Learning center

DCE Tofts: concentration inputs, parameter units and numeric map exports

Follow one six-voxel synthetic extended-Tofts example through explicit timing, plasma AIF, fit scope and separate NPY/NIfTI exports.

Written guide to the DCE workflow in Medical Image Processing’s Perfusion Analysis app. One noiseless synthetic example contains six known voxels and a constant plasma input curve. It supports a bounded review of the selected extended-Tofts calculation and numeric export package. It contains no patient data and does not establish clinical performance or acceptance of a new release.

1. Preserve the dynamic array, timing, mask and concentration domain

Use Perfusion Analysis, identified as medical_image/perfusion_analysis. The reviewed workflow uses the app’s LOAD DYNAMIC, LOAD METADATA, LOAD MASK and LOAD AIF actions. Keep those files together. Medical upload access and the full launcher journey remain separate from this local app-specific example.

The saved dynamic.npy has shape 17 × 2 × 3, declared as TYX: seventeen times and a two-row, three-column plane. Its values are already tissue concentrations in millimolar (mM). The mask is a matching 2 × 3 array selecting all six voxels. These are six constructed curves, not a segmented organ, a 3-D volume or a measured perfusion scan.

The metadata provides seventeen frame times from 0 to 240 seconds at 15-second intervals, explicit seconds units, spatial spacing and an affine. The app converts the times to minutes for kinetic fitting. Preserve the distinction between frame times, spatial axes and physical coordinates; changing an axis label or time unit changes the calculation.

This example bypasses signal-to-concentration conversion because concentration is supplied explicitly. Raw SPGR signal requires additional calibration and baseline information in the current contract. Do not label arbitrary image intensities as mM. The existing input-preparation lesson explains why a loaded array alone does not supply its units or acquisition meaning.

2. Declare the input curve and model assumptions

The saved aif.json contains seventeen values of 1 mM and explicitly declares the plasma compartment. This constant step input makes the synthetic reference transparent; it is not a physiological population AIF or a measured arterial curve. The AIF must align with the supplied time samples and concentration units.

Current source also has a whole-blood input path that requires hematocrit for conversion to plasma. That path is outside this native example. Preserve the compartment declaration instead of treating plasma and whole-blood concentration as interchangeable.

Select DCE and extended_tofts. This model includes a plasma-volume term in addition to extravascular exchange; the standard-Tofts option omits that term. The implementation integrates a piecewise-linear sampled AIF and assumes an empty extravascular compartment at the first supplied time. It does not infer earlier input history or frame-averaged measurements.

For the reviewed staged-package sequence, use LOAD DYNAMIC for dynamic.npy, then LOAD METADATA, LOAD MASK and LOAD AIF for their matching files. Set Mode to DCE and DCE model to extended_tofts. Leave Axis order and Time units at From metadata so the saved TYX and seconds declarations apply. The fixture explicitly enabled Method/profile confirmed, Motion/QC acceptable or corrected and Auxiliary grids aligned. These confirmations declare checked assumptions; they do not create missing acquisition evidence. Preserve the actual grid and method information supporting them.

3. Select the five reviewed routes

The saved run selects DCE signal/concentration curve, DCE Ktrans map, DCE ve map, DCE kep map and Tofts fit evidence. Other DSC, ASL and perfusion routes remain outside this example. The vp map is included through Tofts fit evidence; it is not a sixth selected route.

After Run, inspect each selected route’s status and the fit evidence. The saved run completed all five routes, with no unavailable, error, resource-limited or cancelled route. Its receipt still says validated: false. Successful computation is distinct from scientific validation of real acquisition data.

The native viewer contains seven plots and five tables for this run. Its fit curve represents a selected converged voxel; one displayed curve is not a visual review of every voxel. Preserve the per-voxel maps and diagnostic outcomes alongside the figure.

4. Keep parameter units and reference scope explicit

The Tofts standardization paper defines the kinetic quantities and their relationship. In this implementation, Ktrans and kep use inverse minutes, while ve and vp are fractions. They are model parameters, not direct cerebral-blood-flow measurements.

ParameterUnit and first synthetic voxel
KtransTransfer constant, 1/min. The known first-voxel value is 0.3.
veExtravascular extracellular volume fraction. The known value is 0.2, or 20% when explicitly expressed as a percentage.
vpPlasma volume fraction. The known value is 0.02, or 2%.
kepRate constant, 1/min, with kep = Ktrans / ve. The known first-voxel value is approximately 1.5.

The existing root receipt compares all 24 parameter values across six voxels. Its absolute tolerances are 0.00002 for Ktrans, ve and vp, and 0.0002 for kep, in their respective units. These are comparison tolerances for this fixture, not uncertainty intervals or clinical thresholds.

The saved first Ktrans estimate is 0.2999999999826566. A displayed 0.3 and the saved estimate have different precision. The public evidence inventory retains the full existing values and errors; no fit is repeated for this guide. The OSIPI extended-Tofts examples use other reference data and conditions. Reading that reference does not turn this constant-input fixture into an OSIPI dataset comparison.

