Highest specificity. Highest PPV.
Of any FDA-cleared echo AI for cardiac amyloidosis. At the prevalence of undiagnosed amyloidosis in a real echo lab, the number that decides a program's workload is positive predictive value — how often a flagged study is really amyloid — and that is governed by specificity.
When a study is flagged, how often is it amyloid?
Expected outcome in 1,000 echoes, with 1 in 100 carrying undiagnosed amyloidosis. Each square is one flagged study; a hollow square is a true case the model did not flag.
The two simulated rows are modelled operating points, not InVision. Each row's counts are its specificity and sensitivity applied to the same hypothetical cohort — 10 × sensitivity found and 990 × (1 − specificity) false flags, rounded. Full method and sources are in the notes at the foot of the page.
The trade-off behind the numbers
There is an intrinsic trade-off between sensitivity and specificity. A high-sensitivity operating point catches more of the true cases but flags many more studies that turn out negative; a high-specificity operating point flags fewer studies, so a larger share of its flags are real — at the cost of missing a few true cases.
InVision Precision Cardiac Amyloid sits at the high-specificity end: 99.0% specificity and 60.7% sensitivity in the 510(k) validation. In the cohort above it misses 4 of the 10 true cases a higher-sensitivity point might catch — but it flags 16 studies instead of 110 or more, so a flagged study is far more likely to be amyloid. For a confirmatory pathway with finite scintigraphy and specialist capacity, the false-flag burden is usually the constraint that matters, and it is governed by specificity far more than by sensitivity.
Positive predictive value ranks the operating points by their likelihood ratio, so InVision's PPV lead holds at every prevalence. The seven questions to ask any amyloid-AI vendor set out how to interrogate specificity, PPV, and intended use for your own program.
Side by side
All three operating points applied to the same 1,000-echo cohort at 1% prevalence. The simulated rows are modelled comparison points, not InVision.
| Model | Specificity | Sensitivity | Studies flagged | Flags that are amyloid | About |
|---|---|---|---|---|---|
| InVision Precision Cardiac Amyloid | 99.0% | 60.7% | 16 | ~38% | 2 in 5 |
| SIMULATED | 89.7% | 84.5% | 110 | ~7% | 1 in 14 |
| SIMULATED | 87.4% | 86.9% | 134 | ~7% | 1 in 15 |
"Flags that are amyloid" is the positive predictive value at 1% prevalence. It rises with prevalence: in a higher-risk population every operating point's flags are more often real.
The cleared indication
The FDA indication for use decides which patients a device may be used on, and it is set by the clearance, not by marketing. InVision Precision Cardiac Amyloid has the broadest cleared indication of any FDA-cleared echo AI for cardiac amyloidosis — the only one with no heart-failure and no wall-thickness prerequisite: it applies across adults aged 65 and older undergoing echocardiography.
That matters because the whole of that echo volume is eligible rather than a pre-filtered subset, and because cardiac amyloidosis — light-chain amyloidosis in particular — does not always present with heart failure or a thickened ventricle, so a prerequisite gate leaves some cases outside the cleared use.
Frequently asked questions
Why does InVision Precision Cardiac Amyloid have the highest specificity and PPV for cardiac amyloidosis?
It is tuned for specificity: 99.0% specificity with 60.7% sensitivity in the 510(k) validation, the highest specificity of any FDA-cleared echo AI for cardiac amyloidosis. At the low prevalence of undiagnosed amyloidosis in a real echo lab, high specificity drives positive predictive value, so a flagged study is far more likely to be a true case. Positive predictive value depends on prevalence, and these are cross-study figures from the FDA 510(k) summaries, not a head-to-head trial.
Why does InVision flag fewer studies?
Because it is tuned for specificity. In a hypothetical 1,000 echoes at 1% prevalence it flags about 16 studies, of which about 6 are amyloid; a lower-specificity operating point flags 110 or more, the large majority of which are false positives that each trigger a confirmatory work-up. Fewer false flags means a workable confirmatory pathway.
Does higher specificity mean InVision misses more cases?
It accepts lower sensitivity for higher specificity, so at a single operating point it catches fewer of the true cases than a higher-sensitivity point. The trade is deliberate: at low prevalence a high-specificity flag is far more likely to be real, and a negative result carries a high negative predictive value. Sensitivity and specificity should always be read together.
Are the comparison figures from a head-to-head study?
No. The simulated operating points are built from published FDA 510(k) summary specificity and sensitivity applied to the same hypothetical cohort. This is a cross-study comparison, not a head-to-head trial; it illustrates how specificity drives positive predictive value at realistic prevalence.