The effect of reconstruction algorithms (iterative versus filtered backprojection) on the diagnosis of single pulmonary nodules using Thallium-201 and Technetium-99m MIBI SPECT

Ambayi, Rudo (2004-04)

Thesis (MScMed)--Stellenbosch University, 2004.

Copy not signed by author.

Thesis

ENGLISH ABSTRACT: This study involved 33 patients, 19 men and 14 women. The age range was wide (20-90 years) and median age was 57 years. These patients had a single pulmonary nodule (SPN) defined radiologically as a well defined, round or oval intrapulmonary lung lesion not associated with atelectasis or adenopathy on chest radiography or computed tomography. Patients were investigated with Tc-99m MIBI and TI-201 (25 patients) and with Tc-99m MIBI alone (8 patients). Single photon emission computed tomography images were reconstructed using both iterative reconstruction (Ordered Subsets - Expectation Maximisation: aSEM) and filtered backprojection (FBP), on the Hermes system. Transverse, coronal and sagittal slices were displayed on the screen using a grey scale. The aSEM and FBP images for each study were co-registered semi-automatically using the multimodality programme on the Hermes. The best slice for the lesion was chosen according to the best view used to locate the SPN on chest radiograph. Regions of interest (Ral) were drawn manually outside the outer margin of the detected lesion, first on the aSEM image. This was automatically mirrored on the co-registered FBP image. For most patients, the background was automatically mirrored horizontally on the contralateral side, again, first on the OSEM then automatically on the FBP image. Automatic vertical mirroring or manual horizontal mirroring was used when background was found to be in a visually 'hot' area like the heart or vertebrae. The average counts and standard deviation of the Ral and background were generated automatically. Semi-quantitative image analysis was done by calculating the signal-to-noise ratio (SNR) and tumour-to-background (TIB) ratio using the following formulae: SNR = Mean counts ROI(lesion) - Mean counts background Standard deviation background TIB rati.o = -M---e-a-n-'--c-o--u-n-'t-s- ROI(lesion) Mean counts background Detection was found to be the same for the two reconstruction algorithms, that is, every lesion detected by using OSEM could also be detected by using FBP. However lesion detection did differ between Tl-201 and Tc-99m-MIBI. Sensitivity and specificity were calculated for different thresholds of SNR and TIB ratios. Receiver operating characteristics (ROC) curves were drawn to represent the different sensitivities and specificities at each threshold. Tuberculosis (TB) was not included in this analysis as uptake of Tl-20l was found to be significantly high and comparable to that of malignant nodules. However the effect of OSEM and FBP on the 'positive' TB nodules was assessed separately. By calculating the area under the ROC curves, TI-201 using OSEM was shown to be more accurate at differentiating malignant nodules from benign ones than FBP. Although this difference was not statistically significant (p=0.1 0), there was a clear tendency. The two reconstruction algorithms were found to be almost equally accurate, when using Tc-99m-MIBI, the difference between them being considerably insignificant. In conclusion, it was shown that there is a tendency that OSEM outperforms FBP for studies using Tl-201 but not for Tc-99m-MIBI.

AFRIKAANSE OPSOMMING: Hierdie studie sluit 33 pasiënte in, 19 mans en 14 vroue. Die ouderdomme wissel tussen 20 en 90 jaar met 'n gemiddelde ouderdom van 57 jaar. Elkeen van die pasiënte het 'n enkel longnodule (SPN) op borskas X-straal en/of rekenaar tomografie getoon, wat radiologies gedefinieer word as 'n goed omskrewe, ronde of ovaal intrapulmonale longletsel wat nie met atelektase of adenopatie geassosieer is nie. Pasiënte is met Tc-99m MIDI en TI-201 (25 pasiënte) of slegs met Tc-99m MIBI (8 pasiënte) ondersoek. Enkelfoton emissie rekenaar tomografiese (EFERT) beelde is met beide iteratiewe rekonstruksie (Ordered Subsets - Expectation Maximisation: OSEM) en gefilterde terugprojeksie (FBP) met die Hermes sisteem gerekonstrueer. Transvers, koronale en sagittale snitte is in grysskaal op die sisteem vertoon. Die OSEM en FBP beelde vir elke studie is semi-outomaties gekoregistreer met behulp van die multimodaliteitsprogram op die Hermes. Die optimale snit vir elke letsel is gekies volgens die beste aansig op die borskas X-straalom die SPN te lokaliseer. Gebiede van belang (ROl) is met die hand buite-om die buitenste rand van die letsel getrek op die OSEM beeld en daarna outomaties in die ooreenstemmende area op die gekoregistreerde FPB beeld geplaas. Vir die meeste pasiënte is die agtergrond outomaties as horisontale spieëlbeeld op die kontralaterale kant geplaas, eers op die OSEM en dan outomaties op die FBP beeld. 'n Outomatiese vertikale spieëlbeeld of manuele horisontale verskuiwing van die agtergrondsarea is gedoen indien die agtergrond oorvleuel het met 'n 'warm' area soos die hart of werwels. Die gemiddelde tellings en standaardafwyking van die ROl en agtergrond is outomaties gegenereer. Semi-kwantitatiewe beeldanalise is gedoen deur berekening van die sein-tot-agtergrond verhouding (signal-to-noise ratio - SNR) en tumor-tot-agtergrond (TIB) verhouding met behulp van die volgende formules: SNR = gemiddelde tellings ROI(letsel) - gemiddelde tellings agtergrond Standaard afwyking van agtergrond TIB rati.o = -g=em--id-d-e-l-d-e--te=ll-in-g-s__R:_O-I(-le-t-s'e-l) gemiddelde tellings agtergrond Opsporing is soortgelyk bevind vir die twee rekonstruksie algoritmes, dit wil sê elke letselopgespoor met behulp van OSEM kon ook met FBP opgespoor word. Letselwaameming het egter verskil tussen TI-201 en Tc-99m-MIBI. Sensitiwiteit en spesifisiteit is vir verskillende drempels van SNR en TIB verhoudings bereken. 'Receiver operating characteristics' (ROC) kurwes is getrek om die verskillende sensitiwiteite en spesifisiteite by elke drempel te verteenwoordig. Tuberkulose (TB) is nie in hierdie analise ingesluit nie aangesien opname van Tl-201 beduidend hoog en vergelykbaar met die van maligne nodules was. Die effek van OSEM en FBP op die 'positiewe' TB nodules is egter apart beoordeel. Deur berekening van die area onder die ROC kurwes, is getoon dat OSEM van Tl-201 tomografiese data meer akkuraat as FBP was om maligne van benigne nodules te onderskei. Alhoewel hierdie verskil nie statisties betekenisvol was nie (p=0.10), is daar wel 'n duidelike neiging gevind. Die twee rekonstruksie algoritmes was byna ewe akkuraat wanneer Tc-99m-MIBI gebruik is, met duidelik geen betekenisvolle verskil tussen die algoritmes nie. Gevo lgtrekking In hierdie studie is dit getoon dat daar 'n neiging is dat OSEM beter vaar as FBP vir studies met tallium-201 maar nie vir Tc-99m-MIBI nie.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/50144
This item appears in the following collections: