Vendor-neutral software · PET · PET-CT · PET-MRI
Claritas NucMed builds enhancement, automated segmentation and quantification software for nuclear medicine. Thresholds are set by the clinician. Values are computed at voxel level. Nothing is added to the image that the acquisition did not contain.
ISO 13485 certified · Portfolio cleared by FDA, TGA, ANVISA, CDSCO and HSA · Vendor-neutral · Isotope-agnostic
Illustrative phantom for demonstration. Not patient data and not a diagnostic display. In use, iPETcertum computes these values from the loaded study at voxel level against clinician-defined minimum and maximum SUV and volume parameters.
Three cleared software devices spanning acquisition, reading and quantification in nuclear medicine. All brand-agnostic, all deployed into existing scanner and PACS environments rather than alongside them, all developed under an ISO 13485 quality system.
Enhancement & reconstruction
Noise suppression and reconstruction that recovers diagnostic quality from shorter acquisitions and lower administered activity. The original scan is preserved unaltered; only noise is suppressed and detail already captured is enhanced.
Segmentation & quantification
Segments regions of interest against clinician-defined SUV and volume parameters, and quantifies each by uptake and volume at voxel level. Runs on the original scan and on the iPET-enhanced scan, so the two can be compared directly.
Three regions returned above threshold, each colour-coded and quantified for volume, SUV and SUVmax.
Segmentation follows the clinician's SUV and volume range — physiological uptake outside those bounds is not returned.
Brain quantification
Detection, localisation and quantification across the modalities used in neurological assessment — including fused SPECT-MRI subtraction — referenced to normal population data, so a regional value carries a comparison and not only a measurement.
Stated per product and per jurisdiction, because that is the only form in which the question can be answered honestly. Clearance in one market is not clearance in another, and we do not aggregate the two.
Current as at August 2026 — full regulatory annex on request
Cleared and available for sale Under review · Jurisdictions not listed have no current application.
QUALITY SYSTEM · ISO 13485 certified — DE0507340
VALIDATION · Clinical evaluation conducted across sites in the UK, Asia, India and the Americas
DATA · HIPAA, GDPR and DPDPA compliant; residency configurable by deployment
Performance figures for scan-time and dose reduction are protocol-, scanner- and isotope-dependent. Evaluation data and the conditions under which the figures were obtained are available to qualified partners on request. Claritas does not publish headline performance claims without the acquisition conditions attached.
Two constraints govern a nuclear medicine department: how long a patient occupies the scanner, and how long a physician spends drawing round lesions. Claritas works on both, and leaves the diagnosis where it belongs.
Without
With Claritas
Scanner time is the binding constraint on a PET service, and it is capital. Shorter acquisitions raise the number of studies a scanner completes in a day without adding a scanner. Reduced administered activity lowers isotope spend on every study — a recurring cost, not a one-off.
Manual segmentation is labour, not judgement. Removing it returns the time to diagnosis and treatment decisions, and removes the step where a small or faintly avid lesion is most likely to be passed over. The reader still sets the criteria and still makes the call.
Less time on the table, which matters most to the patients least able to tolerate it, and a lower administered dose for the same diagnostic answer.
Response assessment depends on measurements being comparable between timepoints and between sites. Fixed criteria applied identically across a study remove reader drift as a source of variance in the endpoint.
5% acquisitionunprocessed
RMSE 219 · SNR 10
5% acquisitioniPET processed
RMSE 102 · SNR 49
20% acquisitionunprocessed
RMSE 109 · SNR 40
20% acquisitioniPET processed
RMSE 71 · SNR 95
Different SUV statistics behave differently under enhancement, and we report them separately rather than as a single number. SUVmin is preserved to within 1% — the noise floor falls without the baseline moving. SUVmean over a defined region is substantially preserved. SUVmax is an extreme-value statistic and is expected to move, because in a short, noisy acquisition the hottest voxel is hot partly because of noise; suppressing that noise brings the value closer to the long-acquisition reference, not further from it. Gains are protocol-, scanner- and isotope-dependent: up to 4–5× SNR improvement and 50% RMSE reduction on low-dose or reduced-time acquisitions, around 20% SNR and 10% RMSE on full-dose studies, and performance holds for high-BMI patients. Full evaluation data, with the acquisition conditions and smoothing parameters under which each figure was obtained, is available to qualified partners.
