The evidence here is unusually strong for open models: a risk model validated across seven international sites with a 1-year AUC around 0.84, and a malignancy classifier at AUC 0.895 with a reader study showing hybrid human-plus-model improvement.
What no open model matches is cleared-device mammography CAD. Where a cleared product is required, that is a commercial purchase; these models serve risk stratification, research and audit work.
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 a full-field digital mammogram, normally four views. 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.
Mirai
Vendor: MIT CSAIL / Jameel Clinic
What it does: estimates one-to-five-year breast cancer risk from the mammogram itself, validated across seven international sites — which is what allows screening intervals to be set by risk rather than by age alone.
5-year risk C-index 0.76–0.81; 1-year AUC about 0.84, validated across 7 international sites (independent, Science Translational Medicine).
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
NYU breast cancer classifier (DMV-CNN)
Vendor: NYU Center for Data Science and community
What it does: scores malignancy per breast at AUC 0.895, with a reader study showing that radiologist and model together beat either alone.
AUC 0.895 for malignancy on the NYU screening set; reader study showed hybrid human-plus-AI improvement (independent).
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 the question is "is there cancer now", the classifier is the model. If it is "who should come back sooner", the risk model is, and it is the more defensible of the two in a programme context. Both take a full four-view study, which is what the sizing below assumes.
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 breast cancer screening. 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.