demand for more accurate, patient-tailored risk prediction,” they conclude. A score like this “allows the interrogation of individualized predictive models by clinicians or researchers unfamiliar with machine learning techniques.”

Assuming the tool withstands further validation studies and eventually gets the required regulatory approval, Dey said she can see the score used on all patients who undergo CCTA. “We also tested it in patients with first degree coronary stenosis, so 1-25%, and the risk of ischemia is generally low,” she said.

Dey said she felt confident in the score given the external validation they had done, but acknowledged that “any additional validation in other cohorts is always welcome before entering clinical practice” .

Comments are closed.