Model reference — AI Medical · Cardiology

AI models for Echocardiography analysis

Echocardiography models fall into two families. Task-specific networks return one measurement very accurately — ejection fraction, wall thickness — and are cheap enough to run on every study. Multi-view and vision-language models cover many findings at once and add retrieval, at a higher cost per study.

The published accuracy is unusually good in this area: EF error of around four percentage points is close to inter-reader variability, which is why this is one of the first places automated measurement has reached routine use.

Every model on this page runs as part of a managed AI pipeline in our GPU clusters, with a dedicated private cluster in our cloud or an on-premise installation where medical governance requires it. Output is decision support for a qualified professional to review, not a diagnosis.

Echocardiography analysis service AI Medical services Pricing

Input type — Echocardiogram video

Models in this group take an echocardiogram loop or a full multi-view study. Each table gives three hardware tiers — Minimum, the smallest configuration on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is what one server of that tier processes per hour. Use these figures for initial sizing only. Before production we benchmark your own data to confirm accuracy, latency, throughput and cost.

EchoNet-Dynamic

Vendor: Stanford University (Ouyang / Zou lab)

What it does: measures ejection fraction from an apical-four-chamber loop and returns the segmentation that produced it. Its 4.1-point error is close to inter-reader variability, which is why this is one of the first measurements to reach routine automated use.

EF mean absolute error 4.1 percentage points, R² 0.81; AUC 0.97 for detecting reduced EF (independent, Nature 2020 plus external replications).

Requirement	Minimum	Medium	High
GPU type	RTX 3060	RTX 4090	A100 80 GB
VRAM	12 GB	24 GB	80 GB
vCPUs	6	12	24
RAM	16 GB	32 GB	64 GB
Server	1× RTX 3060 12 GB	1× RTX 4090 24 GB	1× A100 SXM 80 GB
Rate (studies/hour)	≈ 900	≈ 3,200	≈ 8,100

EchoNet-LVH

Vendor: Stanford University (Ouyang / Zou lab)

What it does: measures left ventricular wall thickness and scores for amyloidosis and hypertrophic cardiomyopathy — conditions that are missed precisely because the measurement is tedious.

AUC 0.98 cardiac amyloidosis, 0.93 hypertrophic cardiomyopathy (developer-reported, Circulation 2022).

Requirement	Minimum	Medium	High
GPU type	RTX 3060	RTX 4090	A100 80 GB
VRAM	12 GB	24 GB	80 GB
vCPUs	6	12	24
RAM	16 GB	32 GB	64 GB
Server	1× RTX 3060 12 GB	1× RTX 4090 24 GB	1× A100 SXM 80 GB
Rate (studies/hour)	≈ 900	≈ 3,200	≈ 8,100

EchoNet-Peds

Vendor: EchoNet / Stanford collaborators

What it does: is the paediatric model: adult-trained echo networks do not transfer to children, so this is the one to use rather than a general model with a caveat.

Task-dependent; performance varies by paediatric echocardiography task.

Requirement	Minimum	Medium	High
GPU type	RTX 3060	RTX 4090	A100 80 GB
VRAM	12 GB	24 GB	80 GB
vCPUs	6	12	24
RAM	16 GB	32 GB	64 GB
Server	1× RTX 3060 12 GB	1× RTX 4090 24 GB	1× A100 SXM 80 GB
Rate (studies/hour)	≈ 900	≈ 3,200	≈ 8,100

EchoPrime

Vendor: Stanford University / Cedars-Sinai

What it does: reads a full multi-view study and returns measurements, report text and a searchable embedding, outperforming task-specific baselines across 23 echo tasks from one deployment.

Outperforms task-specific baselines across 23 echo tasks, AUC 0.80–0.96 (developer-reported).

Requirement	Minimum	Medium	High
GPU type	RTX 3090	L40S 48 GB	A100 80 GB
VRAM	24 GB	48 GB	80 GB
vCPUs	8	16	32
RAM	32 GB	64 GB	128 GB
Server	1× RTX 3090 24 GB	1× L40S 48 GB	1× A100 SXM 80 GB
Rate (studies/hour)	≈ 300	≈ 1,100	≈ 2,700

EchoCLIP / EchoCLIP-R

Vendor: Cedars-Sinai (CarDS lab)

What it does: scores findings from a text prompt with no training, and indexes an echo archive so it can be searched by description — AUC 0.86–0.90 zero-shot for pacemaker, severe aortic stenosis and low EF.

AUC 0.86–0.90 zero-shot for pacemaker, severe aortic stenosis and low EF (independent, Nature Medicine 2024).

Requirement	Minimum	Medium	High
GPU type	RTX 3090	L40S 48 GB	A100 80 GB
VRAM	24 GB	48 GB	80 GB
vCPUs	8	16	32
RAM	32 GB	64 GB	128 GB
Server	1× RTX 3090 24 GB	1× L40S 48 GB	1× A100 SXM 80 GB
Rate (studies/hour)	≈ 300	≈ 1,100	≈ 2,700

Choosing between them

If you need one number on every study, a task-specific model is the cheapest route to it. If you need many findings, or search across an archive, the multi-view models earn their hardware. Paediatric work needs the paediatric model — adult-trained networks do not transfer.

Accuracy figures above are those the producers and independent evaluations report, on their own test sets. They are a shortlist tool, not a prediction of what you will see. At the start of a project we run a short proof of concept on a sample of your own data, which replaces them with real figures — so the cost and the schedule for the full engagement are known before anything is committed.

Echocardiography analysis service AI Medical services Pricing

From benchmark to production

Send a representative sample, your expected volume and your latency target for echocardiography analysis. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster — with the cost per unit of work known before you commit.