PET Imaging - Science topic

Thanks to everybody who recommended or followed the post to make it more visible!

Hello Mahyar Radak

The question you mention is very broad! As for which is so-called "better", it really depends what type of tumour you are testing for or evaluating. As we know, different types of tumours are better analysed with different types of PET scans and I know my colleagues generally think they (PET scans) are all the same but you and I know that is not true. Most commonly, they expect if someone is said to have had a PET scan, it most commonly refers to using F18-FDG and clinicians are often surprised when it turned out to be something else (!)

Take a look in our database in our new free medical imaging question and answer forum ( www.imagingQA.com ). You can filter on 'ct' and 'spect' as needed at the link below:

Feel free to sign up too, and join the growing community.

If you can't find what you're looking for, please feel free to open a new topic at the link below :

Can someone refer to any text or website that provides a step by step guide for image reconstruction in PET using list-mode data

make it more dilute,

make stock in 5% ethanol

further dilute prepare fresh working stock with 0.1% ethanol.

Please read this pdf.

Just an add-on software from the previous answer could be the Analyze by the Biomedical Imaging Resource (BIR) at Mayo Clinic.

The only disadvantage is that it accepts images where the format has to be in Analize (.img and the respective .hdr).

Plus you have to create an object map of the brain for GM and WM in order to get concentration values.

Hi,

Currently, International Working Group (IWG) and National Institute on Aging Alzheimer's Association (NIA-AA) has proposed several biomarkers as diagnostic criteria which includes cerebro spinal fluid (CSF) amyloid beta (Aβ) and tau, atrophy on MRI, glucose metabolism on [18F]-fluorodeoxyglucose positron emission tomography (FDG-PET) and fibrillar Aβ burden on amyloid PET.

There are few regions, which acts as definite Hallmark regions for AD. You could refer to below attached papers.

Structural MRI and Amyloid PET Imaging for Prediction of Conversion to Alzheimer's Disease in Patients with Mild Cognitive Impairment: A Meta-Analysis

Biomarker-based prediction of progression in MCI: Comparison of AD signature and hippocampal volume with spinal fluid amyloid-β and tau

I hope these article will help you.

Further, regarding the medication you could look into these articles

Drugs for Alzheimer’s Disease: Are They Effective?

Gud Luck !!!!

Perfect! Thanks!

There is always a resolution/noise tradeoff when using filtering. If you use larger kernels, you will reduce more noise but small structures will loose contrast. If you are trying to quantify organ level distributions, then filtering is ok and won´t change your results. On the other hand, if you are quantifying small lesions, your measurements will get lower as you increase the filter size due to partial volume effects. 

Furthermore, due to the resolution of PET you could not differentiate the binding site in a cell........

Hi

Have a look at FMRIB FSL. There is also a large amount of atlases registered on MNI template.

Good luck.

Dear Dr. Ramakrishnan  and dear Dr. Bradshaw

Thanks for your answers. Dr. Ramakrishnan is right, I am looking for a metric to find metabolic energy distribution (metabolic map) in the  tumor. So I need a tracer that is involved in metabolic process of tumor cells. But I only have Tl and Tc SPECT :(

The larger the reconstruction matrix, the lesser the so-called tissue-fraction effect will be. Indeed this phenomenon is due to the fact that each image pixel may represent the contribution of different tissue, and is more pronounced with larger pixels / voxels. The only interest in going for a smaller reconstruction matrix is to reduce pixel noise level, but this will be at the expense of increased resolution (i.e., partial volume) effects, hence  potentially leading to lesser SUV values. It all comes down to that infamous accuracy / precision trade-off.

You would need to establish how the metabolite formation is influenced by other factors.

I have worked on a tracer (11C-SA4503) which has low (but not as low as in your case) intact parent at 90min, and the metabolite does not cross BBB. The metabolite formation was influenced, at least preclinically, by various factors- age, treatment, species under consideration etc. This meant that kinetic modelling and metabolite correction was required for every study, which is not ideal as you can imagine. Metabolite correction is not exactly something you would want to do regularly in clinic, but if there are no better tracers for that particular purpose it will be used anyway. If another tracer comes along that does not require metabolite correction, but is otherwise similar, that would be preferred. 

i appreciate your considerations.

Look at the Decay Correction flag in the DICOM header (0054,1102). 

NONE = no decay correction

START= acquisition start time

ADMIN = radiopharmaceutical administration time

PET scanners in our centre (GE Discovery 710, Siemens mMR) will correct for the positron branching ratio. When you select the isotope, the overall calibration factor of the scanner is adjusted by the branching ratio. Older scanners might not have these presets, but you could verify this for yourself on your system if you scan a phantom with a known activity concentration and see what you get in the images. 

