SCALE Consolev0.4.5
Windowsx64 · Windows 10 or 11
SCALE-Installer, per user, no administrator rights · 219 MB
AAPM Working Group SCALE
SCALE Console
A desktop application for participating sites. It turns a site's DICOM data into one lesion mask per tumor in the shared MNI brain space, plus a clinical table with no identifiers, and checks the data between every step. Status: in development.
Pipeline
Eight steps in three zones
Identifiable data never leaves the site computer. De-identified data is checked after every step. Only the purple zone leaves the institution, and the site sees it file by file before it is created.
Stays on the site computerIdentifiable data
- 1Find casesFinds each MRI and its tumor contours in the DICOM folder, including inside zip files.
Unpack zips → Read headers → Pair MRI with contours
Checked before the next step
- Case list with every case and its structures
- Identifiable: stays on this computer
Needs: 2 CPUs · 2 GB · 10 GB disk
- 2Name structuresMaps every contour name to the AAPM TG-263 standard. A person confirms the names it is unsure about.
Match names to TG-263
Checked before the next step
- Every structure has a decision
- Names come from the TG-263 vocabulary
Needs: 1 CPU · 1 GB
- 3De-identifyRemoves all patient identifiers and checks every file before anything is released.
Rebuild files → Verify every file → Publish release
Checked before the next step
- No identifier in any file (privacy check)
- Inventory and checksums
- MRI and contours still match
Needs: 4 CPUs · 4 GB · disk ≈ size of the MRIs
De-identifiedChecked after every step
- 4Map to MNIRegisters each MRI to the MNI brain and moves the tumor masks into it.
Convert → Fix intensity → Brain mask → Register to MNI → Move tumor masks
Checked before the next step
- Every mask on the 197 × 233 × 189 MNI grid
- One mask per tumor
- Volumes match the original contours
Needs: NVIDIA GPU ≥ 8 GB (required) · 32 GB RAM · ≈ 0.5 GB per patient
- 5Check qualityMeasures how well each brain was registered and flags the cases a person should look at.
Measure registration → Draw overlays → Flag outliers
Checked before the next step
- Every case accepted or excluded with a reason
Needs: 4 CPUs · 8 GB
- 6Add clinical dataA clinical workbook is filled for the site's patients. Only de-identified values are kept.
Check workbook → Keep de-identified values
Checked before the next step
- No names, IDs, dates or comments
- Ages as decades
- One row per patient
Needs: Excel or Numbers
Leaves the institutionMasks and a clinical table only
- 7Prepare submissionPackages the lesion masks and the clinical table, and shows exactly what will leave the institution.
Select accepted lesions → Build package → Check package
Checked before the next step
- Only masks and the clinical table
- Final privacy check
- Checksums
Needs: 1 CPU · disk ≈ 5 MB per patient
- 8Build lesion mapsAt UCSF: the submissions of all sites are checked with the same checker and combined into lesion probability maps per cancer type.
Runs at UCSF
Decisions
A person decides four times
Step 2Structure namesConfirm the contour names the Console could not map to TG-263 with confidence.
Step 3Cases and releaseReview the images of each case for burned-in text or faces, confirm the de-identification policy and that the release is covered by the IRB and data agreements, then approve.
Step 5Quality flagsAccept or exclude each flagged case, for example a low registration score, with a reason.
Step 6Clinical workbookFill the workbook. It shows patient names so staff know who is who; the workbook never leaves the computer.
Everything else runs on its own. A step starts only when the data it needs passed the check and the computer has what the step needs.
Release
What leaves the institution
| File | Content |
|---|---|
| Lesion masks | One per accepted lesion, in MNI space |
| lesions.csv | Volume and MNI position of each lesion |
| clinical.csv | Age decade, sex, primary site, histology, subtype, KPS, treatment flags |
| quality.csv | Accepted or excluded, with reasons |
Not included: MRI images, transforms, names, IDs, dates, comments.
Containers
Each step is a verified container
| Container | Steps | Runs on | Data it may see | Status |
|---|---|---|---|---|
| scale-deid | Find cases · De-identify · Add clinical data | CPU | May see identifiable data · never network | In the Console |
| scale-tg263 | Name structures | CPU | May see identifiable data · never network | Built, not yet in the Console |
| scale-mni | Map to MNI | NVIDIA GPU (required) | De-identified data only | In development |
| scale-qc | Check quality | CPU | De-identified data only | Planned |
| scale-submit | Prepare submission | CPU | De-identified data only | Planned |
| scale-validate | The check between every step | CPU | Independent of all other containers | Built, not yet in the Console |
| scale-lesionmap | Build lesion maps (UCSF) | CPU | De-identified data only | Planned |
Containers ship inside the Console and are checked against their SHA-256 before use. They never have network access, and the DICOM folder is always mounted read-only.
Requirements
What a site computer needs
| Step | Needs |
|---|---|
| 1 Find cases | 2 CPUs · 2 GB · 10 GB disk |
| 2 Name structures | 1 CPU · 1 GB |
| 3 De-identify | 4 CPUs · 4 GB · disk ≈ size of the MRIs |
| 4 Map to MNI | NVIDIA GPU ≥ 8 GB (required) · 32 GB RAM · ≈ 0.5 GB per patient |
| 5 Check quality | 4 CPUs · 8 GB |
| 6 Add clinical data | Excel or Numbers |
| 7 Prepare submission | 1 CPU · disk ≈ 5 MB per patient |
Step 4 runs only on an NVIDIA GPU, so results match the UCSF atlas. On a computer without one, the Console offers to send the de-identified release to UCSF for processing (Route A). On a two-GPU workstation, mapping 2,000 patients takes about two and a half days; every other step takes minutes plus review time.
Download
SCALE Console 0.4.5
Installer, per user, no administrator rights · 219 MB
SCALE Consolev0.4.5
macOSApple silicon (M1 and later)
SCALE-App in a zip · 229 MB
SCALE Consolev0.4.5
macOSIntel
SCALE-App in a zip · 235 MB
SCALE Consolev0.4.5
Linuxx86_64
SCALE-AppImage, no installation · 224 MB
Each download includes
- The SCALE Console desktop application
- SCALE De-identification 0.3.0, built in and installed on first start; no internet download needed
- Getting Started guide and SHA-256 checksums of every file
This computer needs
- Docker Desktop 24 or newer (on Windows with WSL 2); no Docker account
- 2 GB of memory for Docker, 1 CPU and 2 GB of free disk for the de-identification step
- An NVIDIA GPU only for the later mapping step
Released 2026-10-27. The installers are not yet code-signed, so Windows and macOS ask for confirmation on first start (Windows: More info → Run anyway; macOS: right-click the app → Open). Check the SHA-256 of the file you downloaded against the one shown here. All files, checksums and the Getting Started guide: release page.