Xinqidi,Your biological research partner
Xinqidi Biological Technology Co.,Ltd,Wuhan,China 2008-2020
R&D 12th year

Translating research findings into clinical practice: a systematic and critical review of neuroimagi

Issuing time:2020-04-21 08:49

Abstract

A pivotal aim of psychiatric and neurological research is to promote the translation of the findings into clinical practice to improve diagnostic and prognostic assessment of individual patients. Structural neuroimaging holds much promise, with neuroanatomical measures accounting for up to 40% of the variance in clinical outcome. Building on these findings, a number of imaging-based clinical tools have been developed to make diagnostic and prognostic inferences about individual patients from their structural Magnetic Resonance Imaging scans. This systematic review describes and compares the technical characteristics of the available tools, with the aim to assess their translational potential into real-world clinical settings. The results reveal that a total of eight tools. All of these were specifically developed for neurological disorders, and as such are not suitable for application to psychiatric disorders. Furthermore, most of the tools were trained and validated in a single dataset, which can result in poor generalizability, or using a small number of individuals, which can cause overoptimistic results. In addition, all of the tools rely on two strategies to detect brain abnormalities in single individuals, one based on univariate comparison, and the other based on multivariate machine-learning algorithms. We discuss current barriers to the adoption of these tools in clinical practice and propose a checklist of pivotal characteristics that should be included in an “ideal” neuroimaging-based clinical tool for brain disorders.

Introduction

Brain-based disorders, including psychiatric and neurological illnesses, represent 10.4% of the global burden of disease1, and their prevalence within the general population is thought to be increasing2. While the past few decades have seen significant progress in our biological understanding of these disorders, this has had little or no impact on real-world clinical practice3,4,5. This is especially the case in clinical psychiatry, where diagnostic and prognostic assessment is still based on self-reports and clinical ratings, which are associated with low inter-rater agreement and accuracy6. It is recognized that patients suffering from psychiatric and neurological illnesses could benefit from the translation of the research findings into clinical practice. The key question for researchers and clinicians is how to enable this7,8.

Over the past two decades, scientists have invested many resources in the use of brain-imaging to develop objective tests for detecting brain disorders, monitoring their progression over time and optimizing treatment. This has led to several promising findings. For example, in the field of psychiatry, structural neuroimaging has revealed widespread neuroanatomical alterations, including both transdiagnostic and disorder-specific effects9,10,11. Additionally, neuroanatomical measures have been found to account for up to 40% of the variance in clinical outcome, and can even explain some of this variance where clinical variables (e.g. diagnosis) fail to do so12,13,14. These findings have led to the suggestion that structural neuroimaging could be used to develop objective measures of psychiatric disease, in contrast with current nosological criteria which are susceptible to subjective bias7.

However, so far we have not been able to translate the plethora of promising findings into clinically useful imaging-based tests5,15,16,17. One of the main reasons for the current gap between research and clinical practice, is that the former has been dominated by analytical methods that only allow statistical inferences at group-level (e.g. how does the brain differ between a group of people with psychosis and a group of healthy controls?); whilst a clinician has to make diagnostic and treatment decisions at the level of the individual. In recent years, a growing number of studies have attempted to address this issue by using alternative analytical methods that allow statistical inferences at the level of the single case. A large proportion of these studies have been employing machine-learning methods to make inferences at the levels of the individual based on structural18,19 or functional20,21 neuroimaging data. This has resulted in a number of encouraging findings22,23. For example, machine-learning methods appear to be effective in differentiating between patients with brain illness and healthy controls, and in predicting the onset of illness and response to treatment12,22,23. Although this is still an emerging area of research, there is compelling evidence that neuroimaging data allow for more accurate diagnostic and prognostic inferences compared to the use of clinical and psychometric data alone12.

Following these encouraging findings, some research teams have been developing imaging-based tools for making inferences at the level of the individual24,25,26,27. Through these tools, clinicians can upload the brain images of individual patients and receive an automatic report of the brain abnormalities detected. These tools differ greatly with respect to their specific purpose (e.g. what disease is being targeted), their technical characteristics (e.g. what is the underlying statistical model), their robustness (e.g. how the tool was validated) and their availability (e.g. freely vs. commercially available). At present there is no single resource which presents all available tools and systematically compares their aims and characteristics; this means that it can be difficult for a clinician or a researcher to identify the most appropriate tool. In addition, in the absence of a systematic review of their strengths and limitations, the real translational potential of the existing tools is still unclear. To address this gap, we conducted a systematic review of available neuroimaging-based clinical tools for making inferences at single-subject level. Our first aim was to describe and compare how these tools have been developed and validated, with the ultimate goal of assessing their translational potential in real-world clinical settings. Our second aim was to use the findings to develop a checklist of the pivotal characteristics that should be included in an ideal imaging-based clinical tool for brain disorders. We hope that this review will help clinicians and researchers appreciate the aims, strengths, and limitations of the available tools and select the most appropriate option for their investigations.

