Search

Federated learning is a machine learning method that allows decentralized training of deep neural networks among multiple clients while preserving the privacy of each client's data. Federated learning is instrumental in medical imaging because of the privacy considerations of medical data. Setting up federated networks in hospitals comes with unique challenges, primarily because medical imaging data and federated learning algorithms each have their own set of distinct characteristics. This article introduces federated learning algorithms in medical imaging and discusses technical challenges and considerations of real-world implementation of them., (Copyright © 2022 American College of Radiology. Published by Elsevier Inc. All rights reserved.)

With recent developments in medical imaging facilities, extensive medical imaging data are produced every day. This increasing amount of data provides an opportunity for researchers to develop data-driven methods and deliver better health care. However, data-driven models require a large amount of data to be adequately trained. Furthermore, there is always a limited amount of data available in each data center. Hence, deep learning models trained on local data centers might not reach their total performance capacity. One solution could be to accumulate all data from different centers into one center. However, data privacy regulations do not allow medical institutions to easily combine their data, and this becomes increasingly difficult when institutions from multiple countries are involved. Another solution is to use privacy-preserving algorithms, which can make use of all the data available in multiple centers while keeping the sensitive data private. Federated learning (FL) is such a mechanism that enables deploying large-scale machine learning models trained on different data centers without sharing sensitive data. In FL, instead of transferring data, a general model is trained on local data sets and transferred between data centers. FL has been identified as a promising field of research, with extensive possible uses in medical research and practice. This article introduces FL, with a comprehensive look into its concepts and recent research trends in medical imaging., (Copyright © 2022 American College of Radiology. Published by Elsevier Inc. All rights reserved.)

Objectives: The objective of this study was to quantitatively assess the image quality of Advanced Modeled Iterative Reconstruction (ADMIRE) and the PixelShine (PS) deep learning algorithm for the optimization of low-dose computed tomography protocols in midfacial trauma., Study Design: Six fresh frozen human cadaver head specimens were scanned by computed tomography using both standard and low-dose scan protocols. Three iterative reconstruction strengths were applied to reconstruct bone and soft tissue data sets and these were subsequently applied to the PS algorithm. Signal-to-noise ratios (SNRs) and contrast-to-noise ratios (CNRs) were calculated for each data set by using the image noise measurements of 10 consecutive image slices from a standardized region of interest template., Results: The low-dose scan protocol resulted in a 61.7% decrease in the radiation dose. Radiation dose reduction significantly reduced, and iterative reconstruction and the deep learning algorithm significantly improved, the CNR for bone and soft tissue data sets. The algorithms improved image quality after substantial dose reduction. The greatest improvement in SNRs and CNRs was found using the iterative reconstruction algorithm., Conclusion: Both the ADMIRE and PS algorithms significantly improved image quality after substantial radiation dose reduction., (Copyright © 2020 The Authors. Published by Elsevier Inc. All rights reserved.)

Background: To determine the effect of low-dose, high-pitch non-electrocardiographic (ECG)-triggered chest CT on coronary artery calcium (CAC) detection, quantification and risk stratification, compared to ECG-triggered cardiac CT., Methods: We selected 1,000 participants from the ImaLife study, 50% with coronary calcification on cardiac CT. All participants underwent non-contrast cardiac CT followed by chest CT using third-generation dual-source technology. Reconstruction settings were equal for both acquisitions. CAC scores were determined by Agatston's method, and divided dichotomously (0, >0), and into risk categories (0, 1-99, 100-399, ≥400). We investigated the influence of heart rate and body mass index (BMI) on risk reclassification., Results: Positive CAC scores on cardiac CT ranged from 1 to 6926 (median 39). Compared to cardiac CT, chest CT had sensitivity of 0.96 (95%CI 0.94-0.98) and specificity of 0.99 (95%CI 0.97-0.99) for CAC detection (κ = 0.95). In participants with coronary calcification on cardiac CT, CAC score on chest CT was lower than on cardiac CT (median 30 versus 40, p˂0.001). Agreement in CAC-based risk strata was excellent (weighted κ = 0.95). Sixty-five cases (6.5%) were reclassified by one risk category in chest CT, with fifty-five (84.6%) shifting downward. Higher BMI resulted in higher reclassification rate (13% for BMI ≥30 versus 5.2% for BMI <30, p = 0.001), but there was no effect of heart rate., Conclusion: Low-dose, high-pitch chest CT, using third-generation dual-source technology shows almost perfect agreement with cardiac CT in CAC detection and risk stratification. However, low-dose chest CT mainly underestimates the CAC score as compared to cardiac CT, and results in inaccurate risk categorization in BMI ≥30., Competing Interests: Declaration of competing interest The PhD project of the first author is part of the ImaLife project, which is funded by an institutional research grant from Siemens Healthineers and by the Ministry of Economic Affairs and Climate (Netherlands) Policy by means of the PPP Allowance made available by the Top Sector Life Sciences & Health to stimulate public-private partnerships. Matthijs Oudkerk is involved in the company iDNA BV. There are no other conflicts of interest to disclose., (Copyright © 2020 Society of Cardiovascular Computed Tomography. Published by Elsevier Inc. All rights reserved.)

