Knee MRI,Read twice.Every time.

A multimodal AI second read for knee MRI. Twelve abnormalities, one model, structured for the radiologist who signs the report.

FINDINGS
12
binary abnormalities scored per study
TRAINING LABELS
4,400
free-text reports parsed in ~12 languages
SITES
19
scanner sites standardized to one 140 mm crop
01 — THE PROBLEM

Subtle knee findings get missed. Not for lack of skill, for lack of time.

10–25%

miss rate reported for specific knee MRI findings, varying with reader experience.

1.6

résumés per open radiologist vacancy in Russia. New modalities (CT, MRI, PET-CT) keep adding reading time.

90 min

for a detailed read of one study. A neural network returns its analysis in about 3 seconds.

02 — HOW IT WORKS

Standardize the image first. Then read.

Most tools bet on chemistry-grade model tricks. We bet on the physics of the data: every study, whatever the scanner or protocol, becomes the same fixed tensor before a model ever sees it.

STEP 1 · NORMALIZE

Fixed slot tensor

3 planes × 3 slices, 140 mm crop, 224 px. Series are mapped into consistent slots; a missing-slot mask records absent acquisitions instead of faking them.

STEP 2 · LABEL

Noise-tolerant labels

A multilingual parser with negation handling turns ~4,400 free-text reports in ~12 languages into training labels. No patient metadata, no manual structured input.

STEP 3 · MODEL

12 findings, one network

ResNet-18 baseline scaling to DINOv2/v3 ViT-S with per-finding attention pooling. Rank-space ensembling, evaluated by macro ROC-AUC across all 12 findings.

STEP 4 · REVIEW

Radiologist decides

Output arrives as a structured second-read checklist with confidence per finding. The workflow stays the same; the report is still signed by a human.

03 — DEMO

Variable studies in. One review format out.

Sagittal knee MRI, source study
Before19 sites · mixed protocols
Normalized3×3 slots · 140 mm · 224 px
ACL tear0.91
Meniscal tear, medial0.84
Bone marrow edema0.62
Cartilage defect0.37
Joint effusion0.29
PCL tear0.06
Second readillustrative scores
“A detailed analysis of a single image takes about 90 minutes. At the same time, a neural network will spend 3 seconds analyzing and interpreting one examination.”
Nikolai Makeev
Medical Director, Care Mentor AI & Doctor Smart, Cand. Med. Sci.
Source: Medvestnik, “What Hinders the Introduction of AI in Russian Clinics”
04 — TEAM 45 · IW-2026

Five people, one team.

Artem Mitin
Artem MitinAI & ML
Oleg Evtushenko
Oleg EvtushenkoAI & ML
Linda Hedhli
Linda HedhliSoftware & UX
Patricia Perez Hernandez
Patricia Perez HernandezProduct & Strategy
May Ulshin
May UlshinResearch & Partnerships
05 — FAQ

Questions radiologists ask first.

Any vendor. The pipeline standardizes series from 19 sites into a fixed 3×3 slot tensor (three planes, three slices) with a 140 mm crop at 224 px. Sequences that are absent are recorded by a missing-slot mask rather than substituted.

06 — NEXT STEP

Read with us.

We are interviewing radiologists about workflow fit. Twenty minutes, your schedule. Or send a de-identified knee case and we return the 12-finding read.