Brain quantification across the three modalities used in neurological assessment. Detection, localisation and measurement, referenced to normal population data — so a regional value arrives with a comparison attached, not just a number.
PET
Metabolic and amyloid analysis. Regional uptake against age-matched normative data, and amyloid burden on the Centiloid scale.
MRI
Volumetric analysis and lesion segmentation. Whole-brain and tissue-class volumes with percentile placement.
SPECT
Quantitative dopamine transporter analysis by striatal subregion, with normative referencing and serial comparison.
Two studies of the same patient, acquired a year apart, coregistered and compared. Every lesion is classified by anatomical class under the McDonald criteria and by what it did between scans — unchanged, smaller, larger, or new. The clinician sees the map; the report carries the arithmetic.
Unchanged Smaller Larger New
Each lesion is colour-coded by what it did between studies, in axial, coronal and sagittal reconstruction.
| Class | Baseline | Follow-up | Δ vol. |
|---|---|---|---|
| Periventricular | 21 · 3.43 | 19 · 4.64 | +35.3% |
| Deep white matter | 13 · 0.73 | 19 · 0.96 | +31.5% |
| Infratentorial | 0 · 0.00 | 1 · 0.03 | new |
| Total | 34 · 4.16 | 39 · 5.63 | +35.3% |
Counts and volume in cm³. Note the periventricular row: two fewer lesions, 35% more lesion volume. Existing lesions grew and merged. A count alone would have read as improvement.
| Metric | Volume | Pctl. | Annual rate |
|---|---|---|---|
| Grey matter | 618.53 | 45 | −1.56% |
| White matter | 540.93 | 52 | −1.33% |
| Brain parenchymal fraction | 77.71 | 12 | −0.44% |
Volume in cm³, percentile against the normal population. The annual rate is the output that matters clinically — it turns two scans into a trajectory.
FDG PET coregistered to CT, parcellated to an anatomical atlas, and compared region by region against an age-matched normative population. Every region returns a mean value, a percentage deviation, a z-score and a percentile — left and right separately, with the asymmetry computed between them.
2.0–2.5 SD 2.5–3.0 SD 3.0–6.0 SD
Regions falling below the normative range are shaded by severity, not simply flagged. The reader sees how far outside normal a region sits, and where the boundary lies.
| Region | Δ% | Z | Pctl. |
|---|---|---|---|
| Superior frontal gyrus L | −11.40 | −2.88 | 0.00 |
| Middle frontal gyrus L | −6.66 | −2.15 | 3.23 |
| Broca's area L | −7.94 | −1.79 | 3.23 |
| Superior frontal gyrus R | −6.16 | −1.71 | 4.84 |
| Anterior orbital gyrus L | −6.83 | −1.64 | 3.23 |
The top row is the finding: a left superior frontal gyrus at the 0.00 percentile — below every subject in the normative cohort. Its right-side counterpart sits at 4.84, giving a measured asymmetry rather than an impression of one.
Each region is plotted against the age-referenced normative distribution, so the comparison is visible rather than asserted.
DaT-SPECT coregistered to a normative template, with the striatum parcellated into caudate and putamen subregions. Each returns a specific binding ratio, a z-score, a percentile and an asymmetry index against its contralateral pair — the measurements a reader would otherwise estimate by eye.
Patient Normative template
Patient above, age-matched template below, with the same subregion volumes overlaid on both. Binding ratios are computed inside those regions, not from a visual impression of the image.
