AUTOMATICHOT SPOTDETECTIONAND SEGMENTATIONINWHOLE BODY FDG-PETIMAGES
Lists
Only you see this
Abstract
We present asystem forautomatic hotspots detection and segmentation inwholebodyFDG-PETimages. Themain contribution ofoursystem isthreefold. First, ithasanovel body-section labeling module based onspatial Hidden-Markov Models(HMM);this allows different processing policies to beapplied indifferent bodysections. Second, theCompetition Diffusion (CD)segmentation algorithm, whichtakes into account body-section information, converts thebinary thresholding results toprobabilistic interpretation anddetects hotspotregion candidates. Third, arecursive intensity modeseeking algorithm finds hotspot centers efficiently, andgiven these centers, aclinically meaningful protocol isproposed to accurately quantify hotspotvolumes. Experimental results showthat oursystem worksrobustly despite thelarge variations inclinical PETimages. IndexTerms-Imagesegmentation, Object detection Positron emission tomography (PET)using fluorine-18 deoxyglucose (FDG)isanuclear medicine medical imaging technique that produces athree dimensional imageoffunctional processes inthebody.TumorsinFDG-PETappear ashot spots duetoincreased FDG uptake, andbylate 1990s, a large bodyofliterature haveclearly shownthat FDG-PET imaging isessential andeffective fordetection, diagnosis and prognosis oftumors. Itisalso shownapplicable toneurological disorders andcoronary artery disease [1]. Itisexpected that thenumberofFDG-PETimages performed inmostfacilities will exceed all other procedures performed withradioactive compound inthecoming years [1]. AstheamountofPETdata generated every dayincreases, the burden ontheradiologists inreading anddiagnosing thecases increases. Thismayresult inanegative effect, astherateof overlooking anomalies canincrease duetofatigue, andthe risk ofover-diagnosis mayalso rise. Computer-aided detection (CAD)tools haveconstantly demonstrated effectiveness insupplementing thetaskofradiologists; theyimprove the sensitivity ofanomaly detection andreduce theinterpretation variation amongradiologists. Currently, toourknowledge, there arenocommercially available CADtools forFDG-PET *Thefirst author performed theworkwhile atSIEMENS. images. Inthis paper, wepresent asystem design foraPETCADtool that detects suspicious regions wheretheamountof FDGuptake values areabnormally high. Previous works onPETimagestudies havebeenpresented inbothclinical study andmedical imageprocessing [2,3]. Challenges remain, however, inbothclinical study andimage processing. Fromaclinical perspective, themainchallenge inabnormal hotspotdetection isthat normal tissues also uptake different levels ofFDG[2]. Forexample, FDGiscleared primarily through therenal system. Thus, thenormal bladder isoften thehighest FDGuptake organ. Brain hashighFDG uptake. Heartandbowelhavevariant levels ofFDGuptake. Liver, bonemarrowandspleen sometimes also havemildly highFDGuptake whichishigher thansometumors. Considering all these normal tissue uptakes, theintensity range ofthe normal tissue overlaps withthetumors significantly. Inaddition, theshape, size, andlocation oftumors inwhole-body PETimages areunpredictable.
BibTeX
@inproceedings{guan2006automatichot,
author = {Haiying Guan and Toshiro Kubota and Xiaolei Huang and Matthew Turk},
title = {{AUTOMATICHOT SPOTDETECTIONAND SEGMENTATIONINWHOLE BODY FDG-PETIMAGES}},
booktitle = {ICIP},
year = {2006},
url = {https://openreview.net/forum?id=tNsGTYWwfSnL},
}Loading related content...