- Generated by
1.9.1
|
Medical Imaging Interaction Toolkit
2026.06.00
Medical Imaging Interaction Toolkit
|
This command-line tool converts a PET image (in [Bq/mL]) into a Standardized Uptake Value (SUV) image using the Medical Imaging Interaction Toolkit (MITK).
Features
bw - body-weight-normalized SUV (SUVbw). Output [g/mL].lbm-janma - lean-body-mass-normalized SUV (SUVlbm), Janmahasatian (2005) formula. IBSI-SUV-recommended LBM formula. Output [g/mL].lbm-james128 - lean-body-mass-normalized SUV using the James (1976) "James 128" formula. Output [g/mL].ibw - ideal-body-weight-normalized SUV using the Sugawara (1999) sex-specific formula (also IBSI-SUV-recommended). Output [g/mL].bsa - body-surface-area-normalized SUV (SUVbsa), DuBois (1916) formula. Output [cm^2/mL].(0054,1001) Units:BQML - activity concentration [Bq/mL] (standard).GML - pre-computed SUVbw [g/mL] (re-scaled to the target variant).CM2ML - pre-computed SUVbsa [cm^2/mL] (re-scaled to the target variant).CNTS + Philips SUV-scale private factor - pixel * factor yields SUV directly.CNTS + Philips activity-scale private factor - pixel * factor yields [Bq/mL].(0054,1102) Decay Correction is consulted to determine whether residual decay correction is needed (ADMIN, START, NONE).--tracer-index N.PT) is strict by default and can be bypassed explicitly. The units check is integrated into the input-units classifier; legacy --ignore-units-check forces activity-concentration semantics regardless of the tag.If the input is a DICOM PET dataset with the standard tags, the default invocation needs nothing beyond -i and -o.
| Argument | Description |
|---|---|
-i, --input | Input PET image (DICOM dir or any MITK-supported format). |
-o, --output | Output file path for the resulting SUV image (recommend .nrrd). |
| Argument | Description |
|---|---|
--variant | One of bw (default), lbm-janma, lbm-james128, ibw, bsa. Selects the output SUV normalization. See Details below. |
--variant controls the variant of the SUV image the tool produces. For activity-concentration inputs (Units=BQML, the standard case) the tool applies the variant's formula directly. For pre-normalized inputs (Units=GML, CM2ML, or Philips CNTS+SUV-scale) the tool first classifies the input variant from (0054,1001) Units and (0054,1006) SUV Type, then re-normalizes from the source variant to the target --variant. Per the IBSI-SUV spec the canonical, vendor-neutral output of any conversion is SUVbw, so the default bw is the correct choice for processing pre-normalized inputs unless you specifically need the output in a different variant.
These flags override the corresponding DICOM tags. If a tag is absent and no override is given, the run aborts with a clear error.
| Argument | DICOM tag overridden | Required for variant |
|---|---|---|
--injected-activity | (0018,1074) [Bq] | activity-concentration inputs only |
--body-weight | (0010,1030) [kg] | always |
--patient-height | (0010,1020) [m] | lbm-janma, lbm-james128, ibw, bsa |
--patient-sex | (0010,0040) (M / F / O) | lbm-janma, lbm-james128, ibw |
--half-life | (0018,1075) [s] | activity-concentration inputs only |
--nuclide | (alternative to half-life) | activity-concentration inputs only (one of 18F, 68Ga, 11C, 15O) |
--decay-time | uniform decay-time [s] | - (uniform; not a substitute for DICOM strategy NONE) |
Pre-normalized SUV inputs (Units = GML, CM2ML, or Philips CNTS + SUV-scale) bypass the activity-to-SUV math entirely and only re-normalize from the source variant to the target variant; injected dose, half-life, and decay timing are not consulted for those inputs.
| Argument | Description |
|---|---|
--ignore-modality-check | Bypass the (0008,0060) == "PT" check. |
--ignore-units-check | Force activity-concentration semantics regardless of (0054,1001). Equivalent to passing the input through the standard SUV formula as if Units = BQML. Bypasses both the input-units classifier and any pre-normalized re-scale path. |
--strict-dicom | Refuse benchmark-recommended adaptations of borderline DICOM input (see Details). Default: off. |
--tracer-index N | Explicit selection for multi-item Radiopharmaceutical Information Sequence (0054,0016) (0-based). |
| Argument | Description |
|---|---|
-v, --verbose | Verbose output. |
-h, --help | Show help text. |
A DICOM PET series with all the standard tags. Half-life, injected dose, decay timing, and patient weight are all read from DICOM. Output is [g/mL].
