Standardization and quality assurance of advanced MR-based protocols in radiology Robin Antony Birkeland Bugge A thesis presented for the degree of Master in Physics Department of Physics University of Oslo June 2014 Foreword ThisthesiswaswrittenattheInterventionCenteratOsloUniversityHospital during the time period of august 2013 and june 2014. I would like to thank my three supervisors: Atle Bjørnerud, Wibeke Nordhøy and Øystein Bech Gadmar for all their support and engagement in every aspect of the thesis. Also, thanks to Tone Elise Døli Orheim and Magne Mørk Kleppestø for assisting me with experiments and providing a lot of good insight. A big thank you to all students and employees at the biophysics research group at UiO for providing a great work and social environment throughout my masters degree. And thank you to my patient girlfriend Susanne who has been there and supported me through all my work. Oslo, Norway, May 2014 Robin Antony Birkeland Bugge 1 Abstract Standardization and quality assurance (QA) are essential both for routine magnetic resonance imaging (MRI) clinic and for research. Yet, the extent to which QA is performed varies significantly. Several scanner vendors offer to do yearly and quarterly QA, but then utilizing their own software into which the end user rarely gets insight. The Norwegian Radiation Protection Autority made recommendations in 2005 that urged all clinics and centers using MRI to have at their disposal a quality assurance methodology that is independent of the MRI vendor. The methods used in yearly quality assur- ance tests performed by physicists are often varying and manually intensive. Oneconsequenceofthisisthatresultsvarytosuchadegreethatcross-vendor comparisons becomes difficult. This is especially a challenge in larger insti- tutions with hardware from several vendors. Several of the methods used contain sources of error based on human interactions. In this master thesis the main goal has been to remedy this situation by creating an easy to use software package that reduces the amount of manual effort and makes it more feasible for trained personnel to keep track of MRI system performance over time. This software will contain a graphical user interface (GUI) for interacting with images and specifying analysis-specific options. The main focus has been on quality assessment in terms of geomet- ric distortions, signal to noise ratio (SNR), signal uniformity and temporal signal drift in functional MRI (fMRI). All code has been written in Matlab (R2012b) with the Image Processing Toolbox. A major goal of the project was to automate the currently time consum- ing task of measuring geometric distortion based on standardized phantom images. Through the use of pattern recognition methods, phantom markers could be automatically identified thereby facilitating automated analysis of distortion in a reproducible manner with equal accuracy compared to man- ual analysis. The use of Hough algorithms for detecting circular objects in the structured module of phantoms in the program, will reduce subjective measurements and human error, as well as lead to better regularity and con- sistency in QA analysis and reports. The program follows NEMA and IEC procedures and recommendations [1] [5]. In order to compare the results between manual and automatic calculations of the geometric distortion, arti- ficial MRI images with known distortion percentages were constructed. The deviation between the values calculated by the software and the actual values was calculated in over 100 images based on the root mean square for indi- 2 vidually measured points. A similar test was done to compare with human manual measurements. In this case ANOVA tables and t-tests showed no difference between manual and automated anlyses, but mostly due to a few manual anomalies that may or may not be common. The signal to noise ratio measurements were compared with previously re- ported measurements that used a different program written in IDL. There were some discrepancies in the results, which in some cases could go as high as 10 % . The reason for this is thought to be the different methods in object detection, which causes the region of interest to cover slightly different areas. For non-homogenous images this will affect the SNR. In general the QA program operates without any major issues and performs at least as well as a trained MR phycisist. It makes the QA procedures moreseamlessandsubstantiallyfaster, andremovestheneedforcumbersome manualinputbothinthemeasurementandanalysisphaseoftheprocedures. 3 Contents 1 Introduction 8 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.2 Goals of the thesis . . . . . . . . . . . . . . . . . . . . . . . . 9 2 Theory 11 2.1 MRI theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.1.1 Excitation and relaxation . . . . . . . . . . . . . . . . 12 2.1.2 Slice selection . . . . . . . . . . . . . . . . . . . . . . . 