FiveBrane · Datametior
I measure
your training data.
mētior — Latin for “I measure”
Datametior runs 10 clinical-grade measurements on your medical imaging dataset — in 20 minutes, from images and labels alone — and returns a benchmarked report every stakeholder in the room can read and act on.
Datametior · example report
DermSet-7 v2.1
14,382 images · 7 classes · June 2026
67
/ 100
Image quality
58
Class structure
46
Coverage
79
Fairness
80
Fitzpatrick V–VI absent — regulatory risk
18.4% artifact rate — shortcut learning risk
47× class imbalance — reweighting required
Sharpness 82/100 — clinical grade
What is
DATAMETIOR
Dataset intelligence built on FiveBrane's
medical imaging analytics engine.
DATAMETIOR is a dataset due diligence capability built directly on FiveBrane's existing platform infrastructure — photometric analysis, radiomics, and embedding intelligence — applied to the question that matters most before any training run: is this data good enough to build on?
It combines the clinical precision of FiveBrane's analytics with a standardised scoring methodology benchmarked against public medical imaging datasets, and packages the result into a report that travels from your data team to your board, your investor, and your clinical partner — without losing meaning along the way.
What we measure
10 clinical-grade measurements.
Built on FiveBrane's analytics stack — the same engine used across medical imaging AI development — applied to dataset quality for the first time.
01
⚖️
Class balance
Imbalance ratio — majority vs rarest condition
02
🔬
Image sharpness
Laplacian variance — clinical quality gate
03
💡
Brightness
Luminance distribution — exposure consistency
04
🔁
Near-duplicates
Cross-split duplicates — leakage detection
05
🎨
Omics based
Measures coverage — fairness baseline
06
🗺️
Semantic coverage
Embedding grid density — diversity map
07
🎛️
Contrast
RMS contrast per class — lesion visibility
08
🌐
Intra-class diversity
Embedding spread — variety within each class
09
📏
Artifact rate
Ruler marks, ink dots, frame edges
10
🏷️
Label noise
k-NN consistency — mislabel estimate
Why DATAMETIOR
One report. Three conversations.
Designed to be read by people with different questions — and give each one exactly what they need.
Find problems before training finds them for you.
Your model learns from whatever is in the data. Artifacts, leakage, imbalance — they become the model. Datametior surfaces them in 20 minutes instead of 3 months into a training run.
Know which issues move the score most
Shareable artifact for cross-team alignment
Versioned score tracks data improvement over time
Score the data asset, not just the demo.
Most technical DD stops at the model. The data — the actual competitive moat — is evaluated by intuition. Datametior gives you a benchmarked number before the term sheet.
Benchmarked vs 31 public medical datasets
PII, license risk, and leakage flags included
Investment memo section, written and defensible
Walk in with proof. Not a story.
When someone asks "how do you know your data is good?" — you answer with a score, a benchmark, and a methodology. Your data moat becomes a visible, verifiable asset.
Data room PDF ready for due diligence
Score you can defend in any room
Remediation roadmap — what to fix first
Process
From upload to report in 20 minutes.
No compliance paperwork. No data retained after analysis. No integration required.
01 —
Upload images & labels
Drop your image folder and label CSV. DICOM, PNG, JPG. Labels as class names. No preprocessing needed.
~ 2 min
02 —
FiveBrane measures
10 metrics run in parallel. Photometric, embedding, and label analyses from FiveBrane's clinical engine.
~ 15 min
03 —
Score & benchmark
Each metric benchmarked against public medical imaging datasets.
~ 2 min
04 —
Download your report
PDF with scores, findings, and a prioritised remediation roadmap — ready for your data room.
Instant
The report
A document that travels
without you.
Designed to be shared — to your PM, clinical partner, investor, or regulatory team — without a 30-minute explainer.
Benchmarked score, not just a number
61/100 means nothing alone. We tell you it sits below the 31-dataset median of 63 — and exactly why, dimension by dimension, with public reference datasets named.
Drill-down evidence for every flag
Every critical finding links to the raw distribution that triggered it. A technical reviewer can trace the score all the way to pixel-level evidence.
Prioritised remediation roadmap
Not a list of problems — a ranked plan. What to fix before your next training run, and what to fix before clinical deployment. Two separate lists.
Data room ready
PDF with methodology footnotes, benchmark sources, and versioned score history. The artifact investors ask for by name when evaluating data-driven AI companies.
FiveBrane · Datametior
DermSet-7 v2.1
14,382 images · 7 classes · June 2026 · v1.4
Image quality
58
Class structure
46
Diversity
79
Fairness
80
Fitzpatrick V–VI absent — FDA AI/ML SaMD regulatory risk
18.4% artifact rate — documented shortcut learning cause
47× class imbalance — weighted training required
331 near-duplicates cross train/test boundary
Sharpness 82/100 — top quartile vs 31 benchmarks
fivebrane.com · datametior
Download PDF
Benchmark
Your score in context.
Scored against 31 public and proprietary medical imaging datasets with Datametior methodology v1.4.
Your dataset
67
ISIC 2024
71
FiveBrane · Datametior
Your dataset is waiting
to be measured.
Upload your images and labels. Get a full Datametior report — benchmarked, scored, printable — in 20 minutes.