Materials and methods

Studies selection

As the results of the current review might have health-related implications, the protocol of this review has been registered to the International Prospective Register of Systematic Reviews (PROSPERO—Registration Number: CRD42019127819). In accordance with the PRISMA guidelines28,29, an in-depth search was conducted on PubMed and Google Scholar databases up to February 2019. The following terms were used: (brain AND (MRI OR neuroimaging OR “magnetic resonance”) AND (“clinical tool”) AND (psychiatric OR psychiatry OR neurological OR neurology OR disease OR disorder)). All papers describing a neuroimaging-based tool developed to detect brain abnormalities in brain disorders at the level of the individual, regardless of the diagnosis, were included. Furthermore, additional relevant studies were found using different strategies. These included using the “related articles” function of the PubMed database; tracing the references from the identified papers; tracing the key references on the tool websites; and emailing the providers of the clinical tools.

Inclusion and exclusion criteria

The following inclusion criteria were used: (i) articles presenting a neuroimaging-based clinical tool; (ii) articles presenting a tool aimed at detecting abnormalities in the brain (i.e. studies presenting a tool for detecting abnormalities in other organs, for instance the heart, were excluded); (iii) articles presenting a validation of the algorithm or technology that underlie the tool (i.e. studies applying an already validated clinical tool were excluded); (iv) articles published as original articles in peer-reviewed academic journals or conference proceedings (posters from conferences were excluded); (v) articles published or available in English.

Articles were excluded from the review according to the following a priori exclusion criteria: (i) articles that present software for analyzing neuroimaging data without a clear implementation in a translational tool (e.g. Statistical Parametric Mapping18,19); (ii) articles reporting studies that use non-human subjects; and (iii) studies that present clinical tools that are yet to be released.

According to our first exclusion criterion, we excluded platforms which allow the storage and analysis of individual MRI scans, using software such as Freesurfer30, Sienax31, or FSL32, but do not provide a clinically meaningful report including an estimate of neuroanatomical abnormalities at the level of the individual. One example is QMENTA (https://www.qmenta.com/), a cloud-based platform where different neuroimaging modalities (i.e. structural MRI, functional MRI, diffusion tensor imaging, positron emission tomography) can be stored and a different of different statistical analyses can be carried out. For instance, using QMENTA, researchers can investigate gray matter (GM) volume, cortical thickness, structural and functional connectivity, and ventricular volumetry, just to name a few of the multiple analyses which can be implemented via this platform. The advantage of using a platform such as QMENTA is the possibility to run multiple analyses simultaneously on a cloud thereby saving time. However, QMENTA does not provide researchers and clinicians with individualized reports indicating whether or not the brain under investigation deviates from those of healthy controls and what specific alternations might be driving this conclusion.

According to the same exclusion criterion, we also excluded ASSESSA PML (https://ixico.com/technology/data-platforms/assessa-platform/), a platform allowing neurologists to transfer clinical and neuroimaging data to expert neuroradiologists, who will visually inspect the scans to detect the presence of progressive multifocal leukoencephalopathy (PML), an opportunistic infection of the brain emerging as an adverse event of pharmacotherapy to treat multiple sclerosis (MS)33. ASSESSA PML was excluded from the current review as it is not a clinical tool that automatically extracts clinically relevant information from neuroimaging data.

Data extraction

Two authors (C.S. and M.J.H.) extracted and checked the data independently. An additional member of the team double-checked the data in case of discordance between the first two extractions. An independent researcher oversaw the entire search procedure and randomly selected some of the articles for a random double-check. In this process, no critical issues were detected by the independent researcher. A database was created including the following characteristics: general information (authors, year of publication, name of the tool, website) and technical details regarding the tool (type of images analyzed, type of analysis performed, number of subjects


Article classification: Biological abstract
Share to:
Tel:+86-027-87610298
Tel:+86-027-87610297
Add:Room A11-329, 1st Floor, No.1, SBI Venture Street, Optics Valley, East Lake
New Technology Development Zone, Wuhan, China.
Certificate NO.:U18Q28010569R0S