Analysis of the Benefit of Quantification in Myocardial Perfusion Imaging A human observer ROC evaluation using Monte Carlo Simulated Data
(2005)Medical Physics Programme
- Abstract
- "The purpose of this study was to evaluate the benefit of commercial quantification software in perfusion SPECT imaging by simulating a group of patients and their corresponding SPECT acquisitions using a computer phantom and a Monte Carlo program. Based on decisions from experienced physicians and their visual interpretation of the simulated patient images, without and with addition of the information from the quantitative perfusion software, the benefit of such a quantification program was evaluated from a ROC analysis.
Methods: A computer phantom was used to generate twenty male patients of normal size, where six patients were without any perfusion defects in the left ventricular (LV) wall and fourteen patients had perfusion defects... (More) - "The purpose of this study was to evaluate the benefit of commercial quantification software in perfusion SPECT imaging by simulating a group of patients and their corresponding SPECT acquisitions using a computer phantom and a Monte Carlo program. Based on decisions from experienced physicians and their visual interpretation of the simulated patient images, without and with addition of the information from the quantitative perfusion software, the benefit of such a quantification program was evaluated from a ROC analysis.
Methods: A computer phantom was used to generate twenty male patients of normal size, where six patients were without any perfusion defects in the left ventricular (LV) wall and fourteen patients had perfusion defects at seven different size and locations in the LV wall. The activity uptake in the defects was 75% and 85% of the normal activity in the LV wall, giving a total number of fourteen different defects. The size of the defect varied between 4% and 19 % of the LV wall volume. The activity distribution in the phantom was simulated to correspond to a total administered activity of 600 MBq 99Tcm - Sestamibi.
The phantom was connected to a scintillation camera Monte Carlo simulation program (SIMIND) and realistic SPECT projections were simulated. These projections were imported into the clinical reconstruction software (AutoSPECT+™) and evaluated using the same procedures as for a real patient study. A commercial software, AutoQuant™, was used for the reviewing, analysis and quantification of the myocardial perfusion SPECT images. Four experienced physicians interpreted the twenty computer patient images at two different reading sessions, without and with the use of the quantitative perfusion information. To evaluate the aid of the quantitative perfusion information Receiver Operating Characteristic (ROC) analysis was used. The ROC curve is a plot of sensitivity, or true positive fraction (TPF), of a procedure versus its false positive fraction (FPF = 1- specificity) for all possible cut-points. The area under the ROC curve was used as a measure of accuracy of the diagnostic procedure.
Results: The overall ROC curves showed some improvement in the area under the curve when quantitative perfusion information from the software was used. However, this improvement was relatively moderate and not significant. For the individual observers, the result from the ROC analysis varied, probably because of their individual mental thresholds to rank images.
Conclusion: The Monte Carlo method together with a realistic computer phantom and ROC analysis can be very useful to obtain information in how observers take advantage of information from quantitative perfusion software in myocardial SPECT." (Less) - Abstract (Swedish)
- Patienter med misstänkt hjärtinfarkt genomgår ofta hjärtscintigrafi för att undersöka den regionala förändringen av blodets passage genom hjärtmuskeln. Orsaken till de sjukliga förändringarna kan antingen vara ischemi (nedsatt blodflöde pga förkalkningar i blodkärl) eller infarkt (celldöd i muskulaturen). För att utföra en hjärtscintigrafiundersökning används ett radioaktivt läkemedel (99Tcm-Sestamibi) som ges intravenöst till patienten vid maximal beslastning av hjärtat samt vid vila. Hjärtscintigrafi med en scintillationskamera görs med ungefär 24 timmar mellan mätningarna. Läkaren ställer diagnos genom att jämföra scintigrafibilder i vila och arbete mot varandra med hjälp av ett datorprogram, i detta fall AutoQuant™ (ADAC Lab, Milpetas,... (More)
- Patienter med misstänkt hjärtinfarkt genomgår ofta hjärtscintigrafi för att undersöka den regionala förändringen av blodets passage genom hjärtmuskeln. Orsaken till de sjukliga förändringarna kan antingen vara ischemi (nedsatt blodflöde pga förkalkningar i blodkärl) eller infarkt (celldöd i muskulaturen). För att utföra en hjärtscintigrafiundersökning används ett radioaktivt läkemedel (99Tcm-Sestamibi) som ges intravenöst till patienten vid maximal beslastning av hjärtat samt vid vila. Hjärtscintigrafi med en scintillationskamera görs med ungefär 24 timmar mellan mätningarna. Läkaren ställer diagnos genom att jämföra scintigrafibilder i vila och arbete mot varandra med hjälp av ett datorprogram, i detta fall AutoQuant™ (ADAC Lab, Milpetas, USA). Detta program kan även jämföra mätningen mot en databas av normala hjärtan och klassificera olika regionala skillnader genom ett scoringsystem som anger utbredning och svårighetsgrad. Frågeställningen i examensarbetet var då hur stor hjälp man får av detta scoringsystem jämfört med att enbart göra en visuell bedömning utgående ifrån bilderna?
