Model reference — AI Medical · Neurology

AI models for Brain MRI analysis

Three layers are normally combined. Brain extraction removes non-brain tissue and is close to solved — Dice around 0.98 even with pathology present. Anatomical segmentation labels structures, either fast and contrast-specific or contrast-agnostic and slower. Lesion models then work on top.

Contrast-agnostic models are the important development for clinical archives: they segment any MRI sequence, including scans acquired at clinical rather than research quality, which is what makes a retrospective archive usable 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.

Brain MRI analysis service AI Medical services Pricing

Input type — Brain MRI volumes

Models in this group take a brain MRI volume; the sample input is a single T1 or multi-sequence 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.

FastSurfer / FastSurferVINN

Vendor: DZNE Bonn (Reuter lab)

What it does: produces a 95-class whole-brain segmentation with surfaces and thickness statistics in about an hour rather than seven, and with better test-retest reliability than the classical pipeline it replaces.

Dice about 0.90 against FreeSurfer labels; full pipeline in about 1 hour against about 7, with better test-retest reliability (independent, NeuroImage).

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

SynthSeg / SynthSR

Vendor: MIT / Martinos Center (FreeSurfer)

What it does: segments any MRI sequence at any resolution, including clinical-quality scans — which is what makes a retrospective hospital archive usable at all.

Contrast- and resolution-agnostic; Dice within a few points of same-contrast supervised models on any MRI sequence (independent, Medical Image Analysis).

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

HD-BET

Vendor: DKFZ Heidelberg

What it does: strips the skull at Dice 0.98 even with pathology present or sequences missing. Effectively solved, and the first step in most brain pipelines.

Dice about 0.98 brain extraction, robust to pathology and missing sequences (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

HD-GLIO / nnU-Net BraTS models

Vendor: DKFZ Heidelberg

What it does: outlines glioma sub-regions with volumetry attached, at the accuracy level the BraTS leaderboards have converged on.

Dice 0.85–0.92 whole tumour, 0.78–0.87 enhancing tumour on BraTS (independent, challenge leaderboards).

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

LST-AI

Vendor: Technical University of Munich

What it does: segments MS lesions and labels them by McDonald-criteria region, returning count, volume and location — the reporting output a neurologist actually needs.

Dice about 0.70 for MS lesions, exceeding prior public tools; includes McDonald-criteria region labelling (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

BrainSegFounder

Vendor: BrainSegFounder authors

What it does: is a foundation model for neuroimaging segmentation across brain, tumour and lesion targets.

Task-dependent across neuroimaging segmentation datasets.

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

BrainIAC

Vendor: AIM-KannLab

What it does: covers classification, regression, segmentation and prognosis on brain MRI from one representation model.

Task-dependent across classification, regression, segmentation and prognosis.

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

BrainFound

Vendor: BrainFound authors

What it does: classifies neurodegeneration and grades tumours from brain MRI.

Task-dependent across neurodegeneration classification and tumour grading.

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

NeuroVFM

Vendor: MLNeurosurg collaborators

What it does: covers classification, report generation, retrieval, triage and registration on brain CT and MRI, which suits a neuro service consolidating several tools into one.

Task-dependent across classification, report generation, retrieval, triage and registration.

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

If your scans are research-quality T1, the fast pipeline is the obvious choice. If they are mixed clinical sequences, a contrast-agnostic model will hold accuracy where the fast pipeline will not. Tumour and MS work needs the dedicated lesion models; anatomical segmentation does not find lesions.

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.

Brain MRI analysis service AI Medical services Pricing

From benchmark to production

Send a representative sample, your expected volume and your latency target for brain 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.