Part 1 of 6 in Circle CVI's Structural Heart series. Every Scan. One Plan. One Platform.

The annulus measurement decides the case.

Valve choice follows from it. So does sizing, so does the degree of oversizing, so does whether the root tolerates that oversizing, and so does whether you spend the end of an otherwise clean case managing a paravalvular leak. By the time the patient is on the table, that measurement was made days earlier, in a post-processing application the operator may never open, by a reader whose work arrives as a number on a report.



For a procedure this dependent on millimeters, that is worth examining.

The measurement varies more than most operators assume

Not between good readers and bad ones. Between a good reader and the same good reader, on the same scan, on a different day.

In a study of 82 TAVR screening CTAs read twice by each of three readers, a single reader arrived at a different device-size category between their own repeated measurements in 22 to 26 percent of cases. Moving to a multi-reader strategy cut that to 5 to 10 percent. The study was single-center and the readers spanned a radiologist, a 3D-lab technologist and a medical student, but the direction is the point: annulus measurement carries more internal variance than a single number on a report suggests.

Software adds its own. Comparing two established planning packages across 100 patients, mean annulus area came out at 482 mm² on one and 464 mm² on the other, with the difference significant at p < 0.001. Perimeter and coronary heights diverged the same way. Had the fully automated output been accepted without correction, prosthesis size would have changed in 18 percent of patients. Manual correction of the automated contours brought that down to 4 percent.

Two readers, two packages, two answers. The operator inherits one of them.

What sizing error actually costs

The relationship between oversizing and paravalvular regurgitation is not theoretical. In 835 intermediate-risk patients receiving a balloon-expandable valve, degree of oversizing tracked inversely with the frequency and severity of paravalvular regurgitation, with area and perimeter oversizing performing equally as predictors at an AUC of 0.78.

Push oversizing too far in the wrong anatomy and the consequence changes character. The landmark analysis of aortic root rupture during balloon-expandable TAVR identified moderate-to-severe subannular calcification and prosthesis oversizing at or above 20 percent as the anatomic and procedural features associated with rupture.

At the other end, undersizing has its own bill. Across 10,298 TAVR patients at 21 international centers, 1.0 percent required urgent implantation of a supplementary valve, driven principally by residual aortic regurgitation and valve malposition. Device success in those patients was 70.4 percent against 92.2 percent, and 30-day mortality was 11.8 percent against 5.5 percent.

One percent is a small number until it is your case.

Why CT carries the measurement

This is settled, and has been for some time. The SCCT expert consensus on CT imaging for TAVR remains the reference document for acquisition, annulus and LVOT measurement, fluoroscopic angulation reporting, and vascular access assessment. The European Society of Cardiovascular Radiology consensus standardizes the same ground and proposes a reporting template for it. The EACVI Task Force consensus on interventional cardiovascular imaging assigns annulus and aortic root measurement to CT alongside 3D transesophageal echo, and requires that pre-procedural assessment establish annular size and shape along with the extent and distribution of calcification.

The outcome evidence supports the practice. A meta-analysis of six studies comparing CT-based against echo-based annulus sizing found moderate-or-greater paravalvular regurgitation substantially less common with CT-based sizing than with 2D TEE, at an odds ratio of 0.31, with no difference in annular rupture or 30-day mortality between modalities. CT reduces leak. It has not been shown to reduce death, and the honest version of the claim stops there.

What automation should do, and what it should not

The published evidence on automated annulus and landmark detection is real and still maturing. The largest study to date trained on 1,252 pre-TAVR CT scans and validated externally across 19 hospitals, reporting intraclass correlation up to 0.998 for measurement and accuracy up to 0.989 for detection of anatomical risk factors. It is retrospective, and the external cohort was 100 patients. A 2024 review of AI in TAVR imaging is blunt that the field is constrained by small cohorts, limited scope and variable performance, and that rigorous validation remains outstanding.

Which is the correct way to think about it. The value of assisted detection is not that it removes the operator's judgment from the measurement. It is that it removes the part of the work where fatigue and repetition introduce variance, and leaves the judgment where it belongs.

In cvi42, assisted annulus and landmark detection in the Interventional Planning Suite covering aortic, mitral and LAAC planning removes 80 to 100 clicks from a typical case by Circle's own measurement. Every contour it proposes is physician-editable. The analysis runs on-premises, so patient data stays inside the institution. And the 18-percent-to-4-percent finding above is the argument for exactly this design: automation that proposes and a clinician who disposes produces better agreement than either alone.

The volume argument is no longer hypothetical

EARLY TAVR randomized 901 patients with asymptomatic severe aortic stenosis and preserved ejection fraction to TAVR or clinical surveillance across 75 sites. The primary composite of death, stroke or unplanned cardiovascular hospitalization occurred in 26.8 percent of the TAVR arm against 45.3 percent under surveillance, a hazard ratio of 0.50. Read it precisely: the composite was driven by unplanned hospitalization, at 20.9 percent against 41.7 percent. Death and stroke individually did not differ significantly. What the trial establishes is not a mortality case for early intervention but that 87 percent of the surveillance arm crossed over to valve replacement anyway.

Guidelines have started to move. The 2025 ESC/EACTS valvular heart disease guidelines lowered the age at which TAVI carries a Class I, Level A recommendation to 70 in patients with suitable anatomy for transfemoral access, and now state that intervention should be considered in asymptomatic patients with preserved ejection fraction at Class IIa, Level A. In the United States, CMS finalized a revised TAVR national coverage determination on 10 September 2026, extending coverage to asymptomatic severe aortic stenosis under coverage with evidence development and removing the CED requirement for symptomatic patients entirely.

Referral volume grows from here. The constraint moves from procedure capacity to evaluation capacity, and evaluation capacity is a CT and a heart-team decision.

The scans that come after

The index procedure is not the last time that anatomy needs reading.

The 2025 multi-society consensus on CT for prosthetic heart valve assessment is careful about this: cardiac CT is not used for routine screening of valve degeneration. It is indication-driven, and the indications are real. Suspected leaflet thrombosis is one, and hypoattenuated leaflet thickening is common enough to matter, rising from 10 percent at 30 days to 24 percent at one year in the PARTNER 3 CT substudy. Planning a redo procedure is another, where CT establishes coronary occlusion risk and the root anatomy that determines sizing.

That last one deserves attention at the index case, not after it. CT simulation in patients modeled for TAVR-in-TAVR found coronary access technically impossible in 27 percent after a supra-annular index valve against 10 percent after an intra-annular one, with unimpeded access in 8 percent against 33 percent. The study is small and device-generation-specific, but the principle holds: the valve you choose today shapes what is possible in five years.

cvi42 reads all of it in one environment. The planning CT, the diagnostic cardiac CT, the CMR, and the follow-up study when a clinical question warrants one, against the same prior data, with no export and no second application.

The plan is the product

The procedure is a few hours. The plan is what determines how those hours go, and it is built somewhere else, by someone else, in software the operator usually does not touch.

Making that plan reproducible is not a workflow improvement. It is the difference between a measurement you accept and a measurement you would defend.

See how a structural heart case is planned in cvi42 →