14 2.1.3 k-space and sampling . . . . . . . . . . . . . . . . . . . 14 2.1.4 RF-pulse sequences . . . . . . . . . . . . . . . . . . . . 15 Gradient Echo sequences . . . . . . . . . . . . . . . . . 15 Echo Planar Imaging sequences . . . . . . . . . . . . . 16 2.2 MRI system overview . . . . . . . . . . . . . . . . . . . . . . . 17 2.2.1 System components . . . . . . . . . . . . . . . . . . . . 18 2.3 Sources of error . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.3.1 Sources of geometric distortion and inhomogeneities . . 20 2.3.2 Sources of drift in time series . . . . . . . . . . . . . . 21 2.4 General QA methods . . . . . . . . . . . . . . . . . . . . . . . 22 2.4.1 International guidelines and standards . . . . . . . . . 22 NEMA . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 IEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.4.2 Spatial SNR . . . . . . . . . . . . . . . . . . . . . . . . 23 2.4.3 Uniformity . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.4.4 Geometric distortion . . . . . . . . . . . . . . . . . . . 25 Automated GD analysis based on CHT . . . . . . . . . 26 2.5 QA methods for functional MRI . . . . . . . . . . . . . . . . . 27 2.5.1 Temporal SNR . . . . . . . . . . . . . . . . . . . . . . 27 2.5.2 Drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Weisskoff EPI stability test . . . . . . . . . . . . . . . 29 2.6 Neuro-MRI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4 2.7 Geometric distortion requirements . . . . . . . . . . . . . . . . 31 3 Methods and Materials 32 3.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.2 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 3.3 Geometric Distortion method . . . . . . . . . . . . . . . . . . 35 3.3.1 False positive detection . . . . . . . . . . . . . . . . . . 37 3.3.2 Validation against manual expert assessment . . . . . . 37 3.3.3 Synthetic distortion generation and analysis . . . . . . 39 3.4 Drift method . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.5 Uniformity method . . . . . . . . . . . . . . . . . . . . . . . . 41 3.6 Current QA methods . . . . . . . . . . . . . . . . . . . . . . . 42 3.6.1 SNR and uniformity . . . . . . . . . . . . . . . . . . . 43 3.6.2 Manual analysis for geometric distortion determination 43 4 Results 45 4.1 Geometric distortion test . . . . . . . . . . . . . . . . . . . . . 45 4.1.1 Manual geometric distortion test . . . . . . . . . . . . 45 4.1.2 Synthetic geometric distortion test . . . . . . . . . . . 54 4.1.3 Geometric distortion image registration errors . . . . . 59 4.1.4 3 dimensional geometric distortion visualization . . . . 59 4.2 Image stability . . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.3 Analysis comparisons on existing QA data . . . . . . . . . . . 63 4.3.1 SNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.3.2 Uniformity . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.3.3 Geometric distortion . . . . . . . . . . . . . . . . . . . 66 4.4 Graphic User Interface . . . . . . . . . . . . . . . . . . . . . . 68 5 Discussion 72 5.1 Program development and resulting structure . . . . . . . . . 72 5.2 Dataset comparisons for overall program modules . . . . . . . 74 5.2.1 Geometric distortions . . . . . . . . . . . . . . . . . . . 74 5.2.2 SNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 5.2.3 Uniformity . . . . . . . . . . . . . . . . . . . . . . . . . 77 5.2.4 Time series temporal stability . . . . . . . . . . . . . . 77 5.3 Further development . . . . . . . . . . . . . . . . . . . . . . . 78 5.3.1 IEC compliant scaling . . . . . . . . . . . . . . . . . . 78 5.3.2 Weisskoff power spectrum analysis . . . . . . . . . . . 78 6 Conclusion 79 5 7 Matlab Code 80 7.1 Geometric distortion . . . . . . . . . . . . . . . . . . . . . . . 80 7.1.1 FindCircles.m . . . . . . . . . . . . . . . . . . . . . . . 80 7.1.2 Geodist reader.m . . . . . . . . . . . . . . . . . . . . . 82 7.1.3 geometric2D.m . . . . . . . . . . . . . . . . . . . . . . 82 7.1.4 info printer.m . . . . . . . . . . . . . . . . . . . . . . . 88 7.1.5 phillips positions manual.m . . . . . . . . . . . . . . . 88 7.1.6 phillips small neuronal positions manual.m . . . . . . . 89 7.1.7 Plotter.m . . . . . . . . . . . . . . . . . . . . . . . . . 90 7.1.8 TagPrinter.m . . . . . . . . . . . . . . . . . . . . . . . 91 7.2 Time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 7.2.1 Analyzer.m . . . . . . . . . . . . . . . . . . . . . . . . 94 7.2.2 Drift.m . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 7.2.3 Drift printer.m . . . . . . . . . . . . . . . . . . . . . . 96 7.2.4 Drift reader.m . . . . . . . . . . . . . . . . . . . . . . . 97 7.2.5 Drift ROI images.m . . . . . . . . . . . . . . . . . . . . 97 7.2.6 SNR.m . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 7.3 Uniformity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 7.3.1 Homogen.m . . . . . . . . . . . . . . . . . . . . . . . . 100 7.3.2 Homogen reader.m . . . . . . . . . . . . . . . . . . . . 