Assessment of coronary artery stenoses by digital subtraction angiography: a pathoanatomic validation

Automated computer assessment of coronary stenoses from digital subtraction angiographic images comparing geometric and videodensitometric algorithms was performed. Digital subtraction angiograms were acquired on a 512 X 512 X 8 bit pixel matrix at 8 frames/second. Fifteen segments from nine human cadaver coronary arteries, with lesions ranging from 0% to 97%, were analyzed. Hand injections of radiopaque dye were made during the pulsatile infusion of saline solution at physiologic pressures and flows. Individual frames best demonstrating a lesion were digitally magnified and the stenosis was measured; the operator identified only the segment of interest. The artery was then injected with a rapidly hardening gel during the same rate of infusion as that used during image acquisition. Histologic sections were cut at 2 mm intervals after fixation and elastic stains applied. Photographs of the section comparable to the site determined from the angiogram were taken, and hand planimetry by a blinded investigator was performed. There was an excellent correlation between histopathology and videodensitometry (r = 0.93; p less than 0.0001). The two geometric algorithms studied also had very good correlations (r = 0.90 and 0.84) with pathology. Two experienced angiographers, despite excellent agreement with each other, had lower correlations with pathology than any of the three computer algorithms studied (r = 0.79 and 0.83, respectively), although this difference did not attain statistical significance. This in vitro model simulating in vivo conditions validates the use of automated videodensitometric and geometric computer algorithms to interpret coronary angiography and assess severity of stenosis.