Fosvel FDG
Fosvel FDG · Whole-body PET/CT

Finding by finding, scan after scan.

A research model that aims to outline FDG-avid findings on whole-body PET/CT, follow them across a patient's scans, and show why each one was flagged.

Research use only · Not a medical device · Not CE-marked
Public research data
Why it is needed

One scan. Many findings.

A whole-body FDG PET/CT can hold many findings, from head to thigh. Reading it means finding each one, telling it apart from normal uptake in the brain, heart, kidneys, bladder or muscle, measuring it, and at follow-up matching it to the scan before.

It is careful, repetitive work. Small and faint findings are the easiest to miss, and the hardest to follow over time.

What it does today · research
01 · Detect

Head to thigh, in one pass.

One pass over the whole-body scan. The research model aims to outline and measure FDG-avid findings. Small, faint findings are still its hardest case, and the focus of our current research.

Public research data
02 · Follow-up

The same lesion, found again.

On research data, Fosvel aims to match each finding to the previous scan and to show which ones look smaller, larger, stable, no longer seen — or new.

Baseline → Follow-up · Public research data
Smaller
No longer seen
One research finding up close, axial and coronal PET, with the liver reference region nearby
Public research data
03 · Explain

Why it was flagged.

For each research finding: where it is, how its uptake compares with the liver reference and the tissue around it, and the model's score for it — so a reader can check it.

Tested on public data · research

What we measured, and on what.

Training
902 scans
800 patients · public dataset
Internal test
53 scans
46 patients · internal test · public dataset
Lesions ≥ 0.5 mL found
90%
n = 53 scans · internal test · public dataset
Dice median · internal
0.81
n = 27 positive scans · internal test · public dataset
False findings / scan · internal
3.3
n = 53 scans · internal test · public dataset
External test
376 scans
external test · 4 public datasets
Dice median · external
0.62–0.66
n = 100 + 57 scans · external test · 2 public datasets
Key parameter
98–99%
SULpeak within 15% of reference · 4 public datasets

Trained on 902 public PET/CT scans (800 patients); tested on 53 internal and 376 external scans from 4 independent public datasets.

Internal: same dataset as training, other patients (slightly optimistic). External: other public datasets. Dice: overlap with expert outlines, 0 to 1. Agreement with expert outlines, not clinical outcomes. SUV: computed from public data; calibration not independently confirmed. Very small lesions (under 0.5 mL) remain hard for the model. Research use only. Not a medical device. Results on public, retrospective datasets; not validated for clinical use.

What we hope next

A research assistant that keeps count.

Our goal is a research assistant that keeps track of every finding across all of a patient's scans, prepares the measurements a report needs, and points to what changed.

The aim: more of the physician's time for decisions, less for counting. These are goals of our research, not features of a product.

Why it matters to us
FDG is the workhorse of PET. Getting the whole-body read right is the ground everything else at Fosvel stands on.

Show us a hard case.

contact@fosvel.com
Please do not send patient data.