Scientific context: In Switzerland, as in most Western countries, the radiation exposure of the population is steadily increasing due to an increasing utilization of computed tomography (CT). As an example, from 2008 to 2013 the average dose per Swiss inhabitant increased from 0.8 to 1.0 mSv. Abdominal CT scans are responsible for almost a third of that exposure, with a wide variation in doses for the same clinical question. There are two main explanations for the wide dose variability for abdominal CT in the clinical practice. Firstly, large differences in regards to the technical equipment for dose reduction exists in Switzerland, particularly on the background that very effective techniques were introduced recently by the CT manufacturers for dose reduction (e.g. iterative reconstruction (IR) techniques). Secondly, it has to be assumed that numerous CT scanners are not applied in a dose-efficient approach. In other words, the protocols are insufficiently optimized. The optimization process should begin by making sure that a CT unit is as efficient as possible to convert the radiation received by the detectors into useful image information. Then, one should ensure that the level of image quality produced allows the clinical question to be answered (i.e. diagnostic accuracy) avoiding unnecessary patient exposure. In the past, standard image quality assessment used metrics based on information theory that requires linear data processes. The introduction of IR, which includes highly non-linear processes, introduces a new challenge in the field of image quality assessment. As a matter of fact, although highly noise-suppressed CT images obtained with IR may exhibit a higher contrast-to-noise ratio than conventional filtered-back projection (FBP) images, the detection of low-contrast structures/lesions might still not be possible, thereby reducing the diagnostic value of the examination. Thus, the use of new clinically-relevant metrics controlling low-contrast detectability is required.
At the moment, the use of mathematical model observers (MO) to predict low-contrast detectability of simple structures has been proposed by several groups worldwide. This approach is quite enticing since it is a task-based method that satisfies the pre-requisite of an actual optimization procedure of clinical protocols. Manufacturers have started to use these methods to benchmark their CT units, although many questions about the correct use of a CT unit in the clinical world remain unanswered. The choice of the most adequate model observer, phantom geometry, and low-contrast structures to be used in relation to specific clinical CT protocols has to be made. In summary, one needs to define a sound methodology that can be trusted by radiologists to ensure a safe process of patient dose reduction.
Aims of the project: The aims of this project are to:
Expected results: We will begin by acquiring a set of phantom images in various conditions to build an image dataset that will be evaluated by ten human observers using 2D and 3D image sets. Several mathematical model observers will be then tested aiming at matching the human observer results while trying to reduce the number of required acquisitions and noise functions that will have to be added for an acceptable match between the data. From these results the design of a second custom phantom will be proposed together with the most adequate image quality assessment methodology. The new image quality metrics obtained with the proposed methodology will be then assessed in ten major radiological centers to establish a reference set of values of the present practice in abdominal CT and initiate an optimization process at a national level.
Scientific context: In Germany, as in most Western countries, the exposure to the population due to CT examinations has drastically increased between 1996 and 2010. Since then this trend has not reversed and CT remains a major source of concern for Public Health Authorities such as the “Bundesamts für Stahlenchutz” for Germany. In 2010 the average dose per inhabitant due to CT in Germany was 1.9 mSv. A survey performed in 2008 in Switzerland confirms such a trend and like in Germany, CT now represents about 60% of the collective dose delivered from manmade sources. In order to control this situation it is important to work on several aspects. The first one deals with the justification (indication) of the examination. This is a medical matter and this will not be addressed in the framework of this project. The other important aspect to be managed is the optimization of the practice. The first step of the optimization process should ensure that a CT unit is as efficient as possible to convert the radiation received by the detectors into image information. This part deals with a signal transfer problematic. The second part of the optimization process is to ensure that the level of image quality produced ensures the diagnostic question is answered without unnecessary patient exposure. Thus to manage such an optimization process one needs not only to display such kind of dose/risk indicators to the users, but also some kind of image quality indicators.
In this context many efforts have been made to better estimate the risk part of CT examinations by introducing standardized ways to quote patient exposure (CTDIvol, DLP concepts). Then diagnostic reference levels (DRL) have been introduced to reduce the spread of the practice. However, the most important outcome of the examination, that is the clinical image quality, remains subjectively assessed. In such a context the optimization scheme between risk and benefit cannot be properly exercised. One needs a way to objectively assess the performance of the CT unit together with one or several image quality criteria that control the detection/characterization of pathologies. This requirement is particularly critical with the introduction of iterative reconstruction in CT where very low dose images can be produced without the traditional image signs (artefacts, high noise level) that alert radiologists that a low contrast lesion could be missed. To adapt the image quality level to the diagnostic question to be answered, to benchmark protocols and units one should have access not only to a dose report but also to objective image quality criteria.
Aims of the project: The project will begin by analysing the state of the practice of CT using published references and the expertise of the panel of radiologists involved in this project. Then we will propose a method that allows an objective way to characterize and compare CT units of various models. In the past, figures of merit (FOM); concepts such as the “Q-value” (introduced by ImPACT in the UK) that combines a set of image quality and dose parameters already allowed to evaluate and compare the performance of CT units. The approach was, at that time, quite efficient but it has to be revisited since CT technology has evolved significantly and previous FOMs lack in sensitivity. Moreover, one should focus on image quality assessments that are more related to a clinical context. Working with a set of defined detection tasks will certainly simplify communication with radiologists and allow better optimization strategies. We will focus in particular on the detection of small-size high contrast structures (1 – 3 mm) and “large-size” low contrast structures (5 to 8 mm – 10 to 20 HU of contrast). Our aim is characterize the performance of CT using some mathematical model observers and this set of detection tasks. The first type of model observer investigated will try to assess the maximum amount of information available when using standard filtered backprojection (FBP) image reconstructions. We will test two types of “quasi-ideal” model observers: NPW (non-prewhitening match filter) and CHO (Channelized Hotelling model observer) and try to establish a link between the ImPACT “Q-value” and the detectability indexes d’ obtained with these model observers. After having proposed a method that allows a CT benchmarking based on specifics tasks, we will focus on the evaluation of image quality parameters that are closely related to what human observers actually can detect on images that are processed not only with FBP but also with iterative approaches. We plan develop a method based on CHO proposing a set of parameters (number of channels, number of noise images required for channel training, type and level of internal noise to be used…) that allow an efficient and reliable image quality assessment. In parallel we will establish with a panel of radiologists a list of general indications with image quality requirements expressed according to the simple tasks used in our model observers. Finally we will propose a simple way to quote the image quality level of a CT acquisition (in terms of tasks used for CHO evaluations) that could be evaluated with the usual standard dose report.
Expected results: The first part of the project will allow an objective qualification of the image quality the unit is able to reach for a given exposure. These parameters could be combined with to propose an expected DEI the manufacture could use to help their customer in choosing an efficient unit. The second part of the project should provide radiologists a tool that indicates them the range of lesion they can expect to detect or not. This criterion combined with the diagnostic question to be answered is the pre-requisite for an actual patient exposure optimization. In summary we aim at developing a simple way to produce an image quality report to allow a better control of the structures that can be detected or not with the clinical setting used whether iterative reconstruction are used or not. We plan to involve two CT units per manufacturer (8 CT of at least 16 detector row) and concentrate on six protocols: Pediatric: ventricular size assessment and head routine; thoraco-abdominal: trauma and primary malignancy diagnosis; Adult: head trauma and degenerative disease; chest: pulmonary embolism and malignant tumour (staging); abdominal: pancreatic cancer and urinary tract calculi.