Abstract
While civilisation continues to innovate remedies to human illness, it can be argued that we can never achieve a complete remedy, mortal human beings as we are. Yet, advancement we continue to make. Among these advancements, medical imaging has been and continues to be an indispensable and growing resource. There are few human diseases that have not benefited from this resource, and as medicine progresses, we continue to see new ways that imaging can provide value in disease detection and management. However, no form of medical imaging, or – for that matter – no form of medical action, is without some inherent risk. One such risk, in the case of medical imaging technologies that use ionising radiation, is the potential harm that can come from the associated radiation exposure. The question that ensues is: what are we to do with this potential harm? Avoiding medical imaging is certainly not the answer, as avoiding imaging deprives us of its enormous value. There is only one answer: optimisation. Here, this can be thought of as mitigating radiation risk within the broader context of the value that we derive from medical imaging. The necessity of optimisation is based on the essential mandate of medicine, the Hippocratic Oath of ‘primum non nocere’ (‘first, do no harm’). In practising medicine, we are obliged to minimise harm, regardless of whether we consider it actual, potential, statistical, or theoretical. Specifically, when ionising radiation is used in medicine, ICRP recommends keeping doses as low as reasonably achievable, commensurate with achieving the clinical purpose.
While necessary and certainly possible, optimisation is not trivial due to the multiplicity of concerns that need to be considered concurrently. First among them is the fact that optimisation involves a balance of harm and benefit. That is easy to say but, in itself, raises major ambiguities. Neither harm nor benefit can be known prior to an examination with 100% certainty. On the side of radiation harm, the data are often statistical in nature and cannot be ascribed deterministically to most procedures, except to higher exposures associated with some interventional procedures. Ascribing even a statistical risk figure to a common, low level of exposure is also subject to significant uncertainty, which has led some to denote this risk as theoretical (AAPM, 2022); however, such a label does not negate its relevance. On the benefit side, the value of imaging can also be labelled as theoretical, as one cannot know a priori the exact value of the imaging in the care of the patient before performing the examination. In fact, the very reason why the procedure is conducted is to learn what we do not yet know. Added to this ambiguity is the challenge of balancing a benefit with a harm – what ICRP would refer to as ‘maximising the net benefit’ – as harms and benefits are not always easily relatable, unless they can be put into a scale comparable to one another (Samei et al., 2018; Ria et al., 2022).
Optimisation in medical imaging involves quantifying the harm side of the equation. Many metrics are used, ranging from modality-specific metrics [e.g. computed tomography dose index (CTDI)] to anthropomorphised metrics (e.g. effective dose). Modality-specific metrics are more reflective of what ‘goes into the patient’, while anthropomorphised metrics are currently defined for standardised patients of defined form. No two CTDI values or effective dose values reflect equal impact on different patients of differing size, age, and sex. Recognising this, ICRP has established Task Group 128 on Individualisation and Stratification in Radiological Protection: Implications and Areas of Application. Within the broad mandate of this effort, going well beyond medical practice, there is particular emphasis on the protection of patients, an area that may benefit greatly from a more individualised approach (Samei et al., 2022).
On the other side of the optimisation leger is the benefit, often characterised in terms of image performance or quality, best defined in the context of the clinical task at hand. However, this invokes an existential challenge as we do not know the exact task involved in an imaging examination in advance, which is the reason why the examination is performed in the first place. Even if the indication is known (e.g. detecting a possible lesion), the task is not certain ahead of imaging (e.g. the size or morphology of the lesion, if present, is not known). Added to this is the multiplicity of tasks in a medical image; an imaging case is rarely used to ascertain a single task. Moreover, many incidental findings are not anticipated in advance. The benefit of imaging should also be recognised when an examination is negative, ruling out a potential issue. These realities highlight the theoretical nature of the benefit of medical imaging. However, they further necessitate the need for task-generic quantities that can ascertain the generalised representation of information content offered by an imaging case. There is no single quantity (e.g. noise) that can serve that function, but the information content can be represented through a collection of image quality features. Such features can then be integrated, through the overall framework of observer models, to capture the utility of a case to provide the desired medical information in such a way that is applicable and relatable to a range of potential tasks of relevance to the patient condition.
Provided that we sort out the ambiguities and interdependencies of quantities that guide optimisation, in practice, a large part of what is being optimised is the technology, either in its pre-clinical design or in its post-clinical use. The design obviously takes place in the developmental stage by the manufacturer. Design decisions, often unchangeable once implemented, are ideally informed by the optimisation mindset. The clinical space offers more fluidity, posing a host of choices by which the user can optimise an examination. Fundamentally, medical imaging is the use of radiative energy to harness care-relevant information for effective clinical outcome. Optimisation should seek the minimum radiative energy (in terms of x-ray flux and energy related to a relevant metric of harm) that should be used to capture the information required (task-generic metrics related to task-specific quantities of benefit). The hardware captures this ‘care-relevant information’, which, processed by the software component of the technology, aims to curate the extracted information for effective interpretation by either human or machine intelligence. This curation process, taking advantage of mechanistic and machine-intelligence methods, can improve image quality through methods such as noise reduction, enabling a reduction in radiation exposure. However, any image acquisition should still provide enough unique and original patient-specific content to start from. Here lies the essence of optimisation: a processing-aware determination of the minimum information required (constrained by the associated theoretical harm) for the targeted theoretical benefit.
Optimisation is often not the most attractive medical initiative to receive funding or resources compared with new equipment, new facilities, and such. This, in part, has led to differing optimisation practices, creating a heterogeneity that can negatively influence clinical quality and trust in radiological care. This emphasises the need for a common consensus and optimisation framework that would include defendable, verified, patient-centric quantities of potential harm and benefit; and practical and efficient means to measure and benchmark them (through in-vitro and in-vivo methods), and monitor them across practice and time. Such a solution is essential to assure quality care; an ethical mandate driven by the imperative to minimise harm to patients while maximising the benefit they gain from diagnosis and treatment using ionising radiation. Ensuring the trust of the public in the way we practice medical imaging should justify the resources needed to enable and maintain an optimised practice.
This publication provides essential guidance on optimisation of radiological protection in digital radiology, where options for image acquisition, post-processing, and presentation continue to grow, and the possibility of applying artificial intelligence methods is just around the corner. It lays the groundwork for a companion publication, developed in parallel, on practical aspects in optimisation of radiological protection in digital radiography, fluoroscopy, and computed tomography, coming soon.