5. Inspect Export Latest as a numeric package

After the five routes finish, the recorded action order is EXPORT LATEST, then OPEN VIEWER. The source also offers EXPORT PARENT to choose an existing destination folder; the recorded workflow used a preset parent, so that picker interaction was not separately tested. The exporter creates a new mri_perfusion_… subdirectory and refuses a missing or symlinked parent. Keep each resulting manifest with its own run.

Use EXPORT LATEST after the selected run completes. Its reviewed directory contains fourteen array files plus manifest.json. There are seven map quantities, each in NPY and NIfTI format:

Map groupSaved quantities
Kinetic parametersKtrans, ve, kep and vp: four maps in two formats, giving eight files.
Fit diagnosticsFit-state codes, dimensionless R-squared and RMSE (mM): three maps in two formats, giving six files.
ManifestFourteen file identities, source-manifest identity, route statuses and the explicit validation/export limitations.

The NPY maps retain the internal 2 × 3 grid. The NIfTI writer transposes this 2-D grid to 3 × 2 and writes float32 values with the declared affine. Its manifest’s shape field describes the internal map, so a different NIfTI header shape is expected here. Compare coordinates and axis conventions rather than reshaping by eye. The existing native receipt separately checked the affine when reopening the files.

NPY parameter maps retain float64 values; NIfTI maps use float32 and can differ through rounding. Root’s additive readback compared all four parameter maps in both formats with the retained truth: 48 format-specific values, representing the same 24 parameters. The other diagnostic maps were reopened against the app’s saved route arrays; this is not a separate independent validation of their scientific meaning.

The guide rechecks the fifteen file identities and sixteen small NPY/NIfTI headers, including the dynamic and mask inputs. It reuses the existing numerical comparisons. This package contains no viewer CSV or figure export and is not a DICOM Parametric Map, clinical report or reloadable app session. Opening the viewer does not establish acceptance of every viewer export action.

Exact map names in the accepted package workflow

The 12 November 2026 package run retained these seven filename stems. Each has one .npy file and one .nii.gz file, plus a single manifest.json for the directory. The leading numbers follow the exporter’s sorted map names, not the order in which routes were selected. Use the manifest for a different route selection.

MapExact filename stemNPY type
kep (1/min)001_dce_kep__kep_map_per_minfloat64
Ktrans (1/min)002_dce_ktrans__ktrans_map_per_minfloat64
Fit-state code003_dce_tofts_model__fit_state_mapuint8
R-squared004_dce_tofts_model__r_squared_mapfloat64
RMSE (mM)005_dce_tofts_model__rmse_map_mMfloat64
vp fraction006_dce_tofts_model__vp_mapfloat64
ve fraction007_dce_ve__ve_mapfloat64

All seven NIfTI counterparts use float32. In particular, the NPY fit-state map contains uint8 codes; the corresponding NIfTI stores those codes as floating values. Preserve the code meaning rather than interpreting it as a continuous fitted parameter. Match the export’s run_id and source_manifest_sha256 to the completed run receipt. The latter identifies that run’s source manifest; it is not the package’s release.json hash.

6. Retain the input and the model boundary

Keep the original dynamic array, metadata, mask and AIF separately from the exported maps. The package withholds source paths and identifiers in this reviewed configuration, but that is not a general de-identification certification for other inputs or companion files.

The synthetic example tests neither measurement noise nor motion, unknown bolus history, signal calibration, patient anatomy or scanner conformance. It does not establish the suitability of an AIF, threshold or model for a patient. The corrected integration rule is exact for its stated piecewise-linear AIF representation; that numerical property is distinct from physiological accuracy.

Read complete displayed rows

In the later viewer, select a row in the main table or side table, then choose View full row. With the table focused, Enter also opens the selected row; you can also double-click it. The Full table row window shows each column heading and its complete displayed value in read-only, wrapped text. Select and copy the text you need, and scroll if necessary. Press Escape or choose Close to dismiss it. If no row is selected, select one first. Long result names also have a horizontal scrollbar.

The later 12 November 2026 unsigned candidate contains viewer e3b4baac… in archive a0b478da…, identified exactly in the evidence download. A separate viewer-only check at 1420 × 940 and 1040 × 820 exercised row opening, exact displayed-text copying, Escape and result-name scrolling. Inline cells still have compact widths, and the full-row window preserves display rounding; use the native numeric exports for further comparisons. The earlier ad0f2983… calculation and export receipts remain dated evidence: the rebuilt archive was not given a fresh end-to-end workflow run. This does not establish general accessibility or other operating-system support.

The separate Medical histogram workflow and its recording cover different inputs and calculations. No scientific run, movie, hosted workflow or whole-product acceptance was added for this guide.

Worked examples describe particular inputs, methods and recorded results. Check current availability and input requirements before planning an evaluation.

← Back to the learning center