Health & Medicine · The Record
FDA cleared Heartvue's cardiac MRI software after a 167-day review
The company's founder called it a first-submission success with a change control plan included. FDA's file records the submission on March 12, 2026 and the decision on August 26, 2026 - 25 days longer than the 142-day median the advising firm publishes in its own year-in-review dataset.
The claim attached to this clearance is about the path through the agency as much as the product at the end of it. Heartvue.ai's founder described getting the company's cardiac MRI software cleared on its first submission, with a change control plan authorized alongside it. FDA's public file on that submission contradicts neither half. It does carry two dates that the coverage did not print: the agency recorded the submission on March 12, 2026, and the decision on August 26, 2026.
The AI Journal published the founder's statement under the name and title "Jeffrey Dendy, MD, CEO and Founder of Heartvue.ai". It opens on the clinical case: "As a practicing cardiologist, I built Heartvue.Proton to give clinicians back the time they spend on manual measurements and to make quantitative cardiac MRI available to every patient who needs it." The rest of the passage is about regulatory process - a first submission, a plan included, and credit to the Innolitics partners the statement says guided the company. Medical Device Network carried an overlapping fragment, "Clearing FDA on our first submission, with a PCCP included, is a testament to our team", attributed there to "Jeffrey Dendy, Heartvue founder and CEO". Two outlets, one voice, and it is the company's: none of the captured coverage carries a statement from Innolitics itself about this clearance.
What the software is said to do is described consistently across the trade press. AuntMinnie calls it "a machine learning-powered software platform for viewing, postprocessing, and quantitative evaluation of cardiovascular MR images", and reports that results stay under physician review. Diagnostic and Interventional Cardiology reports that the product automates measurements in three domains - 2D linear dimensions, ventricular volumes and ejection fraction, and blood flow quantification - and that manual analysis of a single study "can take up to an hour or more". That last figure describes current practice as the coverage presents it. It is not something the agency measured.
FDA's 510(k) database entry for K260811 lists Heartvue.Proton with Heartvue.ai, Inc. of Brentwood, Tennessee as the applicant. Innolitics, LLC of Austin, Texas appears in a different role: correspondent on the filing. The entry records the decision as "Substantially Equivalent (SESE)" under regulation 21 CFR 892.2050, with classification product code QIH and subsequent product code LLZ, review panel Radiology, review type Traditional, and no third-party review. It also records that a Predetermined Change Control Plan was authorized. On that point the announcement and the record agree exactly.
The dates are where the record says something the coverage does not. The entry gives a received date of March 12, 2026 and a decision date of August 26, 2026. March 12 to August 26 is 167 days. That arithmetic is this desk's, not a field in the file. Diagnostic and Interventional Cardiology told readers only that "The 510(k) was cleared approximately five and a half months after submission", with no start date, which leaves nothing a reader can check.
The benchmark comes from the advising firm's own publication. Innolitics' year-in-review page on AI/ML medical device 510(k) clearances reports "Median clearance time: 142 days" and an average of 150 days across 295 total clearances from 221 unique manufacturers. It also reports that 24% cleared in under 90 days and 22% took over 200 days for complex submissions. Measured against those figures, 167 days is 25 days longer than the median and 17 days longer than the average, and sits in neither of the tails the firm describes.
One caution about that comparison, and it cuts against the announcement rather than for it. The Innolitics page is a year in review, and the captured page does not state the period its dataset covers. K260811 was received in March 2026 and decided in August 2026. The windows are not known to match, so this is not established as a like-for-like comparison - the benchmark is the closest published yardstick the advising firm itself puts on the record, not a known same-period cohort. A median covering the window this review ran in, when someone publishes one, is the number that would settle it.
View the data
| Value (days) | |
|---|---|
| Heartvue.Proton, 2026 review | 167 days |
| Median clearance time, Innolitics review | 142 days |
| Average days to clearance, Innolitics review | 150 days |
Source: FDA 510(k) database K260811, interval between the recorded received and decision dates; Innolitics year in review, median and average clearance time · Daily Pol graphic
The change control plan is the part of the story the announcement treats as a distinction, and it is uncommon without being unprecedented. Innolitics' review reports "10% authorized with PCCPs for iterative updates". So roughly one AI/ML clearance in ten in that dataset arrived with the same instrument Heartvue.Proton has. The founder's statement is what frames it as a mark of the submission's quality; the captured trade coverage reports its presence flatly.
What such a plan permits is narrower than the language around AI products usually suggests. AuntMinnie describes the mechanism for this device as leave to "update its machine-learning models within pre-specified bounds without requiring a new 510(k) submission for each update" - bounded, agreed in advance, and written down. That is not a model that learns from the clinical data flowing past it. FDA's own product classification entry for code QIH, the code on this clearance, defines the category as covering "nonadaptive machine learning algorithms" trained on clinical or artificial data. The agency cleared software in the nonadaptive class and authorized a pre-specified update path. Read together, the two records set a ceiling on how much change is licensed here.
There is also a limit on what a substantial equivalence finding establishes. The captured database entry records the decision, the dates, the codes and the authorized plan. It carries no performance data, no indications-for-use text and no predicate device name. This desk sought a readable text version of a decision summary for K260811 and could not produce one, so nothing about the testing behind the clearance - what the software was measured against, on whose images, with what agreement to expert readers - is in evidence here and none of it is characterized above. The hour-per-study figure and the workflow claims come from the company's account as the trade press carried it. Beyond those published quotations, this desk holds no statement from Heartvue.ai or Innolitics about this clearance.
Two things a reader can check without waiting on anyone. The K260811 entry is public and will keep showing the received and decision dates that produce the 167-day figure. And the same database will show whether Heartvue.ai files again: if the authorized plan works as described, updates inside its bounds arrive without a new submission number. This desk's expectation, checkable on the FDA 510(k) database by March 1, 2027, is that no additional cleared submission listing Heartvue.ai, Inc. as applicant will appear by that date.