Same input as Example 1 but with the lean-body-mass variant (Janmahasatian, IBSI-SUV-recommended LBM formula). Patient height (0010,1020) and sex (0010,0040) are read from DICOM in addition to the body weight. Output is [g/mL] (the LBM substitutes weight in the SUV scale numerator).
The alternative lbm-james128 variant uses the older James (1976) "James 128" formula instead.
Body surface area normalization (DuBois formula). The patient-height DICOM tag is missing in this dataset, so it is supplied via the override. Output is [cm^2/mL].
The Radiopharmaceutical Information Sequence has more than one item. The tool refuses to silently pick one; the user names the desired item explicitly.
A non-DICOM input; all parameters supplied via CLI. --decay-time 0 matches the DICOM ADMIN semantics (pixel data already decay-corrected to administration time). Modality and units checks are bypassed since NRRD does not carry the corresponding DICOM tags.
| Variant | Normalization formula | Output unit |
|---|---|---|
bw | Body weight: scale numerator = W_kg * 1000. Reference: Strauss & Conti, J Nucl Med 1991. | g/mL |
lbm-janma | Janmahasatian (2005): LBM_male = (9270*W)/(6680+216*BMI), LBM_female = (9270*W)/(8780+244*BMI), then *1000. IBSI-SUV-recommended LBM formula. | g/mL |
lbm-james128 | James (1976): LBM_male = 1.10*W - 0.0128*W^2/H^2, LBM_female = 1.07*W - 0.0148*W^2/H^2 (W in kg, H in m), then *1000. | g/mL |
ibw | Sugawara (1999): IBW_male = 48.0 + 1.06*(H_cm - 152), IBW_female = 45.5 + 0.91*(H_cm - 152), then *1000. Below 152 cm the formula returns values below the constant; the tool does not clamp. Also adopted verbatim by the IBSI-SUV benchmark. | g/mL |
bsa | DuBois (1916): BSA_m2 = 0.007184 * W^0.425 * H_cm^0.725, then *10000 (cm^2). | cm^2/mL |
The math kernel is the same for all variants on activity-concentration inputs:
Pre-normalized inputs (Units = GML, CM2ML, or Philips CNTS+SUV-scale) skip the kernel and re-normalize from the source variant to the target variant:
For the sex-specific variants (lbm-janma, lbm-james128, ibw), patient sex O (Other) is handled by the IBSI-SUV-recommended adaptation: the SUV is computed as the mean of the male- and female-specific scale numerators. By default this adaptation is applied silently with a MITK_WARN. Under --strict-dicom it is refused and the tool exits with code 8 and an AmbiguousPatientSexAdaptationRefusedException message; supply --patient-sex M|F to disambiguate or relax the policy.
The tool classifies the input's pixel semantics from (0054,1001) Units and (0054,1006) SUV Type per the IBSI-SUV spec, then dispatches to the appropriate pipeline. (0054,1001) carries the pixel unit (g/mL, cm^2/mL, [Bq/mL], count); (0054,1006) carries the actual SUV variant the pixel represents.
(0054,1001) | (0054,1006) | Pipeline | Notes |
|---|---|---|---|
BQML | (ignored) | activity-to-SUV | Standard case; pixel is [Bq/mL]. |
GML | empty / absent / BW | renormalize from SUVbw | Pixel is SUVbw [g/mL]; re-scale to target variant. |
GML | LBMJANMA | renormalize from SUVlbm-janma | Pixel is SUVlbm (Janmahasatian). |
GML | LBMJAMES128 | renormalize from SUVlbm-james128 | Pixel is SUVlbm (James 128). |
GML | IBW | renormalize from SUVibw | Pixel is SUVibw. |
GML | LBM (Morgan) | refuse | IBSI-SUV calls this obsolete and not covered. Re-export the input with a specific LBM formula tag. |
GML | BSA | refuse (inconsistent) | BSA SUV uses Units = CM2ML, not GML. |
CM2ML | absent / BSA | renormalize from SUVbsa | Pixel is SUVbsa [cm^2/mL]. |
CM2ML | other | refuse (inconsistent) | CM2ML is strictly coupled to BSA. |
CNTS + Philips SUV-scale | (consulted; default BW) | renormalize from SUVbw | Manufacturer must be Philips; pixel * (7053,xx00) factor yields SUVbw, then re-scale. |
CNTS + Philips activity-scale | (ignored) | activity-to-SUV | Manufacturer must be Philips; pixel * (7053,xx09) factor yields [Bq/mL], standard math. |
CNTS without Philips factor | (any) | refuse | Both Lenient and Strict raise MissingPhilipsPETScaleException. |
CNTS non-Philips manufacturer | (any) | refuse | Both Lenient and Strict raise UnsupportedPETUnitsException. |
| Anything else / missing | (any) | refuse | The IBSI-SUV catalogue defines no fallback; both policies raise. |
The Philips private factors are lifted automatically by the PET DICOM reader and attached as named properties (mitk.pet.PhilipsSUVScale, mitk.pet.PhilipsActivityScale).
