Halligan, SM;
(2015)
Towards greater clarity for the analysis of imaging studies: Development & validation of an alternative to the area under the receiver-operator characteristic curve.
Doctoral thesis , UCL (University College London).
Preview |
Text
HALLIGANphdFINAL.pdf Download (13MB) | Preview |
Abstract
This thesis arose from a 2006 study performed by the author and his collaborators that attempted to gain regulatory approval for computer-assisted detection (CAD) software. The USA Food & Drug Administration (FDA) obliged us to use the change in the area under the receiver-operator characteristic curve (ROC AUC) as our primary outcome. Despite its wide dissemination in radiology research, we found implementation of ROC AUC very problematic. This thesis explores the hurdles we encountered and argues for an alternative approach. Chapter 1 describes the rationale for and against ROC AUC as a measure of diagnostic performance. An alternative analysis based on net benefit is proposed on the basis that it is more transparent and simpler to interpret. Chapter 2 uses the net benefit method to analyse a multi-reader multi-case (MRMC) study of CAD for CT colonography. The analysis requires an estimate of relative misclassification costs for false-negative versus false-positive diagnoses; “W”. This study used a conservative value for W, arrived at via consensus. In Chapter 3 an evidence-based value for W in the context of screening for colorectal cancer and polyps by CT colonography is arrived at via a discrete choice experiment (DCE) of patients and healthcare workers. Chapter 4 uses the value for W obtained in Chapter 3 in a net benefit analysis to compare observer performance in two MRMC studies of CAD for CT colonography. Chapter 5 obtains W by DCE for a different clinical context – detection of extracolonic pathology by CT colonography. Chapter 6 describes a systematic review that aims to determine whether reporting of MRMC ROC AUC methods in the radiological literature is comprehensive. Chapter 7 then provides guidelines for the comprehensive reporting of MRMC ROC AUC studies. The thesis finishes with a summary of the work performed and suggestions for further research.
Type: | Thesis (Doctoral) |
---|---|
Title: | Towards greater clarity for the analysis of imaging studies: Development & validation of an alternative to the area under the receiver-operator characteristic curve. |
Event: | UCL |
Open access status: | An open access version is available from UCL Discovery |
Language: | English |
Keywords: | colonography, receiver operator curve, ROC AUC, computer assisted detection, CAD |
UCL classification: | UCL UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Medical Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Medical Sciences > Div of Medicine UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Medical Sciences > Div of Medicine > Department of Imaging |
URI: | https://discovery.ucl.ac.uk/id/eprint/1472422 |
Archive Staff Only
View Item |