The same patient, fourteen months later. Progression is quantified subregion by subregion rather than described.
| Subregion | SBR L | Z L | SBR R | Z R |
|---|---|---|---|---|
| Ventral caudate | 0.81 | −3.42 | 0.81 | −2.48 |
| Anterior caudate | 0.79 | −3.23 | 0.73 | −2.89 |
| Anterior putamen | 1.11 | −2.43 | 0.81 | −3.85 |
| Posterior putamen | 0.53 | −3.18 | 0.42 | −4.31 |
The posterior putamen is the most affected subregion and the right side more than the left — the rostro-caudal gradient and asymmetry that characterise a degenerative parkinsonian syndrome, here as numbers rather than as a reader's impression. Both caudate percentiles on the left sit at 0.00.
| Subregion | Base | F/up | Δ |
|---|---|---|---|
| Anterior caudate L | 0.79 | 0.45 | −54.8% |
| Anterior putamen L | 1.11 | 0.80 | −32.5% |
| Posterior putamen R | 0.42 | 0.29 | −36.6% |
| Ventral caudate L | 0.81 | 0.61 | −28.2% |
Two studies, one year apart, on the same reconstruction pipeline and the same normative reference. Serial DaT imaging is only interpretable if the measurement is stable between timepoints — which is a software property, not a scanner one.
Ictal and interictal perfusion SPECT, subtracted from one another and coregistered to the patient's own MRI. The difference is thresholded, clustered, and mapped onto an anatomical atlas — so the candidate focus arrives with a location, a volume and a significance value rather than as a region a reader points at.
The primary focus outlined on the patient's own MRI, segmented by lobe — the anatomy a surgical team plans against.
SISCOM >2 SD Ictal Interictal
Thresholded clusters above, with the ictal and interictal studies they were derived from beneath. The subtraction is auditable against its own inputs.
Primary focus
Inferiolateral parietal R
15.51 cm³
| Region | Z | Extent |
|---|---|---|
| Inferiolateral parietal R | 5.91 | 13.15% |
| Cerebellum R | 5.76 | 11.69% |
| Middle frontal gyrus R | 5.48 | 7.49% |
| Inferiolateral parietal L | 6.00 | 4.77% |
| Postcentral gyrus L | 6.00 | 2.85% |
Significance alone would rank the left parietal and postcentral regions equal first. Extent — the proportion of the region actually occupied by the cluster — separates a large hyperperfused territory from a small intense one. Both numbers are needed, and the report gives both.
| SPECT to MRI | 0.9686 |
| MRI to reference | 0.9914 |
| Ictal / interictal correlation | 0.99 |
Subtraction is only meaningful if the two studies are aligned. The registration metrics are reported alongside the result rather than assumed.
Regional SUVR against a reference region, converted to Centiloid — the standardised scale on which 0 is a young healthy brain and 100 the typical Alzheimer's presentation. It is the unit anti-amyloid trials report in, and increasingly the unit eligibility is judged on, so the result travels between centres and between studies.
Patient Low-uptake reference High-uptake reference
The patient above, bracketed by low- and high-uptake reference templates with the same cortical regions outlined on all three. Grey–white differentiation is preserved in the top row and lost in the bottom — the visual signature the number encodes.
| Region | SUVR L | SUVR R | Z R |
|---|---|---|---|
| Prefrontal | 0.96 | 1.00 | −0.58 |
| Parietal | 1.02 | 1.15 | 0.45 |
| Precuneus | 0.90 | 0.93 | 0.77 |
| Cingulate | 0.81 | 0.82 | −0.41 |
| Composite · cerebellar ref. | SUVR 0.96 | −8.07 CL | |
Four reference regions are computed in parallel — cerebellum, pons, cerebellum plus brainstem, and cerebellar grey matter — because the choice of reference changes the answer, and a centre that uses a different one can still compare like with like.
Also available
Amyloid PET/MRI
Regions drawn on the patient's own MRI rather than a template, for cases where atrophy makes template-based parcellation unreliable. SUVR reported against the tracer-specific threshold.
neuroCloud is a quantification tool. Its output supports, and does not constitute, a diagnostic report. Cleared by ANVISA; FDA and other jurisdictions in progress — see the regulatory register above.
