Here, we summarise the unresolved debate about
p
value and its dichotomisation. We present the statement of the American Statistical Association against the misuse of statistical significance as well ...as the proposals to abandon the use of
p
value and to reduce the significance threshold from 0.05 to 0.005. We highlight reasons for a conservative approach, as clinical research needs dichotomic answers to guide decision-making, in particular in the case of diagnostic imaging and interventional radiology. With a reduced
p
value threshold, the cost of research could increase while spontaneous research could be reduced. Secondary evidence from systematic reviews/meta-analyses, data sharing, and cost-effective analyses are better ways to mitigate the false discovery rate and lack of reproducibility associated with the use of the 0.05 threshold. Importantly, when reporting
p
values, authors should always provide the actual value, not only statements of “
p
< 0.05” or “
p
≥ 0.05”, because
p
values give a measure of the degree of data compatibility with the null hypothesis. Notably, radiomics and big data, fuelled by the application of artificial intelligence, involve hundreds/thousands of tested features similarly to other “omics” such as genomics, where a reduction in the significance threshold, based on well-known corrections for multiple testing, has been already adopted.
One of the most promising areas of health innovation is the application of artificial intelligence (AI), primarily in medical imaging. This article provides basic definitions of terms such as ...“machine/deep learning” and analyses the integration of AI into radiology. Publications on AI have drastically increased from about 100–150 per year in 2007–2008 to 700–800 per year in 2016–2017. Magnetic resonance imaging and computed tomography collectively account for more than 50% of current articles. Neuroradiology appears in about one-third of the papers, followed by musculoskeletal, cardiovascular, breast, urogenital, lung/thorax, and abdomen, each representing 6–9% of articles. With an irreversible increase in the amount of data and the possibility to use AI to identify findings either detectable or not by the human eye, radiology is now moving from a subjective perceptual skill to a more objective science. Radiologists, who were on the forefront of the digital era in medicine, can guide the introduction of AI into healthcare
.
Yet, they will not be replaced because radiology includes communication of diagnosis, consideration of patient’s values and preferences, medical judgment, quality assurance, education, policy-making, and interventional procedures. The higher efficiency provided by AI will allow radiologists to perform more value-added tasks, becoming more visible to patients and playing a vital role in multidisciplinary clinical teams.
Worldwide interest in artificial intelligence (AI) applications is growing rapidly. In medicine, devices based on machine/deep learning have proliferated, especially for image analysis, presaging new ...significant challenges for the utility of AI in healthcare. This inevitably raises numerous legal and ethical questions. In this paper we analyse the state of AI regulation in the context of
medical device
development, and strategies to make AI applications safe and useful in the future. We analyse the legal framework regulating medical devices and data protection in Europe and in the United States, assessing developments that are currently taking place. The European Union (EU) is reforming these fields with new legislation (General Data Protection Regulation GDPR, Cybersecurity Directive, Medical Devices Regulation, In Vitro Diagnostic Medical Device Regulation). This reform is gradual, but it has now made its first impact, with the GDPR and the Cybersecurity Directive having taken effect in May, 2018. As regards the United States (U.S.), the regulatory scene is predominantly controlled by the Food and Drug Administration. This paper considers issues of accountability, both legal and ethical. The processes of medical device decision-making are largely unpredictable, therefore holding the creators accountable for it clearly raises concerns. There is a lot that can be done in order to regulate AI applications. If this is done properly and timely, the potentiality of AI based technology, in radiology as well as in other fields, will be invaluable.
Teaching Points
•
AI applications are medical devices supporting detection/diagnosis, work-flow, cost-effectiveness.
•
Regulations for safety, privacy protection, and ethical use of sensitive information are needed.
•
EU and U.S. have different approaches for approving and regulating new medical devices.
•
EU laws consider cyberattacks, incidents (notification and minimisation), and service continuity.
•
U.S. laws ask for opt-in data processing and use as well as for clear consumer consent.
We are currently facing extraordinary changes. A harder and harder competition in the field of science is open in each country as well as in continents and worldwide. In this context, what should we ...teach to young students and doctors? There is a need to look backward and return to "fundamentals", i.e. the deep characteristics that must characterize the research in every field, even in radiology. In this article, we focus on data integrity (including the “declarations” given by the authors who submit a manuscript), reproducibility of study results, and the peer-review process. In addition, we highlight the need of raising the level of evidence of radiological research from the estimation of diagnostic performance to that of diagnostic impact, therapeutic impact, patient outcome, and social impact. Finally, on the emerging topic of radiomics and artificial intelligence, the recommendation is to aim for cross-fertilization with data scientists, possibly involving them in the clinical departments.
Breast density is an independent risk factor for the development of breast cancer and also decreases the sensitivity of mammography for screening. Consequently, women with extremely dense breasts ...face an increased risk of late diagnosis of breast cancer. These women are, therefore, underserved with current mammographic screening programs. The results of recent studies reporting on contrast-enhanced breast MRI as a screening method in women with extremely dense breasts provide compelling evidence that this approach can enable an important reduction in breast cancer mortality for these women and is cost-effective. Because there is now a valid option to improve breast cancer screening, the European Society of Breast Imaging (EUSOBI) recommends that women should be informed about their breast density. EUSOBI thus calls on all providers of mammography screening to share density information with the women being screened. In light of the available evidence, in women aged 50 to 70 years with extremely dense breasts, the EUSOBI now recommends offering screening breast MRI every 2 to 4 years. The EUSOBI acknowledges that it may currently not be possible to offer breast MRI immediately and everywhere and underscores that quality assurance procedures need to be established, but urges radiological societies and policymakers to act on this now. Since the wishes and values of individual women differ, in screening the principles of shared decision-making should be embraced. In particular, women should be counselled on the benefits and risks of mammography and MRI-based screening, so that they are capable of making an informed choice about their preferred screening method.
Key Points
•
The recommendations in Figure 1 summarize the key points of the manuscript
•Strategies how to develop AI applications as clinical decision support systems are provided.•We focus on differences between radiomic machine learning and deep learning application domains.•Pros and ...cons, recommendations and references to software tools are provided.
Artificial intelligence (AI) models are playing an increasing role in biomedical research and healthcare services. This review focuses on challenges points to be clarified about how to develop AI applications as clinical decision support systems in the real-world context.
A narrative review has been performed including a critical assessment of articles published between 1989 and 2021 that guided challenging sections.
We first illustrate the architectural characteristics of machine learning (ML)/radiomics and deep learning (DL) approaches. For ML/radiomics, the phases of feature selection and of training, validation, and testing are described. DL models are presented as multi-layered artificial/convolutional neural networks, allowing us to directly process images. The data curation section includes technical steps such as image labelling, image annotation (with segmentation as a crucial step in radiomics), data harmonization (enabling compensation for differences in imaging protocols that typically generate noise in non-AI imaging studies) and federated learning. Thereafter, we dedicate specific sections to: sample size calculation, considering multiple testing in AI approaches; procedures for data augmentation to work with limited and unbalanced datasets; and the interpretability of AI models (the so-called black box issue). Pros and cons for choosing ML versus DL to implement AI applications to medical imaging are finally presented in a synoptic way.
Biomedicine and healthcare systems are one of the most important fields for AI applications and medical imaging is probably the most suitable and promising domain. Clarification of specific challenging points facilitates the development of such systems and their translation to clinical practice.
Objectives
To evaluate the diagnostic performance of dual-energy computed tomography (DECT) with regard to its post-processing techniques, namely linear blending (LB), iodine maps (IM), and virtual ...monoenergetic (VM) reconstructions, in diagnosing acute pulmonary embolism (PE).
Methods
This meta-analysis was conducted according to PRISMA. A systematic search on MEDLINE and EMBASE was performed in December 2019, looking for articles reporting the diagnostic performance of DECT on a per-patient level. Diagnostic performance meta-analyses were conducted grouping study parts according to DECT post-processing methods. Correlations between radiation or contrast dose and publication year were appraised.
Results
Seventeen studies entered the analysis. Only lobar and segmental acute PE were considered, subsegmental acute PE being excluded from analysis due to data heterogeneity or lack of data. LB alone was assessed in 6 study parts accounting for 348 patients, showing a pooled sensitivity of 0.87 and pooled specificity of 0.93. LB and IM together were assessed in 14 study parts accounting for 1007 patients, with a pooled sensitivity of 0.89 and pooled specificity of 0.90. LB, IM, and VM together were assessed in 2 studies (for a total 144 patients) and showed a pooled sensitivity of 0.90 and pooled specificity of 0.90. The area under the curve for LB alone, and LB together with IM was 0.93 (not available for studies using LB, IM and VM because of paucity of data). Radiation and contrast dose did not decrease with increasing year of publication.
Conclusions
Considering the published performance of single-energy CT in diagnosing acute PE, either dual-energy or single-energy computed tomography can be comparably used for the detection of acute PE.
Key Points
• Dual-energy CT displayed pooled sensitivity and specificity of 0.87 and 0.93 for linear blending alone, 0.89 and 0.90 for linear blending and iodine maps, and 0.90 and 0.90 for linear blending iodine maps, and virtual monoenergetic reconstructions.
• The performance of dual-energy CT for patient management is not superior to that reported in literature for single-energy CT (0.83 sensitivity and 0.96 specificity).
• Dual-energy CT did not yield substantial advantages in the identification of patients with acute pulmonary embolism compared to single-energy techniques.
European Radiology Experimental
reached the first 100 articles published in two years. Rejection rate was 30%, publication rate increased from 3.5/month in the first 12-month period to 4.8/month in ...the second 12-month period. The journal metrics were: 25 days from submission to first decision, 96 days from submission to acceptance, and 69 days from acceptance to publication. At the end of May 2019, we accumulated a total of 82,367 article accesses, 541 Altmetric score, and 110 citations for 92 published articles. Europe accounted for 85% of article origin. One third of corresponding authors were not radiologists/radiology residents, but were rather mainly physicists, engineers, or computer scientists. The distribution among subspecialties/body parts was well balanced; 9% of the topics regarded patient’s safety, radioprotection, or contrast media. Magnetic resonance imaging (MRI) and computed tomography (CT) accounted for 71% of the articles. Twenty-two percent of original articles/technical notes reported on animal models, 15% on phantoms, 3% on
in silico
, 2% on human cadavers, and 2% on cells. Nine articles regarded artificial intelligence and/or radiomics, and 2 regarded augmented reality. Of 100 articles, 57 declared funding sources. A total of 517 independent reviews were performed by 92 reviewers. The five articles quoted the most regarded augmented reality, spectral photon-counting CT, artificial intelligence, MRI radiomics, and diffusion tensor imaging of the musculoskeletal and peripheral nerve systems. The journal is complying with aims and scope of its “experimental” profile.