This is an earlier-stage area than CT or MRI: accuracy is reported per dataset with Dice and IoU, and no model claims a general clinical standard.
Two routes are in use. A foundation model pretrained on ultrasound supports classification, segmentation and enhancement from one base. Prompted segmentation models, adapted from general promptable segmentation, produce a mask from a point or box with no training at all.
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.
Models in this group take an ultrasound frame or a prompted region within one. 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.
USFM
Vendor: OpenMedLab / Fudan University
What it does: is pretrained on ultrasound rather than adapted from photography, and covers classification, segmentation and image enhancement from one base — three tasks, one deployment.
Task-dependent across ultrasound disease classification, tissue segmentation and image-enhancement benchmarks.
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 (images/hour)
≈ 4,000
≈ 14,000
≈ 36,000
UltraSam
Vendor: CAMMA / University of Strasbourg collaborators
What it does: segments from a click or a box on an ultrasound frame, which gets usable masks without an annotation project first.
Dataset-dependent; evaluated using Dice and IoU across ultrasound segmentation datasets.
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 (images/hour)
≈ 4,000
≈ 14,000
≈ 36,000
SAMUS
Vendor: SAMUS authors
What it does: is a prompted ultrasound segmenter tuned for the modality, evaluated across several public ultrasound datasets.
Dataset-dependent; evaluated across multiple ultrasound segmentation datasets with Dice and IoU.
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 (images/hour)
≈ 4,000
≈ 14,000
≈ 36,000
Choosing between them
If you need several tasks from one deployment, the foundation model is the more economical base. If you need masks quickly on varied anatomy, a prompted model gets there without an annotation project. We benchmark on your own device output, since probe and preset variation moves these numbers more than model choice.
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.
Send a representative sample, your expected volume and your latency target for ultrasound 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.