Background: To determine the impact of high-pitch spiral acquisition on radiation dose and cardiovascular disease (CVD) risk stratification by coronary artery calcium (CAC) assessment with computed tomography in individuals with a high heart rate., Methods: Of the ROBINSCA trial, 1990 participants with regular rhythm and heart rates >65 beats per minute (bpm) were included. As reference, 390 participants with regular heart rates ≤65 bpm were used. All participants underwent prospectively electrocardiographically(ECG)-triggered imaging of the coronary arteries using dual source CT at 120 kVp, 80 ref mAs using both high-pitch spiral mode and sequential mode. Radiation dose, Agatston score, number of positive scores, as well as median absolute difference of the Agatston score were determined and participants were stratified into CVD risk categories., Results: A similar percentage of participants with low heart rates and high heart rates had a positive CAC score in data sets acquired in high-pitch spiral (low heart rate: 57.7%, high heart rate: 55.8%) and sequential mode (58.0%, 54.7%, p = n.s.). The median absolute difference in Agatston scores between acquisition modes was 14.2% and 9.2%, for the high and low heart rate groups, respectively. Excellent agreement for risk categorization between the two data acquisition modes was found for the high (κ = 0.927) and low (κ = 0.946) heart rate groups. Radiation dose was 48% lower for high-pitch spiral versus sequential acquisitions., Conclusion: Radiation dose for the quantification of coronary calcium can be reduced by 48% when using the high-pitch spiral acquisition mode compared to the sequential mode in participants with a regular high heart rate. CVD risk stratification agreement between the two modes of data acquisition is excellent., (Copyright © 2018 Society of Cardiovascular Computed Tomography. Published by Elsevier Inc. All rights reserved.)

Aim: To assess the association of coronary artery geometry with the severity of coronary artery disease (CAD)., Methods: 73 asymptomatic individuals at increased risk of CAD due to peripheral vascular disease (18 women, mean age 63.5 ± 8.2 years) underwent coronary computed tomography angiography (coronary CTA) using first generation dual-source CT. Curvature and tortuosity of the coronary arteries were quantified using semi-automatically generated centerlines. Measurements were performed for individual segments and for the entire artery. Coronary segments were labeled according to the presence of significant stenosis, defined as >70% luminal narrowing, and the presence of plaque. Comparisons were made by segment and by artery, using linear mixed models., Results: Overall, median curvature and tortuosity were, respectively, 0.094 [0.071; 0.120] and 1.080 [1.040; 1.120] on a per-segment level, and 0.096 [0.078; 0.118] and 1.175 [1.090; 1.420] on a per-artery level. Curvature was associated with significant stenosis at a per-segment (p < 0.001) and per-artery level (p = 0.002). Curvature was 16.7% higher for segments with stenosis, and 13.8% higher for arteries with stenosis. Tortuosity was associated with significant stenosis only at the per-segment level (p = 0.002). Curvature was related to the presence of plaque at the per-segment (p < 0.001) and per-artery level (p < 0.001), tortuosity was only related to plaque at the per-segment level (p < 0.001)., Conclusion: Coronary artery geometry as derived from coronary CTA is related to the presence of plaque and significant stenosis., (Copyright © 2018 Society of Cardiovascular Computed Tomography. Published by Elsevier Inc. All rights reserved.)

