The technological evolutions of diagnostic X-ray imaging modalities enable to radiologists improve
diagnosis quality and patient care. In this context, the number of X-ray examinations like conventional
radiography, fluoroscopy or computed tomography (CT), is increasingly used in patient care. The risk
associated with the use of ionizing radiation in medical imaging is the risk of inducing cancer, a risk which is
by the Linear No-Threshold model traditionally developed for patient radiation protection. In addition, CT
imaging contributes to roughly 70 % of the total annual effective dose delivered by X-ray imaging to the
population. Because of this, many efforts have been made to decrease patient exposure to ensure that the
risk benefit balance clearly lies on the benefit side. Nevertheless, while the risk of inducing cancer cannot
be neglected, the major risk for the patient, if the justification process is respected, was the non-detection
of a pathological lesion.
The goal of this work was to propose a strategy to optimise patient exposure while maintaining diagnostic
accuracy using a task-based methodology that is pertinent in a clinical context when dealing with CT
imaging.
In this context, objective image quality should be developed and should take into account the following
four elements: (1) It should be linked to a task; (2) the properties of signals and backgrounds have to be
defined in accordance with their statistical properties; (3) the observer should be specified and (4) a figure
of merit should be precisely defined and quantified. In this sense, model observers, which are
mathematical tools potentially used as a surrogate for human observers are well suited to objectively
estimate image quality at the diagnostic accuracy level. They can indeed perform a task (e.g. lesion
detection) for a given type of image and signal (e.g. noisy uniform background) and allow a quantitative
performance estimation using for example the area under the receiver operating characteristic curve. In
addition, the advantage of model observers is that they are economical, both in terms of time and money
and they are consistent unlike the human observers.
This work shows that using a task-based approach to benchmark CT units and clinical protocols in terms of
image quality and patient exposure becomes feasible with model observers. Such an approach may be
useful for adequately and quantitatively comparing clinically relevant image quality and to estimate the
potential for further dose reductions offered by the latest technological developments.
The methodology developed during this PhD thesis enables medical physicists to convert clinically relevant
information defined by radiologists into task-based image quality criteria.