A B S T R A C TLiver MR is a well-established modality with multiparametric capabilities. However, to take advantage of its full capacity, it is mandatory to master the technique and optimize imaging protocols, apply advanced imaging concepts and understand the use of different contrast media. Physiologic artefacts although inherent to upper abdominal studies can be minimized using triggering techniques and new strategies for motion control. For standardization, the liver MR protocol should include motion-resistant T2-w sequences, in-op phase GRE T1 and T2-w fast spin echo sequences with fat suppression. Diffusion-weighted imaging (DWI) is mandatory, especially for detection of sub-centimetre metastases. Contrast-enhanced MR is the cornerstone of liver MR, especially for lesion characterization. Although extracellular agents are the most extensively used contrast agents, hepatobiliary contrast media can provide an extra-layer of functional diagnostic information adding to the diagnostic value of liver MR. The use of high field strength (3T) increases SNR but is more challenging especially concerning artefact control. Quantitative MR belongs to the new and evolving field of radiomics where the use of emerging biomarkers such as perfusion or DWI can derive new information regarding disease detection, prognostication and evaluation of tumour response. This information can overcome some of the limitations of current tests, especially when using vascular disruptive agents for oncologic treatment assessment. MR is, today, a robust, mature, multiparametric imaging modality where clinical applications have greatly expanded from morphology to advanced imaging. This new concept should be acknowledged by all those involved in producing high quality, high-end liver MR studies.

Note: This copy is for your personal, non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at www.rsna.org/rsnarights.

Existing quantitative imaging biomarkers (QIBs) are associated with known biological tissue characteristics and follow a well-understood path of technical, biological and clinical validation before incorporation into clinical trials. In radiomics, novel data-driven processes extract numerous visually imperceptible statistical features from the imaging data with no a priori assumptions on their correlation with biological processes. The selection of relevant features (radiomic signature) and incorporation into clinical trials therefore requires additional considerations to ensure meaningful imaging endpoints. Also, the number of radiomic features tested means that power calculations would result in sample sizes impossible to achieve within clinical trials. This article examines how the process of standardising and validating data-driven imaging biomarkers differs from those based on biological associations. Radiomic signatures are best developed initially on datasets that represent diversity of acquisition protocols as well as diversity of disease and of normal findings, rather than within clinical trials with standardised and optimised protocols as this would risk the selection of radiomic features being linked to the imaging process rather than the pathology. Normalisation through discretisation and feature harmonisation are essential pre-processing steps. Biological correlation may be performed after the technical and clinical validity of a radiomic signature is established, but is not mandatory. Feature selection may be part of discovery within a radiomics-specific trial or represent exploratory endpoints within an established trial; a previously validated radiomic signature may even be used as a primary/secondary endpoint, particularly if associations are demonstrated with specific biological processes and pathways being targeted within clinical trials. Key Points • Data-driven processes like radiomics risk false discoveries due to high-dimensionality of the dataset compared to sample size, making adequate diversity of the data, cross-validation and external validation essential to mitigate the risks of spurious associations and overfitting. • Use of radiomic signatures within clinical trials requires multistep standardisation of image acquisition, image analysis and data mining processes. • Biological correlation may be established after clinical validation but is not mandatory.

Purpose: This work aims to develop a computer-aided diagnosis (CAD) to quantify the extent of pulmonary involvement (PI) in COVID-19 as well as the radiological patterns referred to as lung opacities in chest computer tomography (CT).Methods: One hundred thirty subjects with COVID-19 pneumonia who underwent chest CT at hospital admission were retrospectively studied (141 sets of CT scan images). Eighty-eight healthy individuals without radiological evidence of acute lung disease served as controls. Two radiologists selected up to four regions of interest (ROI) per patient (totaling 1,475 ROIs) visually regarded as well-aerated regions (472), ground-glass opacity (GGO, 413), crazy paving and linear opacities (CP/LO, 340), and consolidation (250). After balancing with 250 ROIs for each class, the density quantiles (2.5, 25, 50, 75, and 97.5%) of 1,000 ROIs were used to train (700), validate (150), and test (150 ROIs) an artificial neural network (ANN) classifier (60 neurons in a single-hidden-layer architecture). Pulmonary involvement was defined as the sum of GGO, CP/LO, and consolidation volumes divided by total lung volume (TLV), and the cutoff of normality between controls and COVID-19 patients was determined with a receiver operator characteristic (ROC) curve. The severity of pulmonary involvement in COVID-19 patients was also assessed by calculating Z scores relative to the average volume of parenchymal opacities in controls. Thus, COVID-19 cases were classified as mild (<cutoff of normality), moderate (cutoff of normality ≤ pulmonary involvement < Z score 3), and severe pulmonary involvement (Z score ≥3).Results: Cohen's kappa agreement between CAD and radiologist classification was 81% (79–84%, 95% CI). The ROC curve of PI by the ANN presented a threshold of 21.5%, sensitivity of 0.80, specificity of 0.86, AUC of 0.90, accuracy of 0.82, F score of 0.85, and 0.65 Matthews' correlation coefficient. Accordingly, 77 patients were classified as having severe pulmonary involvement reaching 55 ± 13% of the TLV (Z score related to controls ≥3) and presented significantly higher lung weight, serum C-reactive protein concentration, proportion of hospitalization in intensive care units, instances of mechanical ventilation, and case fatality.Conclusion: The proposed CAD aided in detecting and quantifying the extent of pulmonary involvement, helping to phenotype patients with COVID-19 pneumonia.