Dear Helmut and other experts,



I was pondering about the same things as Bianca.



I've attached two pretty standard pattern that I find in my analyses (in both cases they're thresholded at p<0.05, FEW). One it's a (pretty standard and known) group analysis of DLB patients vs  controls in PET, the other it's a correlation study between PET methabolism and a neuropsychological score.



I'd find that reporting the cluster peak location would be extremely misleading in both cases, as the effect can be seen in very large (and specific) brain regions and not in a "point".



In which way would you describe the observed results in a paper?





Thank you,

Luca



-----Messaggio originale-----

Da: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] Per conto di MRI More

Inviato: lunedì 4 settembre 2017 16:26

A: [log in to unmask]

Oggetto: Re: [SPM] reporting cluster-corrected results



Dear Bianca,



> though I am still worried that this practice of reporting of

> cluster-peaks leads to misunderstandings on the nature of cluster correction (and leads people/readers to infer spatial specificity that is not there).



Also don't forget the smoothing. Peaks, clusters are usually labeled based on unsmoothed maximum probability maps (same for region of interest analyses), and T maps are often mapped onto unsmoothed anatomical high-res data, which implies anatomical precision that simply doesn't exist and can also be very misleading in case the geometric artefacts in the EPIs are rather large (e.g. OFC). It's an established routine though. It's also established to present thresholded data, hiding the rest.



> In the past, I have used my strategy 1 for this, but this then means

> no coordinates are reported



IMO this is bad. In case one wants to find out the exact location of the cluster one has to rely on figures then, which can be very misleading (especially surface renderings). For the gross localization labels can do it, but their use is limited as well as parcellations can differ quite a bit (e.g. how many frontal or cingulate regions). Thus one would also have to consult the corresponding atlas file first, which might not be implemented in the default software. Thus I'd suggest to report *some* coordinates, be it peak coordinates or center of mass. In that case the reader only has to find out which MNI version you are refering to and/or whether preprocessing has led to any deviations (e.g. a non-MNI Dartel template) ;)



> where I have no cluster-peaks to report



I would report the center of the mass then, which might be sufficient in most instances. In case the center came to lie outside the cluster I'd prefer a different measure though. For larger clusters the range seems to make sense, although it might be very difficult to infer the exact form of the cluster without an additional figure.



Best regards



Helmut

[BannerOSR_731X129-01]<http://www.5xmille.org/>

IL TUO 5XMILLE AL SAN RAFFAELE DI MILANO

Devolvi il tuo 5 per mille all’Ospedale San Raffaele: perché al centro della Ricerca ci sei TU. CODICE FISCALE: 07636600962, nel riquadro RICERCA SANITARIA. Non c’è cura, senza ricerca. Non c’è ricerca, senza il tuo 5xmille. Scopri come su http://www.5xmille.org





Rispetta l’ambiente: non stampare questa mail se non è necessario.

Respect the environment: if it's not necessary, don't print this mail.