--ignore-units-check is the legacy escape hatch: it forces activity-concentration semantics regardless of the actual (0054,1001) value. Use it only if you know the pixels really are [Bq/mL].
The tool sets (0054,1001) Units and (0054,1006) SUV Type on the output to match the chosen target variant:
| Target variant | Output (0054,1001) | Output (0054,1006) |
|---|---|---|
bw | GML | BW |
lbm-janma | GML | LBM |
lbm-james128 | GML | LBM |
ibw | GML | IBW |
bsa | CM2ML | BSA |
Both DICOM-defined LBM variants share the standard SUV Type code LBM; the chosen formula is recorded externally (e.g. via filename or in a study log). (0028,1052) RescaleIntercept is set to 0.0 and (0028,1053) RescaleSlope to 1.0 since SUV is already in physical units.
The tool reads (0054,1102) Decay Correction and follows the IBSI-SUV- benchmark-recommended interpretation:
ADMIN - pixel data is already decay-corrected to the administration time. The residual decay term is 2^0 = 1 (no further correction).START - pixel data is decay-corrected to a vendor-specific reference time. See DC=START fallback chain below.NONE - pixel data is not decay-corrected. The voxel value represents the count rate averaged over the frame, which equals the instantaneous rate at AcquisitionDateTime + T_ave, with T_ave the closed-form average count-rate time T_ave = (1/lambda) * ln((lambda*T) / (1 - exp(-lambda*T))), lambda = ln(2)/T_half, T = (0018,0x1242) ActualFrameDuration. The residual decay duration is therefore (AcquisitionDateTime + T_ave) - InjectionDateTime per slice. Inputs that lack (0018,0x1242) are rejected with a MissingDICOMPropertyException rather than fall back to the biased (t_acq - t_inj) form (the bias is ~3 % for typical 5-minute frames); use --decay-time to supply timing manually.The injection time is read from (0018,1078) Radiopharmaceutical Start DateTime if available (unambiguous) or (0018,1072) Radiopharmaceutical Start Time (TM-only, with a one-shot 24 h rollover correction if the resulting decay would be negative).
The reference time used to compute the residual decay correction is chosen by the IBSI-SUV-benchmark-recommended fallback chain (the first condition whose preconditions are met determines the result):
(0071,0x22) in private block "SIEMENS MED PT".(0009,0x0D) in private block "GEMS_PETD_01". The PET reader lifts these out of the dataset at load time and attaches them as named properties (mitk.pet.SiemensDecayDateTime / mitk.pet.GEScanDateTime). When present and yielding a non-negative decay, the value is used as the uniform reference time across all slices.(0008,0032) AcquisitionTime equals (0008,0031) SeriesTime** in seconds at slice 0, and Manufacturer is one of Siemens, GE, Philips. Per-slice AcquisitionTime is then used as the reference, accommodating sub-second per-slice variation while honouring the SeriesTime tie.AcquisitionTime + T_ave - FrameReferenceTime, with T_ave from the closed-form above. The -FrameReferenceTime term undoes the scanner-applied offset from the per-frame midpoint back to the start of acquisition; T_ave additionally compensates for the average count-rate time inside the frame. Requires per-slice (0008,0032) AcquisitionTime, (0054,0x1300) FrameReferenceTime (non-negative), (0018,0x1242) ActualFrameDuration (positive), and a known half-life.Δt formula. Per-slice reference = AcquisitionTime - FrameReferenceTime. Requires the same per-slice tags as Step 3 (with the same non-negativity / positivity preconditions).If none of these applies (typically an unrecognized manufacturer with no private tag and no per-slice frame timing), the helper raises an AmbiguousDecayTimingException. The tool refuses to silently extend a vendor-specific formula to an input it cannot classify, and exits with code 3. Use --decay-time to supply timing externally.