För detta används en datormodell av ett patient och ett Monte Carlo program för att simulera kamerasystem. Resultaten analyserades sedan på samma sätt som vid riktiga undersökningar av ett antal observatörer (här läkare). För att utvärdera bilderna användes programmet AutoQuant™ genom en sk ROC-analys (Receiver-Operating-Characteristic) där observatören gör bedömning av studie blint och gör graderar sannolikheten för friskt eller sjukt tillstånd. I studien simulerades 20 datorpatienter där 6 hade ett friskt hjärta och 14 patienter hade ischemiska områden fördelade på sju olika storlekar och positioner, samt med två olika nedsättningar i aktivitet. Ischemistorleken varierade mellan 4% och 19% relativt vänsterkammarväggens totala volym.
Bildutvärderingen gjordes under två tillfällen med åtta veckor mellanrum. Vid det första tillfället genomförde läkaren utvärderingen utan att använda informationen som beräknats av AutoQuant™. Vid det andra tillfället använde läkaren all information beräknade av programmet. Läkaren bedömde bilderna efter en skala från 1 till 4; där 1 = normal, 2 = sannolikt normal, 3 = sannolikt patologisk, och 4 = patologisk. Med ROC-analys erhölls parametrarna sensitivitet (sannolikheten att klassificera ett sjuk tillstånd korrekt) och specificitet (sannolikheten att klassificera ett frisk tillstånd korrekt), som användes för framtagande av ROC-kurvor (sensitivitet som funktion av 1-specificiteten). Arean under ROC-kurvan är då ett mått på hur säker diagnosen är och där ett högre värde på arean under ROC-kurvan ger bättre diagnostisk procedur för att upptäcka sjukt tillstånd.
Resultaten i examensarbetet visade på att bildutvärdering med hjälp av det beslutstöd som programmet gav något säkrare svar för några av utvärderarna men inte för alla. Detta kan bero på utvärderarens erfarenhet samt dess tilltro på den scoringinformation som programmet beräknar. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/2156956
- author
- Rouwaida, Yassin
- supervisor
- organization
- year
- 2005
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Nukleärmedicin
- language
- English
- id
- 2156956
- date added to LUP
- 2011-09-14 12:21:59
- date last changed
- 2013-09-05 10:24:38
@misc{2156956,
abstract = {{"The purpose of this study was to evaluate the benefit of commercial quantification software in perfusion SPECT imaging by simulating a group of patients and their corresponding SPECT acquisitions using a computer phantom and a Monte Carlo program. Based on decisions from experienced physicians and their visual interpretation of the simulated patient images, without and with addition of the information from the quantitative perfusion software, the benefit of such a quantification program was evaluated from a ROC analysis.
Methods: A computer phantom was used to generate twenty male patients of normal size, where six patients were without any perfusion defects in the left ventricular (LV) wall and fourteen patients had perfusion defects at seven different size and locations in the LV wall. The activity uptake in the defects was 75% and 85% of the normal activity in the LV wall, giving a total number of fourteen different defects. The size of the defect varied between 4% and 19 % of the LV wall volume. The activity distribution in the phantom was simulated to correspond to a total administered activity of 600 MBq 99Tcm - Sestamibi.
The phantom was connected to a scintillation camera Monte Carlo simulation program (SIMIND) and realistic SPECT projections were simulated. These projections were imported into the clinical reconstruction software (AutoSPECT+™) and evaluated using the same procedures as for a real patient study. A commercial software, AutoQuant™, was used for the reviewing, analysis and quantification of the myocardial perfusion SPECT images. Four experienced physicians interpreted the twenty computer patient images at two different reading sessions, without and with the use of the quantitative perfusion information. To evaluate the aid of the quantitative perfusion information Receiver Operating Characteristic (ROC) analysis was used. The ROC curve is a plot of sensitivity, or true positive fraction (TPF), of a procedure versus its false positive fraction (FPF = 1- specificity) for all possible cut-points. The area under the ROC curve was used as a measure of accuracy of the diagnostic procedure.
Results: The overall ROC curves showed some improvement in the area under the curve when quantitative perfusion information from the software was used. However, this improvement was relatively moderate and not significant. For the individual observers, the result from the ROC analysis varied, probably because of their individual mental thresholds to rank images.
Conclusion: The Monte Carlo method together with a realistic computer phantom and ROC analysis can be very useful to obtain information in how observers take advantage of information from quantitative perfusion software in myocardial SPECT."}},
author = {{Rouwaida, Yassin}},
language = {{eng}},
note = {{Student Paper}},
title = {{Analysis of the Benefit of Quantification in Myocardial Perfusion Imaging A human observer ROC evaluation using Monte Carlo Simulated Data}},
year = {{2005}},
}