MedINRIA is great to normalize PET data on a given template. In addition, this one is free : https://med.inria.fr/

Dear Jorge,

it did these kind of reconstruction with the Philips Ingenuity PET/MR (using external attenuation images). I used a special recon protocol provided by Philips (EXAC reconstruction protocol). EXAC stands for external attenuation. This protocol allows you to use externally genererated attenuation images.

I think the best way would be to ask Philips about this recon protocol.

Georg

this is so easy to answer in hindsight with the retrospectoscope. Or even better on a desk running analysis on a population based retrospective study....however in real life where physicians must decide, in broad day light or night alike; some times under time pressure, some times biased by a recent case, sometimes afraid of medicolegal implications ....it becomes harder to only order the best exam

off course there will be overuse, how else can one be sure that all necessary examinations have been carried out... 

and yes it is frustrating that we have examinations and time and effort spend when in hindsight it could have been done without... however ...better safe than sorry and first don't harm are still relevant 

I do not think it is pretreated

The only approach that one uses with human beings is based on beta cells/islet response to insulin secretory stimuli (primarily glucose). This method has its limitations because of obvious reasons where altered secretion or function of insulin for any reason will effect the secreted insulin levels and given that there is considerable reserves of insulin, in a typical test duration, it is difficult to reliably link the insulin levels (are under the curve etc.) to a specific number of beta cells.

Even if invasive procedures procedures could be contemplated, needle biopsy is not very likely to give an overall picture due to difference in composition of islet cells between head a tail of pancreas. 

As you alluded to, in experimental animals PET has been used for estimates islet mass by loading the animals with labeled compound (dihydrotetrabenazine and fluoropropyl analogue -  DTBZ) that target vesicular monoamine transporter in beta cells. I do recall a clinical trial by Columbia U but I am not sure about the outcome. Perhaps you could search through NIH website.

You may want to give the following review a quick read:

http://diabetes.diabetesjournals.org/content/56/10/2420.full.pdf  

I had the same problem for fMRI analysis in the SPM. They are probably two problems; i.e. estimation parameters in the preprocessing stages, and standard brain masks and templates in SPM. I found useful information in the following paper for my case (fMRI). Further, I think that the last version of SPM (version 12) is more trustable than the old one for processing the brain data sets.

I wish that my information will be useful for you. 

P.S. Paper: Michal Mikla, Radek Mareček, Petr Hluštík, Martina Pavlicová, Aleš Drastich, Pavel Chlebus, Milan Brázdil, Petr Krupa: Effects of spatial smoothing on fMRI group inferences. Magnetic Resonance Imaging 26 (2008) 490–503.

PVELab was developed specifically for partial volume correction (PVC). It is free, but it needs  Matlab environment.

PMOD offers same PVC algorithms as PVELab, although minor differences may exist (for instance, PMOD offers more options for estimating the intensity value of white matter). By the way, PMOD requires licence fee.

I also tried the SFSRR package, unfortunately, I can hardly complete the whole procedure.

Try this: 

James Hampton (City university of London) has done a lot on the conceptual combination issue.

Or Geert Storms in Leuven Belgium. I think that Geert has some norms on this. It sounds familiar.

Hi Roberto,

The biggest pro of florbetapir is the longer half-life. F18 compounds are for many sites the only choice. As a consequence, F18 compounds are the standard in clinical trails. Thus. for now, PiB might still be used in most publications, but this might already be old news and is sure to change in the near future.

As mentioned, the biggest pro of PiB is that has been used the longest. It might have slightly beter sensitivty, but this is not the biggest issue when measuring patients with MCI and AD. (See http://www.ncbi.nlm.nih.gov/pubmed/23166389). Yet, when you are interested in cognitively normal older adults and working in "grey-zone" between Amyloid +/-, I would prefer PiB over florbetapir. See supplemental materials http://www.ncbi.nlm.nih.gov/pubmed/24748674

The more difficult issue is that there are many other F18 compound and some of them are better than others. These other compounds also differ in off-target binding, but I know very little about that.

Good luck - and love to hear what you are planning.

Willem

Hi Rasha, I know Heart Institute (INCOR) from Sao Paulo, Brazil, has that machine. Try to find some researchers from that institution, the most well known is Carlos Rochitte and also Cesar Nomura. They have profiles in ResearchGate. Best of luck.

