Treatment effects were not identifiable in our data
not availableversus a no-treatment baseline
Fit result not available in this data file.
measured Details
In silico tumour growth · treatment what-if
What would this tumour do if the treatment had been different?
A growth simulation seeded from a real patient's scan segmentations and run forward from the last scan. It runs on one local machine, built so patient data never leaves the device.
- measured
- from the scans
- illustrative
- simulated
- sample data
- bundled sample
- AI-written
- checked text
Workspace
No patient
Choose a patient
0
Cell-density seed, illustrative. MRI does not measure cell density.
The 3D view is off because this browser has no WebGL 2. A 2D cross-section is shown instead; the chart and numbers work as usual.
Drag to orbit, scroll to zoom. The cut plane hides one side of the tumour so the inside is visible.
Timeline days since first scan
Day 0
View options: cross-section, opacity, colour scale
Growth curve: tumour burden over time
Patient none selected
Filter and sort
Model parameters
Read-only. These are the values the stored simulation used; this build does not re-run the simulation in the browser.
01Two agents and a rule-based judge
Every sentence carries its receipt
One agent writes the explanation and must cite a verified statement for every sentence. A second agent audits each sentence. A rule-based judge, which is not a language model, checks the numbers and wording and has the last word.
No explanation is stored for this patient.
02Cohort results
What the evidence says
Four results decide how much weight the simulation deserves. They are stated here in full, including the ones that go against the demo.
Mortality risk model
not availablepooled AUC
Risk results not available in this data file.
measured Details
What the scenarios assume
not availablegrowth suppression while an agent is active
Taken from trial progression-free-survival ratios. Not estimated here, and not independent of imaging pseudo-response for bevacizumab.
illustrative Sources
Does the model forecast the next scan?
measured Details
03Intended use, method and limits
Model card
Intended use
Exploring how a treatment assumption changes a growth forecast for a single patient's tumour, and showing what the assumptions are. It is a research demo for a hackathon audience.
Not for choosing or changing treatment, giving a prognosis, or any decision about a person.
What is simulated
A reaction-diffusion model: cells proliferate at rate ρ, spread with diffusion D, and are lost at rate δ. It is seeded from the patient's segmentation masks (core 0.95, enhancing 0.5, oedema 0.2; resection cavity excluded) and run forward from the last scan.
Tumour burden is the sum of density times voxel volume, in cell-equivalent mm³. MRI does not measure cell density, so the density colouring is a modelling choice.
Treatment-effect fit measured
Fit result not available in this data file.
| Agent | Fitted kill rate | Interval | Intervals exposed |
|---|---|---|---|
| not available | |||
Priors used in the scenarios illustrative
| Agent | Growth suppression while active | Source |
|---|---|---|
| Loading | ||
The temozolomide median PFS figure (6.9 vs 5.0 months) was not independently verified; the overall-survival hazard ratio of 0.63 was. Radiation uses a placeholder, not a measured effect.
Mortality risk model measured
This is a separate model from the growth simulation. It ranks patients by mortality risk and does not predict tumour growth.
Risk results not available in this data file.
Forecast check measured
Loading
| Method | Typical error (log ratio) | Typical factor off | Direction right |
|---|
Limits to keep in mind
- Scenario curves are what the model does under assumed effects. They are not what would have happened to this patient.
- Growth rate ρ is calibrated on a single interval per patient. Some fits do not converge; those are flagged and should not be read.
- Density is a seed built from segmentation labels, not a measurement. The 3D view is a 32×32×32 display grid, so fine structure is lost.
- Treatment records come from clinical sheets; where an end date was missing it was assumed, and the assumption is listed per patient.
- Missing values are shown as "not available". They are never replaced by zero.