Reading PET/CT in the browser

PET/CT is the study that exposes a weak viewer fastest.

A plain CT can be read on almost anything that scrolls. A PET/CT needs two very different image sets shown together, numbers that depend on information typed in at reception, and measurements that change depending on how you draw a line around a lesion. Get any of those wrong and the report is wrong, sometimes without anyone noticing. That is why most centres still read PET/CT on a dedicated vendor workstation in one room, even when everything else has moved to a web viewer.

So, let's look at what a PET/CT reading actually needs, how SUV is calculated and why it goes wrong, how threshold-based segmentation works, and why we think the browser is now a reasonable place to do all of it.

What reading a PET/CT needs

A reader moves between three views of the same patient.

1. The CT, for anatomy, at normal windows.
2. The PET, usually in an inverted grey scale or a colour map, plus a rotating maximum intensity projection (MIP) of the whole body to spot uptake at a glance.
3. The fusion, with PET colour laid over the CT, so a hot spot can be placed on a specific node, bowel loop or bone.

All three have to scroll together, in all planes. Clicking a point on the MIP should take you to that point in the axial, coronal and sagittal views. The colour scale has to be controllable. For response assessment, the EANM guideline recommends viewing studies on the same fixed colour scale, for example SUV 0 to 10, so a baseline and a follow-up are visually comparable.

Then comes the part that makes PET different: numbers.

SUV, in plain terms

The standardised uptake value (SUV) compares how much tracer a region took up with what you would expect if the injected dose had spread evenly through the body. For the common body-weight version:

SUV = tissue activity concentration (kBq/mL) ÷ [injected activity (MBq) ÷ body weight (kg)]

The injected activity and the image activity must be decay-corrected to the same reference time, which in DICOM PET may be the start of acquisition or the time of administration; the viewer has to read which one the scanner used. An SUV of 1 means "average". A lesion with an SUV of 8 has taken up roughly eight times the average. The EANM guideline also recommends a lean-body-mass version, called SUL, alongside SUV. (EANM procedure guidelines for tumour imaging, version 2.0)

A worked example helps. Say a 70 kg patient has 370 MBq of FDG on board at scan time (after decay correction), and a node measures 40 kBq/mL. The expected average is 370 ÷ 70, about 5.3 kBq per gram of tissue, which for soft tissue is roughly 5.3 kBq/mL. The node's SUV is 40 ÷ 5.3, about 7.6. Now suppose reception entered the weight as 60 kg instead of 70. The same node becomes 40 ÷ (370 ÷ 60), about 6.5. Nothing in the image changed; one typed number moved the SUV by more than one unit.

A few variants show up in reports:

• SUVmax: the single hottest voxel in the lesion.
• SUVmean: the average inside a region you have drawn.
• SUVpeak: the average inside a small sphere, about 1 mL (1.2 cm across) in the EANM definition, placed where that average is highest. Less noisy than SUVmax.

The formula is simple; the inputs are where trouble starts. A viewer calculates SUV from values stored in the DICOM header: patient weight, injected dose, injection time, the radionuclide's half-life and the scan time. If any of those is wrong, the SUV is wrong, even though the images look perfect. In Indian centres, the usual culprits are ordinary:

1. Weight. Entered at reception from memory, or carried over from a previous visit. The EANM guideline says weight must be measured directly before each scan, because it changes during treatment.
2. Injection time. Written on paper and typed in later, or the clock on the dose calibrator does not match the scanner's clock. FDG's fluorine-18 has a half-life of about 110 minutes, so a ten-minute error in timing shifts the SUV by roughly six per cent.
3. Uptake time. EANM recommends about 60 minutes (acceptable range 55 to 75), and for repeat studies in the same patient, the same interval within 10 minutes. A busy scanner room drifts.
4. Blood glucose and patient preparation. High glucose lowers tumour uptake. A methodology review in the European Journal of Nuclear Medicine and Molecular Imaging also describes how reconstruction settings, motion and scanner calibration change SUV.

None of these are viewer problems as such. But a viewer that shows the weight, dose and timings it used, next to the SUV, makes them easy to catch.