A short "summary" about drug excretion, also mentioning drug weight and lipophilocity : http://www.merckmanuals.com/professional/clinical-pharmacology/pharmacokinetics/drug-excretion 

and for those who want to read more: http://classes.biology.ucsd.edu/bimm118.WI11/PPT%20Lecture%20Notes/Lecture%203.pdf

Interesting and good discussion!

Thank you. Please be aware the Eckert Ziegler has filed a Drug Master File with USFDA for their Ge68/Ga68 Generator

There are several approaches to retrieve good images. Thirst question is what do you really need/expect? You won't get any SUV from PET/CT until you made your own calibrations and very time consuming scans. To get nice images we use 25-30 min per bed position (about 18 cm with Philips Gemini TF and Time of Flight) for patients. I know that Philips was working on a calibration factor so that you can easily choose Y-90 like F-18 or Ga-68 in the acquisition protocol. I'm not sure if they succeeded or if such a calibration factor is available from GE. Maybe you've to contact GE Service. SPECT/CT is well described in several papers. One method is to use medium energy collimators and a wide energy window (20-30%) @75 keV to use Bremsstrahlung for imaging. This works fine for our Siemens Symbia T6 with 60 Frames over 360 degree with 15s per frame, continuous acquisition mode and auto contour. Reconstruction is 3DOSEM 8 iterations and 8 subsets and no filtering. It is necessary to perform scatter and attenuation correction. Maybe you have to set the energy to 75 keV for scatter and attenuation correction as well in your recon protocoll. I want to make clear that this is for routine use in patients, not for quantifying activity concentrations. There you may have to add more frames and more time per frame and play a little with reconstruction parameters. Hope this helps.  

PBR28 compound of either [18F]fluoromethyl-PBR28 or [11C]PBR28 have been used to study Parkinson's disease. A recent study described its use to image a biomarker translocator protein (TSPO): Bioconjugate Chem., 2014, 25: 442-450.

Hi Rabin,

Thanks for your really detailed response I've been doing some reading around dual process theory (Pat Croskerry has written a couple of nice articles), but haven't come across the naturalistic model, so I'll look that up thanks.

I'm working up an observational study investigating how clinical staff review images and make clinical/patient management decisions in the oncology setting.

I'll check out those references thanks and would appreciate any other thoughts you have.

Good luck with the publication

Best wishes

Mark

The ground truth is always pathology. The sensitivity and specificity of an agent is always measured against the ground truth. Brain surgery after PET imaging is often done and helps prove the utility of certain agents. Though MRI is the gold standard for brain tumors, the contrast enhanced MRI does not specifically image brain tumor cells but the result of infiltrated tissue ie. Build up of fluid 

Not 90% sure; but I think so. Take a look at: Nuclear Medicine and Biology 30 (2003) 253–260, P. Bernhardt et al

How  would  you  be  able  to  effectively  use  images  which  parameters  for  generating  them  you  might  not  know? Shouldn't  it  be  better  for you  to  generate  some  yourself  and  use  them  as  guide/standards?

Quantification in terms of Bq/ml is usually only performed by a solid state camera (rotating heads are not fast enough to pick up the dynamics involved). Using a DSPECT system with 99mTc-MIBI, peak myocardial activity concentration is on the order of 5 Bq/ml (with injection of 800 MBq), while of course the input function from the ventricle should be much higher. These numbers are patient-dependent, but are certainly on the order of Bq/ml for SPECT (only 2% of MIBI is taken up in the myocardium), even in PET studies a maximum of few hundred kBq/ml are to be expected (depending on tracer of course)

A recent paper from UCL in London explains myocardial perfusion quantification in SPECT further:

Hope it helps

Well, in recents papers (such as Glenn Wells et al. 10.2967/jnumed.114.139782) the potential of fast cardiac SPECT cameras to acquire in a dynamic way to "track" the delivery  and uptake of perfusion agents is investigated (i.e. the tracer is injected while the patient is sitting/lying in/under the camera). Based on such data, quantification of absolute myocardial blood flow similar to PET would be possible. Theoretically this allows - amongst others advantages - the detection of balanced three vessel disease as the stress flow is reduced in absolute terms (ml/min/gram) but the (static) images look almost normal. The problem, however, is that conventional SPECT tracers are much less extracted by the myocardium as typical PET agents - thus this quantification is much less reliable.

Thanks for all the good answers. We cannot rescan the patients in an MRI, we only have an anonymized database of PET+CT exams. This is true that the CT is mostly used for the attenuation but clinicians do use it for anatomical reference as well.

