MAISI-v2

Run this model →
Version 1.1 · Licence Open Model License · Released 2026-08-23
CT / MRI

MAISI-v2 generates synthetic CT or brain MRI images without an input scan. Create CT images with optional anatomy masks, or brain MRI images in one or several contrasts.

Available tasks (2)

Task options define what the model does. Choose one when you submit a scan or configure generation.

CT generation — image / image + mask

Default

Synthetic CT generation (rflow-ct). Image-only, or image + paired anatomy mask with controllable organ and tumour sizes. X/Y ∈ {128,256,384,512}, spacing 0.5–5 mm, FOV ≥ 256 mm.

Input
No input scan. Choose the generation settings.
Output
Synthetic CT image, optionally with a paired anatomy mask (.nii.gz).
View 124 output classes
1 liver3 spleen4 pancreas5 right kidney6 aorta7 inferior vena cava8 right adrenal gland9 left adrenal gland10 gallbladder11 esophagus12 stomach13 duodenum14 left kidney15 bladder17 portal vein and splenic vein19 small bowel22 brain23 lung tumor24 pancreatic tumor25 hepatic vessel26 hepatic tumor27 colon cancer primaries28 left lung upper lobe29 left lung lower lobe30 right lung upper lobe31 right lung middle lobe32 right lung lower lobe33 vertebrae L534 vertebrae L435 vertebrae L336 vertebrae L237 vertebrae L138 vertebrae T1239 vertebrae T1140 vertebrae T1041 vertebrae T942 vertebrae T843 vertebrae T744 vertebrae T645 vertebrae T546 vertebrae T447 vertebrae T348 vertebrae T249 vertebrae T150 vertebrae C751 vertebrae C652 vertebrae C553 vertebrae C454 vertebrae C355 vertebrae C256 vertebrae C157 trachea58 left iliac artery59 right iliac artery60 left iliac vena61 right iliac vena62 colon63 left rib 164 left rib 265 left rib 366 left rib 467 left rib 568 left rib 669 left rib 770 left rib 871 left rib 972 left rib 1073 left rib 1174 left rib 1275 right rib 176 right rib 277 right rib 378 right rib 479 right rib 580 right rib 681 right rib 782 right rib 883 right rib 984 right rib 1085 right rib 1186 right rib 1287 left humerus88 right humerus89 left scapula90 right scapula91 left clavicula92 right clavicula93 left femur94 right femur95 left hip96 right hip97 sacrum98 left gluteus maximus99 right gluteus maximus100 left gluteus medius101 right gluteus medius102 left gluteus minimus103 right gluteus minimus104 left autochthon105 right autochthon106 left iliopsoas107 right iliopsoas108 left atrial appendage109 brachiocephalic trunk110 left brachiocephalic vein111 right brachiocephalic vein112 left common carotid artery113 right common carotid artery114 costal cartilages115 heart116 left kidney cyst117 right kidney cyst118 prostate119 pulmonary vein120 skull121 spinal cord122 sternum123 left subclavian artery124 right subclavian artery125 superior vena cava126 thyroid gland127 vertebrae S1128 bone lesion132 airway

MRI brain generation — image / cross-contrast

Synthetic brain MRI generation (rflow-mr-brain). Image-only in a chosen sequence (T1/T2/FLAIR/SWI, whole-brain or skull-stripped), or cross-contrast — the same synthetic brain rendered in several sequences as a co-registered set. X/Y ∈ {128,256,384,512}, spacing 0.5–5 mm.

Input
No input scan. Choose the generation settings.
Output
Synthetic brain MRI image or co-registered multi-contrast images (.nii.gz).

Research details

Training data
MAISI-v2. CT weights trained on large multi-institutional CT cohorts. Brain MRI weights are NVIDIA's NV-Generate-MR-Brain v1 (August 2026), trained on MR-RATE batches 0–27: 73,516 unique subjects / 665,371 scans across T1w, T2w, FLAIR, SWI and MRA, whole-brain and skull-stripped. Adult human anatomy.
Intended use
Research generation of synthetic 3D CT or brain MRI volumes for dataset creation and augmentation, with optional paired CT masks. Synthetic images must not be used as real patient data.
Known limitations
Output quality depends strongly on field of view (size × spacing) — out-of-distribution fields of view can produce unrealistic volumes even when accepted. CT masks contain only the organs you select. Brain MRI cross-contrast generation uses the same seed to synthesise several sequences; it does not translate a real input scan.
Species
human

Citations & references

Citing Nalvera.AI

Please cite the platform alongside the model's own paper listed above — the model authors deserve the credit for the method.

@misc{nalvera,
  title        = {Nalvera.AI: a cloud platform for medical-imaging AI},
  author       = {{van der Heyden, Brent}},
  year         = {2026},
  howpublished = {\url{https://nalvera.ai}},
  note         = {Model: MAISI-v2 v1.1. Accessed YYYY-MM-DD}
}
Research use only — not a medical device. Outputs are not validated for clinical diagnosis or treatment.
Run this model → ← Back to models