Machine learning tops common angiography methods in spotting heart blockages

Recent research found a machine learning algorithm outperformed coronary CT angiography and quantitative coronary angiography in identifying heart blockages.

In the study, published April 10 in Radiology, researchers compared fractional flow reserve (FFR) derived from coronary CT angiography based on computational fluid dynamics (FFRCFD) and FFR-derived from coronary CT angiography based on machine learning (FFRML) against coronary CT angiography and quantitative coronary angiography (QCA).

The retrospective study included 85 individuals. Each had undergone coronary CT angiography followed by invasive FFR. 

Results were as follows:

  • On a per-lesion and per-patient level, FFRML showed sensitivity of 79 percent and 90 percent, along with a specificity of 94 percent and 95 percent respectively for detecting lesion-specific ischemia.
  • FFRCFD achieved 79 and 89 percent sensitivity, and a specificity of 93 percent per-lesion and patient.
  • On a per-lesion level, FFRML and FFRCFD achieved an area under the receiver operating characteristic curve (AUC) of 0.89 compared to that of coronary CT angiography (0.61) and QCA (.69).
  • On a per-patient level, FFRML and FFRCFD performed better (AUC .91) than coronary CT angiography (0.65) and QCA (0.68).
  • The processing time for FFRML was shorter than FFRCFD (40.5 minutes compared to 43.4 minutes).

“Traditionally, FFR derived from coronary CT angiography has been computationally demanding and typically implemented off-site by using supercomputers in core laboratories,” wrote U. Joseph Schoepf, MD, with the department of medical imaging technologies at Siemens Healthcare, and colleagues.

 “Our study demonstrates that the FFRML algorithm performs equally in detecting lesion specific ischemia compared with the FFRCFD algorithm,” wrote Schoepf et al. “Both algorithms outperform accuracy of coronary CT angiography and QCA in the detection of flow-limiting stenosis.”