Source-guided kinetic workflow. Medical Image Processing Kinetic Modeling, medical_, separates the inputs needed to attempt a route from the evidence needed to interpret its result. This lesson reviews current controls and export contracts against retained GUI regression evidence. It adds no new fit, independent parameter comparison or recording.
1. Keep curves, timing and acquisition evidence together
A time–activity curve (TAC) is a measured activity series with an explicit time coordinate. Keep the target tissue, input and reference curves distinct. A loaded dynamic image does not automatically supply a target region, a calibrated arterial input or the scientific assumptions of a reference-tissue model.
| Input | What the current workflow needs |
|---|---|
| Target | A measured target TAC, or dynamic PET pixels with an explicitly selected target ROI. Preserve activity units, frame timing and source identity. |
| Arterial input | A measured plasma curve or a separately validated and calibrated image-derived input function (IDIF). An input mask alone produces an explicitly uncalibrated IDIF; it does not establish the corrections or scale needed by arterial models. |
| Reference tissue | A measured reference TAC or corresponding mask, with the reference-tissue assumptions reviewed. The control names lack of specific binding and compatible nondisplaceable kinetics; selecting a region does not demonstrate those assumptions. |
| Blood and metabolites | Preserve whole-blood and plasma samples separately. A vascular-fraction model needs measured whole blood. Applying parent-fraction correction needs measured parent-fraction samples; a checkbox does not create them. |
Use LOAD CURVE / BLOOD TABLE for the intended measured table, and PREPARE KINETIC STUDY to review image-derived roles. The TARGET ROI MASK, INPUT / IDIF MASK, REFERENCE ROI MASK and PARAMETRIC MAP MASK controls identify different uses. Review masks on the acquisition’s physical grid and keep the preparation receipt. Parametric support is not automatically a target or reference-tissue definition.
2. Preserve time origins and frame boundaries
The PET-BIDS specification defines timing relative to TimeZero, which need not equal injection time. The app’s kinetic models require the relevant injection-relative timing explicitly. Keep the image and blood-file associations, unit declarations, decay reference and transformations into that time basis. Do not silently treat scan start, series reference and injection as the same event.
The kinetic core represents time in minutes. In the interface, t* (min) uses minutes, while Input delay (s; + shifts later) uses seconds. Preserve that distinction when recording settings. The core’s physically calibrated activity curves are normalized to Bq/mL; retain the input contract’s unit and physical-unit flags with the exported values.
Frame duration alone does not prove that a reported frame time is a midpoint. Current source derives start/end boundaries from time and duration only when midpoint semantics are explicitly declared. Otherwise, supplied durations remain available for weighting, while those boundaries are not invented. Start and end vectors must be supplied together, ordered and non-overlapping. Missing boundaries affect whether a model predicts frame averages or samples at representative times.
3. Check the selected route’s requirements
Open Kinetic Modeling in Nuclear Medicine. Set the scientific controls, then use CHECK ROUTE READINESS before selecting the needed analyses. The current default regime is unspecified, weighting is duration, and automatic t* and parametric-map creation are initially off. These defaults are interface state, not a recommended scientific protocol.
| Route family | Distinct requirements in current source |
|---|---|
| 1-TCM / 2-TCM | Target, measured plasma or validated IDIF, and injection-referenced time. The 2-TCM route additionally requires an explicit reversible or irreversible regime. |
| Patlak / Ki | Arterial input, an irreversible regime, and explicit t* or automatic t* with QC. A late straight segment alone does not supply the declared kinetic regime. |
| Logan / Vt | Arterial input, a reversible regime, and explicit t* or automatic t* with QC. These requirements differ from the Patlak selection. |
| Full Ref / SRTM / SRTM Basis | Target and measured reference tissue, injection-referenced timing and reviewed reference-tissue assumptions. Arterial-model prerequisites cannot substitute for a suitable reference region. |
Weighting has its own evidence requirements. Duration weighting requires measured durations. Inverse-variance weighting requires explicit external per-frame observation variance. Spatial variation among voxels inside an ROI is retained as QC; it is not automatically the variance of the measured TAC. A positive fixed vascular fraction or a fitted vascular fraction additionally needs measured whole blood.
Creating Patlak, Logan or SRTM-basis maps also requires dynamic PET pixels, an explicit parametric-support mask and exact source-content binding. A successful regional fit does not establish every voxel’s fit, and an ROI overlay is not a parameter map.
4. Read readiness and scientific qualification separately
The readiness table has five columns: Route, Preflight status, Required contract, Current blocker(s), and Scientific interpretation qualification. eligible_ means that the presence checks found no listed blocker; compute-time QC still applies. The same row can retain critical qualification unresolved.
Review tracer identity, correction evidence, ROI visual review and other study-level qualifications alongside the model prerequisites. A source flag named scientific_ describes those recorded checks; it is not independent scientific validation. Preserve missing or conflicting evidence and the actual run/route statuses instead of turning eligibility into a claim of accurate parameters.
The retained selected GUI receipt has four passing cases: two launcher/timing cases, one app/viewer case and one preparation-workspace case. Its FXS2 loader case retained a lazy 30-frame image with the malformed unit token BQ_, missing injection-time normalization and no measured target TAC. That loading result was not a valid kinetic fit. The current launcher differs from the receipt’s launcher; seven kinetic/support files and the test source still match. No new current-launcher or numerical acceptance is inferred.
5. Keep the bundle’s curves, contracts and results together
The Export route writes what is available at that point in the workflow. Inspect the manifest and run context rather than expecting a fixed file count. Exporting preserves results and their limitations; it does not validate them.
| Saved object | Contents and limits |
|---|---|
| Measured curves | measured_ contains frame time in minutes and target activity, with reference activity or target variance when present. Plasma, whole blood, parent fraction and HPLC recovery have separate measured tables when available. |
| Input contract | input_ retains frame times, starts/ends/durations, activity units, input kind, time origin, warnings, selected model settings, preparation and study qualification. The measured-curves CSV alone is not the complete input contract. |
| Results and tables | kinetic_ retains serialized route results. tables.json retains complete structured tables. Available figures are PNGs; ROI masks and parameter maps have separate arrays and context. The manifest records actual file sizes and SHA-256 hashes. |
| In-run receipt | The current run receipt is published after an in-run export completes. Its absence at that point is stated by receipt_; it is not a completed current-run receipt. The Export summary is added after the bundle, so do not expect that new summary inside the bundle it describes. |
The dedicated viewer labels a parameter map as a single 3-D volume. Its direct volume export uses the inherited paired NPY/JSON path and writes float32 values; keep the metadata with the array. ROI overlays are not embedded in that direct array export. This is distinct from the results bundle, which saves available structured tables, masks and map metadata.
Export metadata retains computed_ and publishable_. This lesson checks those source contracts; it does not claim a newly generated bundle, every exported numerical cell, a reopened analysis session or a new parameter reference comparison.