108 7.3.3 Infoprinter homogen . . . . . . . . . . . . . . . . . . . 109 7.3.4 Tagprinter homogen.m . . . . . . . . . . . . . . . . . . 110 7.4 Artificial MR image generator . . . . . . . . . . . . . . . . . . 115 7.4.1 Artificer.m . . . . . . . . . . . . . . . . . . . . . . . . . 115 7.4.2 Artificer drift.m . . . . . . . . . . . . . . . . . . . . . . 117 7.4.3 Artificer repeat.m . . . . . . . . . . . . . . . . . . . . . 118 7.4.4 Easy reader.m . . . . . . . . . . . . . . . . . . . . . . . 121 7.4.5 philips artificer.m . . . . . . . . . . . . . . . . . . . . . 121 7.4.6 warp artificer.m . . . . . . . . . . . . . . . . . . . . . . 122 7.4.7 warp artificer rand.m . . . . . . . . . . . . . . . . . . . 122 6 Abbreviations and acronyms BOQA Book of Methods in MR QA CHT Circular Hough Transform EPI Echo planar imaging fMRI Functional magnetic resonance imaging GD Geometric distortion MRI Magnetic resonance imaging OUS Oslo Universitetssykehus (Oslo University Hospital) QA Quality assurance QAP Quality assurance program, developed for this thesis ROI Region of interest SD Standard deviation SE Spin echo SNR Signal to noise ratio TE Echo time TR Repetition time 7 Chapter 1 Introduction 1.1 Background The physicists in the clinic carry out different tasks related to their field of expertise, some of which are routine based performance services that relate the need for calibration of various systems. In medical technology these tasks are referred to as quality assurance, QA for short, and is an integrated part of the clinical work environment. The Norwegian Radiation Protection Authority(NRPA)havesince2005requiredallfacilitiesutilizingclinicalMRI to have an available qualified phycisist and to have an independent system for quality assurance [12]. This thesis has focused on establishing automated quality assurance routines that will function as auxilliary and independant methods within a subset of the regular QA-routines. A standard QA procedure often includes mea- surement of SNR, signal uniformity, contrast resolution, ghosting level and geometric distortion, which is part of the routine tasks for the MR physicists, whereas the aim of this thesis was to develop and implement robust and au- tomatic software for some of the most important QA methods in MRI. These are, in addition to basic image quality methods such as spatial SNR and sig- nal uniformity, geometric distortion and temporal signal stability methods. The field of diagnostic radiology, and especially MRI, is one that requires good QA routines that are carried out on a regular basis. Currently, at the OUS (Oslo University Hospital) the MR physicists are carrying out a standardized QA procedure once a year on each and every MRI system in the hospitals in the South-Eastern Health Region of Norway. In addition, the 8 various manufacturers perform their own QA routines every 3 to 6 months. Automated procedures are generally preferred if sufficiently accurate since they eliminate user bias and potentially save time. One example is geometric distortion, where manual measurements are subject to user bias, which is dependent on gauging the positions and lengths in a geometric structure. There is also time saved from drastically reducing the number of steps that a human must intervene in the process. As already mentioned, most suppliers of MRI equipment provide QA at reg- ular intervals as part of their services that differ in content and frequency amongst vendors. Despite of this it is of great importance for both research and the clinical work to have independent methods for testing. This is not only useful as a means of self-assurance, but is essential for any center that needs to ensure that their MR machine is stable and produce high quality images, and for early discovery of various errors that are not yet grown large enough to become visually identifiable in the images. There are too many things that can go wrong with a MRI system to give an extensive overview of various sources of error in this thesis. That is everything from malfunction- ing gradient coils to a patient dropping a metallic or magnetic coin in the machine interior. These are situations that can occur in the time between QA runnings and needs to be rectified, which are possible to detect using the appropriate on-site QA method. 1.2 Goals of the thesis The goals of this thesis is to: Reduce user bias in QA assessment of multi- vendor MRI data through the development of a validated semi-automated QA software tool. In order to achieve this the following specifications and part-goals were uti- lized: (cid:136) InvestigatingcurrentmethodsforperformingqualityassuranceinMRI. (cid:136) Creating new methods or altering existing methods to better suit the needs of this project. (cid:136) Incorporate these methods in a Matlab based semi-automatic QA pro- gramthatwillfunctionasasupplementtothecurrentqualityassurance procedures. 9
Description: