Machine learning in cardiovascular radiology: ESCR position statement on design requirements, quality assessment, current applications, opportunities, and challenges

Abstract

Machine learning offers great opportunities to streamline and improve clinical care from the perspective of cardiac imagers, patients, and the industry and is a very active scientific research field. In light of these advances, the European Society of Cardiovascular Radiology (ESCR), a non-profit medical society dedicated to advancing cardiovascular radiology, has assembled a position statement regarding the use of machine learning (ML) in cardiovascular imaging. The purpose of this statement is to provide guidance on requirements for successful development and implementation of ML applications in cardiovascular imaging. In particular, recommendations on how to adequately design ML studies and how to report and interpret their results are provided. Finally, we identify opportunities and challenges ahead. While the focus of this position statement is ML development in cardiovascular imaging, most considerations are relevant to ML in radiology in general. Key Points • Development and clinical implementation of machine learning in cardiovascular imaging is a multidisciplinary pursuit. • Based on existing study quality standard frameworks such as SPIRIT and STARD, we propose a list of quality criteria for ML studies in radiology. • The cardiovascular imaging research community should strive for the compilation of multicenter datasets for the development, evaluation, and benchmarking of ML algorithms.

Guidelines and quality criteria for artificial intelligence-based prediction models in healthcare: a scoping review

“…One of the very first aspects of developing and validating an AIPM as recommended in literature is to clearly specify the medical problem and context that the AIPM will address, and to identify the healthcare setting(s) in which the AIPM is to be deployed 3,[6][7][8][9][10][11][12][13][14][15] . Before starting actual AIPM development, it is advocated to first conduct a thorough investigation into the current standard of care, context and workflow [7][8][9][10][11][14][15][16][17][18] , and to provide a clear rationale for why the current approach falls short. For example, via analysis of the needs of targeted end users through observations and interviews, and by involving them from the start in the developmental process 11,12,[17][18][19][20] .…”

“…For example, via analysis of the needs of targeted end users through observations and interviews, and by involving them from the start in the developmental process 11,12,[17][18][19][20] . Once a precise (diagnostic or prognostic) prediction task has been formulated, healthcare actions, treatments or interventions should be defined that are to follow from the AIPM predictions 3,[6][7][8]10,11,13,17,21 . Clinical success criteria must be determined and described 3,6,7,9,11,12,20,22 , including an analysis of the potential risks of prediction errors 6,23 .…”

The Ichikado CT score as a prognostic tool for coronavirus disease 2019 pneumonia: a retrospective cohort study

“…Regardless of the CT scoring system analyzed, all prior studies have identified a CT score as an independent predictor of adverse outcome in patients with COVID-19 pneumonia [ 10 , 15 , 17 , 18 ]. Moreover, CT scores had significant associations with various inflammatory biomarkers known to be predictors of mortality [ 14 , 16 ]. However, to our knowledge, no other study has followed the Ichikado scoring method, which we believe might be superior to previously described methods due to its reproducibility and having both a quantitative and qualitative approach.…”

“…While some authors and radiological societies recommend against routine use of CT scan in patients with COVID-19, our analysis seems to suggest a benefit by allowing early recognition of patients at high risk of decompensation [ 28 , 29 ]. Additionally, the high diagnostic sensitivity may prove to be an asset as CT scans may show characteristic findings of the disease even when RT-PCR is negative (false-negative), or in hospitals with a relatively long turn-around time for RT-PCR results [ 16 , 30 ].…”

Automatic coronary artery calcium scoring from unenhanced-ECG-gated CT using deep learning

“…Several recent studies report different experiences in automating the procedure of segmentation and quantification of the CAC burden, using different deep learning approaches. Indeed, in the last few years semantic segmentation architectures were used to predict dense segmentation maps by extending Convolutional Neural Networks (CNN) to Fully Convolutional Neural Networks (FCNN) [4]. They were first applied to 2D biomedical imagery with the so-called U-Net [5] and later with a straightforward extension to 3D and 4D V-Net [6].…”