Home > Riviste > The Quarterly Journal of Nuclear Medicine and Molecular Imaging > Fascicoli precedenti > The Quarterly Journal of Nuclear Medicine 2002 March;46(1) > The Quarterly Journal of Nuclear Medicine 2002 March;46(1):62-9

ULTIMO FASCICOLO
 

ARTICLE TOOLS

Estratti

THE QUARTERLY JOURNAL OF NUCLEAR MEDICINE AND MOLECULAR IMAGING

Rivista di Medicina Nucleare e Imaging Molecolare


A Journal on Nuclear Medicine and Molecular Imaging
Affiliated to the Society of Radiopharmaceutical Sciences and to the International Research Group of Immunoscintigraphy
Indexed/Abstracted in: Current Contents/Clinical Medicine, EMBASE, PubMed/MEDLINE, Science Citation Index (SciSearch), Scopus
Impact Factor 2,413


eTOC

 

  TECHNOLOGIES AND METHODS IN NUCLEAR MEDICINE
Guest Editors: Todd-Pokropek A., Gilardi M. C.


The Quarterly Journal of Nuclear Medicine 2002 March;46(1):62-9

lingua: Inglese

Advances in computers and image processing with applications in nuclear medicine

Todd-Pokropek A.

From the Department of Medical Physics University College London, London, UK
INSERM U494, Paris, France


FULL TEXT  


The con­tin­u­ing advanc­es in hard­ware per­for­mance had ­made ­many pre­vi­ous­ly com­pu­ta­tion­al­ly unat­trac­tive meth­ods fea­sible, an exam­ple ­being iter­a­tive recon­struc­tion in tomog­ra­phy, ­which is now rou­tine. Dynamic ­SPECT can ­also be per­formed. However the aim of ­image pro­cess­ing is not ­just to pro­duce pret­ty pic­tures, but to ­extract ­good clin­i­cal infor­ma­tion. The meth­ods ­also ­need to incor­po­rate clin­i­cal knowl­edge and be ­defined ­using clin­i­cal con­straints. In gen­er­al ­data in nucle­ar med­i­cine are n-D, ­often 3-D ­plus ­time. Data reduc­tion for exam­ple by the extrac­tion of phys­io­log­i­cal infor­ma­tion, is impor­tant. Such ­data are in any ­case ­hard to visu­al­ise with­out com­pres­sion, for exam­ple ­some ­kind of dimen­sion­al­ity reduc­tion, ­going ­from n-D to a 2-D “func­tion­al” ­image. Both lin­e­ar and non-lin­e­ar oper­a­tions can be con­sid­ered. To ­extract phys­io­log­i­cal ­data, we ­need to fit mod­els. Two class­es of meth­od are impor­tant: ­data driv­en and hypoth­e­sis driv­en. Examples of ­data driv­en meth­ods are prin­ci­pal com­po­nent anal­y­sis and fac­tor anal­y­sis, ­where the mod­el is ­derived ­form the ­data. Hypothesis driv­en meth­ods are all impli­cit­ly or expli­cit­ly ­based on mod­el fit­ting. A pre­lim­i­nary ­data driv­en ­step fol­lowed by an hypoth­e­sis driv­en ­approach ­could be ­called con­strained sta­tis­ti­cal ­image anal­y­sis. Examples are ­shown as ­used in nucle­ar med­i­cine and are ­being extend­ed to MRI. Another impor­tant prob­lem con­sid­ered is ­that of mul­ti-modal­ity ­image reg­is­tra­tion and ­fusion. Although ­many meth­ods ­exist, all ­based on the min­i­misa­tion of an appro­pri­ate dis­tance func­tions ­between 2 ­image ­data ­sets ­such as mutu­al infor­ma­tion, addi­tion­al con­straints are ­required ­when the imag­es are not so sim­i­lar. Additional con­straints can be ­imposed by ­means of clus­ter anal­y­sis of the n-dimen­sion­al fea­ture ­space. In the anal­y­sis of ­such ­data, ­tests ­against ref­er­ence ­data ­sets (atlas­es) are ­required, nor­mal­ly requir­ing warp­ing the ­data ­sets in ­space, for exam­ple by the use of ­optic ­flow, or ­some ­kind of dif­fu­sion equa­tion. Real ­time anal­y­sis of ­data dur­ing acqui­si­tion can ­lead to optim­isa­tion of acqui­si­tion pro­ce­dures. Incorporation of ­such ­image anal­y­sis ­into a deci­sion sup­port ­system is desir­able.

inizio pagina

Publication History

Per citare questo articolo

Corresponding author e-mail