The software goes into the environment you already run — PET, SPECT and MRI alike. No scanner replacement, no PACS migration, no change to how studies reach the reader.
The system does not receive identifiable patient data. That removes the largest single obstacle to information-governance sign-off, and it is a design decision rather than a configuration option.
Independent of scanner OEM and model, file type, radionuclide and PACS. Installed equipment, including older systems, keeps producing reportable studies.
Deploys into existing systems or runs alongside them, whichever suits the department. Clinical workflow is not redesigned around the software.
Source studies remain intact and available for record and comparison throughout. Enhanced and original can be read side by side.
iPET and iPETcertum process inside the hospital network. neuroCloud is delivered as a hosted service, with data residency configured per deployment — so a centre without local compute still gets normative referencing and longitudinal comparison.
neuroCloud's population references and atlases are versioned and updated centrally, so every site quantifies against the same standard and results stay comparable between centres.
HIPAA · GDPR · PDPA compliant across all supported regions.
A tracer produces an image. What a sponsor actually needs is a number — comparable between sites, stable between readers, and defensible to a regulator.
That quantification layer is the part of a launch most often assembled late, per product, by a different vendor each time. We build it once, across a portfolio, so a sponsor's prostate, renal and neuro assets report on one standard rather than three.
We work in trials, in expanded access, and at commercial launch. The software is agnostic to isotope and to scanner manufacturer, which matters when a product's distribution reaches sites that have never run that tracer before.
For theranostic pairs, the same criteria applied across successive scans measure progression rather than describing it — lesion volume, SUV and SUVmax compared before and after treatment on a consistent basis.
A defined dataset, evaluated against the clinical task your label actually claims — not a generic performance benchmark.
Where the variance is: acquisition, reconstruction, reader, or something else. The residual is usually the interesting part.
Assembled to carry a submission, with the run manifest, software version and operator recorded against every result.
Integration into the sites reading your tracer, under one licence across the portfolio rather than one per product.
Most image AI asks to be trusted. Ours asks to be checked. In a field where a generative model can add a lesion that was never scanned, the discipline below is the product.
Segmentation runs against minimum and maximum SUV and volume parameters defined by the reader, not against a model's internal opinion of what counts as a lesion. The decision rule is visible and adjustable, which is what makes a result contestable — and therefore usable.
Values derive from uptake at voxel level. Two readers applying the same criteria to the same study obtain the same figures. Subjectivity is removed from the measurement without being removed from the diagnosis.
Enhancement suppresses noise and brings out detail the acquisition already captured. It does not synthesise structure that was not there. Fine lesions are retained, artefacts are not introduced, and the original is preserved alongside the enhanced image so a reader can check the claim rather than accept it. Fidelity is reported against RMSE and SNR.
Every result carries the software version, the parameters applied and the operator. A finding from eighteen months ago can be reproduced exactly, which is the difference between evidence and output.
Brand-agnostic and isotope-agnostic, integrating into existing scanner and PACS environments. A multi-site study does not need matched hardware to produce comparable numbers.
ISO 13485 certified, with clearances obtained and maintained per product and per jurisdiction rather than claimed in aggregate.
A growing number of tools improve a scan by generating a cleaner version of it. The images look excellent. The difficulty is that a model trained to produce plausible images will produce a plausible image whether or not the underlying signal supports it — and the reader has no way to tell the two cases apart.
The generative problem
How Claritas behaves
This is why the same engine can carry both enhancement and quantification. If enhancement altered the measurement, iPETcertum could not be run on an enhanced scan and the two products would have to stay apart. Because it does not, an accelerated, reduced-dose acquisition still yields a number a physician can report and a sponsor can submit.
The fastest way to evaluate this is to run it on studies you already have and compare the output against your current read. We will tell you where it does not help, which is usually the more useful half of the answer.
Request an evaluation Request the regulatory annex