The --decay-time flag bypasses all of the above and applies the supplied duration uniformly to every voxel. It is intended for inputs without DICOM timing tags (Example 5 above) or to reproduce ADMIN-style behaviour by passing 0. Because it is uniform-by-construction, it is not a substitute for the per-slice handling that DC=START / DC=NONE perform when reading from DICOM.
By default, the tool requires (0008,0060) Modality == "PT" (trim and case-insensitive). Bypass with --ignore-modality-check if the input is a valid PET image whose modality tag is missing or non-standard.
The units check is no longer a binary gate; (0054,1001) Units is classified into one of the supported pixel-semantics categories documented in Input pixel-unit semantics above. Use --ignore-units-check only when you want to force activity-concentration semantics regardless of the tag (the legacy behaviour: pixels are assumed [Bq/mL] and the standard SUV formula applies).
The IBSI-SUV benchmark catalogues a small number of recommendations that help MITK accept real-world PET DICOM input without losing physical meaning. Each one has a clearly bounded trigger, and by default the tool applies the recommendation and emits a MITK_WARN log entry so post-processing pipelines can audit the adaptation.
Pass --strict-dicom to refuse all such adaptations; the tool then exits with code 8 (BenchmarkAdaptationRefused) and an error message identifying the offending input. Strict mode is intended for validation, regulatory, or strict-conformance contexts where silent reinterpretation of borderline input is unacceptable.
The currently implemented adaptations:
| Adaptation | Trigger | Lenient response | Strict response |
|---|---|---|---|
Radionuclide Total Dose (0018,1074) MBq detect | DICOM-side value strictly between 0 and 1e4 | Multiply by 1e6 (interpret as MBq), emit MITK_WARN. | Exit 8 with an ImplausibleRadionuclideDoseException message. |
| DC=START vendor empirical decay-timing fallback | DC=START, no Step 1 / Step 2 match, but Step 3 / 4 preconditions met (Siemens / Philips T_ave / GE -ΔFrameRef) | Apply the IBSI-SUV recommended formula. | Exit 8 with a VendorEmpiricalDecayFallbackRefusedException; supply unambiguous timing or --decay-time. |
Sex ‘(0010,0040) == 'O’mean-of-M-and-F \ilinebr </td> <td class="markdownTableBodyNone"> sex-specific target variant (lbm-janma,lbm-james128,ibw) and patient sex resolves toOther\ilinebr </td> <td class="markdownTableBodyNone"> Compute the mean of the male- and female-specific scale numerators, emitMITK_WARN. \ilinebr </td> <td class="markdownTableBodyNone"> Exit8with anAmbiguousPatientSexAdaptationRefusedException; supply–patient-sex M|F`. |
The threshold for the dose adaptation sits in the empirically empty gap between the two physical regimes (clinical FDG doses cluster around 4e2 MBq and 4e8 Bq), so it cannot misclassify a plausible clinical input. The decay-timing fallback is empirical-not-derivable: the formulas were observed across real Siemens / Philips / GE PET data, so Strict refuses them and forces the caller to supply a spec-clean reference (vendor private datetime, AcquisitionTime equal to SeriesTime) or an explicit --decay-time.
The --injected-activity override bypasses the dose adaptation entirely (regardless of --strict-dicom); the --decay-time override likewise bypasses the decay-timing fallback entirely. Both accept the value verbatim and are treated as opt-outs from the respective heuristic.
| Code | Meaning |
|---|---|
0 | Success |
1 | Generic / unexpected error |
2 | A required DICOM property is missing |
3 | Ambiguous decay timing (cannot reconcile injection / acquisition) |
4 | Invalid CLI arguments |
5 | Multi-item Radiopharmaceutical Sequence without --tracer-index |
6 | A DICOM property holds an unsupported value |
7 | A required SUV input (e.g. height for lbm) is missing |
8 | --strict-dicom: a benchmark-recommended adaptation was refused |
9 | The input image could not be read |
10 | The output image could not be written |
The output is a 3D or 3D+t MITK-loadable image (recommend .nrrd), pixel type double, geometry identical to the input. Time-resolved DICOM properties from the input are carried over so provenance survives in the result.
The output (0054,1001) and (0054,1006) DICOM tags are set per the target variant (see Output DICOM tags above), so downstream tools see a consistent description of the SUV result.
1.9.1