For prostate, the reported figures are strong and independently evaluated: patient-level AUROC 0.85–0.91 for clinically significant cancer, matching radiologists in the challenge reader study. That is one of the few places in radiology where an open model has that level of evidence.
Multi-sequence foundation models cover a much wider range — one reports across 44 MRI clinical tasks — with accuracy that is task-dependent by construction.
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 one or more MRI sequences as a 3D 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.
What it does: produces a lesion heatmap and a patient-level risk score for clinically significant prostate cancer, and matched radiologists in the challenge reader study — the strongest evidence behind any open model on this page.
Patient-level AUROC about 0.85–0.91 for clinically significant prostate cancer; matched radiologists in the PI-CAI reader study (independent).
Requirement
Minimum
Medium
High
GPU type
RTX 3090
A100 80 GB
H100 80 GB
VRAM
24 GB
80 GB
80 GB
vCPUs
12
24
48
RAM
64 GB
128 GB
256 GB
Server
1× RTX 3090 24 GB
1× A100 SXM 80 GB
2× H100 SXM 80 GB
Rate (volumes/hour)
≈ 60
≈ 210
≈ 540
MRI-PTPCa
Vendor: MRI-PTPCa authors
What it does: predicts prostate cancer probability and grade using pathology-aligned context alongside the MRI.
Task-dependent across prostate-cancer diagnosis and grading cohorts.
Requirement
Minimum
Medium
High
GPU type
RTX 3090
A100 80 GB
H100 80 GB
VRAM
24 GB
80 GB
80 GB
vCPUs
12
24
48
RAM
64 GB
128 GB
256 GB
Server
1× RTX 3090 24 GB
1× A100 SXM 80 GB
2× H100 SXM 80 GB
Rate (volumes/hour)
≈ 60
≈ 210
≈ 540
MARS
Vendor: MARS authors
What it does: covers 44 reported MRI clinical tasks — classification, segmentation, registration, reporting and prognosis — from one multi-sequence model, which is the argument for it over several single-task deployments.
Task-dependent across 44 reported MRI clinical tasks.
Requirement
Minimum
Medium
High
GPU type
RTX 3090
A100 80 GB
H100 80 GB
VRAM
24 GB
80 GB
80 GB
vCPUs
12
24
48
RAM
64 GB
128 GB
256 GB
Server
1× RTX 3090 24 GB
1× A100 SXM 80 GB
2× H100 SXM 80 GB
Rate (volumes/hour)
≈ 60
≈ 210
≈ 540
Choosing between them
For a single indication with a public challenge behind it, the challenge-lineage model is both strongest and easiest to validate. For broad coverage, a multi-sequence foundation model is the base, with your own head trained on top. MRI sizing is memory-driven; a multi-sequence study needs the higher tiers.
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 mri 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.