Segmenting a lesion with a threshold

SUVmax describes one voxel. For many purposes you want the whole lesion: its volume, and how much metabolically active tissue it contains. That needs a 3D outline, and two measures depend on it:

• Metabolic tumour volume (MTV): the volume of the outlined region.
• Total lesion glycolysis (TLG): SUVmean inside that region multiplied by the MTV.

The EANM guideline notes that both are of growing interest for response monitoring and prognosis, and should be reported if available.

Drawing a 3D outline by hand on every slice is slow and inconsistent. So most tools use a threshold: every connected voxel above a cut-off belongs to the lesion.

Threshold-based 3D lesion segmentation on fused PET/CT, step by step.

There are two broad ways to set that cut-off, and each has a clear advantage and a clear weakness.

Fixed SUV threshold. Everything above a set SUV value is included; SUV 2.5 is a value often used in published work. It is simple to explain and simple to repeat. In a 2018 study of lymphoma patients, the SUV 2.5 method had the best agreement between observers and was the easiest to apply. The weakness is that it inherits every SUV error above. If the weight or timing is off, the volume changes. A lesion with low uptake may fall below the line entirely, and a hot lesion next to the bladder or the heart can merge with normal uptake.

Percentage of SUVmax. Everything above a fraction of the lesion's own SUVmax is included. The EANM guideline suggests 41% and 50% of the maximum as standard options, and recommends reporting MTV and TLG from both when possible. Because the cut-off moves with each lesion, it adapts to very hot and fairly mild lesions alike, and is less affected by calibration. But it depends on a single voxel, SUVmax, which is one of the noisiest numbers in the study. And EANM points out that in a lesion with uneven uptake, a relative threshold may outline only the most active part.

So which one is right? There is no consensus. A review of MTV methods makes exactly that point, and the lymphoma study above found that cut-offs separating good from poor prognosis varied markedly depending on the method. The practical rules are simpler than the debate. Pick a method for each indication, write it in the report, and use the same method on every follow-up of the same patient. Comparing a 41% volume at baseline with a fixed-threshold volume at follow-up is meaningless.

Once the outline exists, it should be stored as a standard object rather than as a screenshot. Herman Oosterwijk's deep dive into DICOM SEG explains how a segmentation is encoded so another viewer can open it, and our own post on DICOM SEG and PET/CT lesion segmentation applies that to PET/CT.

Why the browser

Until recently, all of this lived on a dedicated workstation in the reporting room. That made sense when the computing was heavy and the network was slow. It makes less sense now.

PET/CT readers in India are spread thin. A nuclear medicine physician may cover several centres. A radiologist reads the CT part from home. An oncologist wants to see the fused images before a tumour board. A workstation tied to one desk does not fit that pattern.

There are honest trade-offs. A whole-body PET/CT is a large study, so the first load depends on bandwidth (our guide to cloud PACS or on-premise has the arithmetic), and very old machines will struggle with 3D work. But for most readers, being able to open the same study, with the same tools, from wherever they are is worth more than shaving seconds off a local workstation.

What this means for a centre

A few practical steps follow from all of this.

1. Fix the inputs first. Weigh patients on the day. Record injection time from a clock that matches the scanner. Note blood glucose.
2. Check SUV against your scanner's workstation. Before trusting any new viewer, compare SUVmax on a set of known studies. They should agree closely.
3. Choose and document a segmentation method per indication, and keep it constant across follow-ups.
4. Put the method in the report. "MTV 41%" or "MTV, SUV 2.5 threshold" tells the next reader what they are comparing against.

The practical upshot is that a PET/CT viewer is only as trustworthy as the numbers typed in at reception and the segmentation method written in the report. Get those two right, and where the viewer runs becomes a question of convenience rather than accuracy.

Related reading

• DICOM SEG and PET/CT lesion segmentation — how a lesion outline is stored and shared.
• What is a zero-footprint DICOM viewer — client-side rendering and its limits.

FAQ

Card image: "Whole body PET/CT scan" by Myo Han, via Wikimedia Commons, licensed CC BY 3.0. It is a general PET/CT viewer screenshot, not a SOMATIQ screen.