Background & Aims: Frail patients with low model for end-stage liver disease (MELD) scores may be under-prioritised. Low skeletal muscle mass, namely sarcopenia, has been identified as a risk factor for waiting list mortality. A recent study proposed incorporating sarcopenia in the MELD score (MELD-Sarcopenia score). We aimed to investigate the association between sarcopenia and waiting list mortality, and to validate the MELD-Sarcopenia score (i.e. MELD + 10.35 * Sarcopenia)., Methods: We identified consecutive patients with cirrhosis listed for liver transplantation in the Eurotransplant registry between 2007-2014 and measured skeletal muscle mass on computed tomography. A competing risk analysis was used to compare survival of patients with and without sarcopenia, and concordance (c) indices were calculated to assess performance of the MELD and MELD-Sarcopenia score. We created a nomogram of the best predictive model., Results: We included 585 patients with a median MELD score of 14 (interquartile range 9-19), of which 254 (43.4%) were identified as having sarcopenia. Median waiting list survival was shorter in patients with sarcopenia than those without (p <0.001). This effect was even more pronounced in patients with MELD ≤15. The discriminative performance of the MELD-Sarcopenia score (c-index 0.820) for three-month mortality was lower than MELD score alone (c-index 0.839). Apart from sarcopenia and MELD score, other predictive variables were occurrence of hepatic encephalopathy before listing and recipient age. A model including all these variables yielded a c-index of 0.851., Conclusions: Sarcopenia was associated with waiting list mortality in liver transplant candidates with cirrhosis, particularly in patients with lower MELD scores. The MELD-Sarcopenia score was successfully validated in this cohort. However, incorporating sarcopenia in the MELD score had limited added value in predicting waiting list mortality., Lay Summary: In this study among patients with liver cirrhosis listed for liver transplantation, low skeletal muscle mass was associated with mortality on the waiting list, particularly in patients who were listed with low priority based on a low MELD score. However, adding these measurements to the currently used system for donor and organ allocation showed no added value., (Copyright © 2017 European Association for the Study of the Liver. Published by Elsevier B.V. All rights reserved.)

Background: Differences in coronary artery calcium (CAC) quantification of successive CT systems of one vendor could impact results of CAC screening and progression studies. The purpose of this study is to compare CAC quantification between three generations of dual-source computed tomography (DSCT) systems., Methods: Three DSCT generations were used to repeatedly scan an anthropomorphic chest phantom and three inserts. The first and second insert contained 100 small and nine large calcifications, respectively, to determine detectability, and the Agatston and (calibrated) mass score, respectively. A third insert containing a moving artificial coronary artery was used to determine impact of movement on calcium scoring. Data were acquired at 120 kVp, 90 reference mAs with prospective electrocardiographic(ECG)-gating at sequential and high-pitch spiral mode, for respectively first and second/third generation DSCT. Differences and variability in detectability and calcium scores were analyzed., Results: Although noise levels differed (p=<0.002), no differences in detectability were found between the three DSCT generations; median (range) for first, second and third generation were 11 (8), 11 (4) and 12 (2) out of 100 calcifications (p > 0.272). Between second and third generation no difference was found in Agatston score for the large calcification phantom (p > 0.05). The intra-scanner variability and inter-scanner median relative difference ranged for Agatston score from 2.1 to 8.3% and 0.5-12.7% and for mass score from 1.4% to 4.4% and 0.7-5.6%. Overall, intra-scanner variability was lowest for third generation DSCT., Conclusion: The three DSCT generations have similar detectability of calcifications. Median Agatston and mass score differed by no more than 12.7% and 5.6%., (Copyright © 2017 Society of Cardiovascular Computed Tomography. Published by Elsevier Inc. All rights reserved.)