Dear Etienne,

as far as I know in a one compartment model, input function must be convolved with K1.exp(-k2.t) which is transfer function of the model.

how many TACs are there in a one compartment model of myocardium? I think in this model there is just one TAC which can be described by the equation C1.exp(-j1.t)+C2.exp(-j2.t)

also,the shape of input function is the way that it can be plotted using this equation. so maybe to have myocardium TAC the equation C1.exp(-j1.t)+C2.exp(-j2.t) must be convolved with the transfer function.

I would be glad if you let me know what you think about it.

M.Reza

Hi,

this might serve as a starting point:

Effect of diabetes mellitus on myocardial 18F-FDG SPECT using acipimox for the assessment of myocardial viability. Schinkel AF, Bax JJ, Valkema R, Elhendy A, van Domburg RT, Vourvouri EC, Bountioukos MA, Krenning EP, Roelandt JR, Poldermans D.

J Nucl Med. 2003 Jun;44(6):877-83.

We do a hyperglycemic clamp in order to maximally standardize the measurement; but I would not say that this is simple ;)

Best regards

Stephan

It might be possible to distinguish the contribution of the two radioisotopes to the images on the basis of their kinetics in the brain. If you have two ligands with well characterised kinetic profiles, then your tissue concentration of ligand build-up will be comprised of a sum of these two profiles. If sufficiently different, you could model this mathematically: the impulse response in the tissue would be a sum of two exponentials rather than one. Non-linear least squares algorithms are best suited for this sort of component analysis. Spectral analysis and principle component analysis of time-activity curves is already being used to look at PET ligand binding to two separate subtypes of the GABA receptor, so it is doable!

In this kind of analysis, the difference in isotope is not important, just the difference in brain behaviour. In fact, doing it with two different isotopes would make things harder: you could do the ligand comparison, but quantification would be hard given your problem with decay-correction.

There are a few ways with which you may compare similarity of images:

The stuctural similarity index (SSIN) is a good method proposed by Bovik et al. There are also other traditional metrics such the MSE, SNR, PSNR and many more.

You may find a complete analysis of all these metrics in a recent publication from our group:

C.P. Loizou, C. Theofanous, M. Pantziaris, T. Kasparis, “Despeckle filtering software toolbox for ultrasound imaging of the common carotid artery”, Comput. Meth. & Progr. Biomed., vol. 114, pp. 109-124, 2014. Best paper, editor’s choice.

There is also a software available to download with which you can perform analysis and get similarity meausres between images. You may download this from the researchgate.

If you do a post filtering or not depends on what you want. Modern PSF reconstructions will enhance your contrast and special resolution (at least they are told to) but it is known that they are more infulenced by noise (e.g. gibbs artefacts). No post filter is good for sharp images but if you use the maximal pixel value (SUVmax) it will be less reproducible than with post filtering. For quantitative measurements post filtering could help. If you do not want post filtering you could use the mean pixel value of an threshold segmentet area or the mean of a fixed area around the max pixel. These values are more stable concerning noise.

TOF reconstructions should reduce noice, but I can't quantivy that reduction.

A common setting for post filtering is a Gaussian filter with 5mm FWHM

In Siemens PET/CTs nowadays monstly the TrueX algorithm (PSF) is used with no post filtering (Allpass). If the "Iterativ" algorithm (without PSF) is used mostly a 5mm FWHM Gaussian filter is used.

GE is using VUE POINT HD or TF(meaning TOF) and you can switch the PSF on and off (SHARP IR on/off). GE uses a Gaussian post filter. I think the recommended FWHM is 6.3mm (entered in the field Cut Off). The Z-Axis filter is a Axial 1:N:1 filter with N=2 (Heavy), N=4 (Standard), N=6 (Light).

For Philips i don't know a common setting. I didn't solve the mysteries of the examination cards till now but i am looking forward that i will somewhen.

The question post filtering or not can't be answert.

The importent thing is that post filtering is changing your measurements. You should be always aware that every change in recon setting or filter setting will influence the outcome. Therefor the choice of settings should be made in accordance with the things (measurements) you want to compare to.

E.g. if you want to monitor therapy response, you must not change the settings between the two studies because you would get values which are not comparable.

greetings

Ivo

I agree that biopsy from mediastinal LN is needed. However, If the mediastinal LN are involved, I doubt that there is a benefit of treating her with curative intent. In most studies that investigated surgical resection of oligometastatic sites, mediastinal LN involvement was associated with significantly inferior prognosis, this applies to variety of carcinomas that metastasize to mediastinal LN.

The most used algorithm in emission tomography is the ML-EM (Maximum Likelihood Expectation Maximization) (Shepp and Vardi, 1982) and it's accelerated version OS-EM (Hudson and Larkin, 1994).