Brian E. Jacobs

High-pitch low-voltage CT coronary artery calcium scoring with tin filtration: accuracy and radiation dose reduction

ObjectivesTo investigate diagnostic accuracy and radiation dose of high-pitch CT coronary artery calcium scoring (CACS) with tin filtration (Sn100kVp) versus standard 120kVp high-pitch acquisition.Methods78 patients (58% male, 61.5±9.1 years) were prospectively enrolled. Subjects underwent clinical 120kVp high-pitch CACS using third-generation dual-source CT followed by additional high-pitch Sn100kVp acquisition. Agatston scores, calcium volume scores, Agatston score categories, percentile-based risk categorization and radiation metrics were compared.Results61/78 patients showed coronary calcifications. Median Agatston scores were 34.9 [0.7–197.1] and 41.7 [0.7–207.2] and calcium volume scores were 34.1 [0.7–218.0] for Sn100kVp and 35.7 [1.1–221.0] for 120kVp acquisitions, respectively (both p<0.0001). Bland-Altman analysis revealed underestimated Agatston scores and calcium volume scores with Sn100kVp versus 120kVp acquisitions (mean difference: 16.4 and 11.5). However, Agatston score categories and percentile-based risk categories showed excellent agreement (ĸ=0.98 and ĸ=0.99). Image noise was 25.8±4.4HU and 16.6±2.9HU in Sn100kVp and 120kVp scans, respectively (p<0.0001). Dose-length-product was 9.9±4.8mGy*cm and 40.9±14.4mGy*cm with Sn100kVp and 120kVp scans, respectively (p<0.0001). This resulted in significant effective radiation dose reduction (0.13±0.07mSv vs. 0.57±0.2mSv, p<0.0001) for Sn100kVp acquisitions.ConclusionCACS using high-pitch low-voltage tin-filtered acquisitions demonstrates excellent agreement in Agatston score and percentile-based cardiac risk categorization with standard 120kVp high-pitch acquisitions. Furthermore, radiation dose was significantly reduced by 78% while maintaining accurate risk prediction.Key points• Coronary artery calcium scoring with tin filtration reduces radiation dose by 78%.• There is excellent correlation between high-pitch Sn100kVp and standard 120kVp acquisitions.• Excellent agreement regarding Agatston score categories and percentile-based risk categorization was achieved.• No cardiac risk reclassifications were observed using Sn100kVp coronary artery calcium scoring.

Comparison of the effect of radiation exposure from dual-energy CT versus single-energy CT on double-strand breaks at CT pulmonary angiography

PURPOSE To compare the effect of dual-source dual-energy CT versus single-energy CT on DNA double-strand breaks (DSBs) in blood lymphocytes at CT pulmonary angiography (CTPA). METHODS AND MATERIALS Sixty-two patients underwent either dual-energy CTPA (Group 1: n = 21, 80/Sn140 kVp, 89/38 mAs; Group 2: n = 20, 100/Sn140 kVp, 89/76 mAs) or single-energy CTPA (Group 3: n = 21, 120 kVp, 110 mAs). Blood samples were obtained before and 5 min after CTPA. DSBs were assessed with fluorescence microscopy and Kruskal-Walls tests were used to compare DSBs levels among groups. Volume CT dose index (CTDIvol), dose length product (DLP) and organ radiation dose were compared using ANOVA. RESULTS There were increased excess DSB foci per lymphocyte 5 min after CTPA examinations in three groups (Group 1: P = .001; Group 2: P = .001; Group 3: P = .006). There were no differences among groups regarding excess DSB foci/cell and percentage of excess DSBs (Group 1, 23%; Group 2, 24%; Group 3, 20%; P = .932). CTDIvol, DLP and organ radiation dose in Group 1 were the lowest among the groups (all P < .001). CONCLUSION DSB is increased following dual-source and single-source CTPA, while dual-source dual-energy CT protocols do not increase the estimated radiation dose and also do not result in a higher incidence of DNA DSBs in patients undergoing CTPA.

Dual-Energy Computed Tomography in Cardiothoracic Vascular Imaging

Dual energy computed tomography is becoming increasingly widespread in clinical practice. It can expand on the traditional density-based data achievable with single energy computed tomography by adding novel applications to help reach a more accurate diagnosis. The implementation of this technology in cardiothoracic vascular imaging allows for improved image contrast, metal artifact reduction, generation of virtual unenhanced images, virtual calcium subtraction techniques, cardiac and pulmonary perfusion evaluation, and plaque characterization. The improved diagnostic performance afforded by dual energy computed tomography is not associated with an increased radiation dose. This review provides an overview of dual energy computed tomography cardiothoracic vascular applications.

Artificial intelligence machine learning-based coronary CT fractional flow reserve (CT-FFRML): Impact of iterative and filtered back projection reconstruction techniques

BACKGROUND The influence of computed tomography (CT) reconstruction algorithms on the performance of machine-learning-based CT-derived fractional flow reserve (CT-FFRML) has not been investigated. CT-FFRML values and processing time of two reconstruction algorithms were compared using an on-site workstation. METHODS CT-FFRML was computed on 40 coronary CT angiography (CCTA) datasets that were reconstructed with both iterative reconstruction in image space (IRIS) and filtered back-projection (FBP) algorithms. CT-FFRML was computed on a per-vessel and per-segment basis as well as distal to lesions with ≥50% stenosis on CCTA. Processing times were recorded. Significant flow-limiting stenosis was defined as invasive FFR and CT-FFRML values ≤ 0.80. Pearsons correlation, Wilcoxon, and McNemar statistical testing were used for data analysis. RESULTS Per-vessel analysis of IRIS and FBP reconstructions demonstrated significantly different CT-FFRML values (p ≤ 0.05). Correlation of CT-FFRML values between algorithms was high for the left main (r = 0.74), left anterior descending (r = 0.76), and right coronary (r = 0.70) arteries. Proximal and middle segments showed a high correlation of CT-FFRML values (r = 0.73 and r = 0.67, p ≤ 0.001, respectively), despite having significantly different averages (p ≤ 0.05). No difference in diagnostic accuracy was observed (both 81.8%, p = 1.000). Of the 40 patients, 10 had invasive FFR results. Per-lesion correlation with invasive FFR values was moderate for IRIS (r = 0.53, p = 0.117) and FBP (r = 0.49, p = 0.142). Processing time was significantly shorter using IRIS (15.9 vs. 19.8 min, p ≤ 0.05). CONCLUSION CT reconstruction algorithms influence CT-FFRML analysis, potentially affecting patient management. Additionally, iterative reconstruction improves CT-FFRML post-processing speed.

The power and limitations of machine learning and artificial intelligence in cardiac CT

Artificial intelligence and machine learning are no longer a science fiction image of the future. In recent years, machine learning has seen increased use in most sciences to solve problems that were historically too complex to address. The latest advancements in computer technology allow machine learning algorithms to process large, real-time, high resolution data sets. The various fields of medical imaging are currently experiencing a major influx of machine learning applications, which are accompanied by valid questions, concerns, and speculations about potential developments in the upcoming decade. Specifically, experts in the field are interested in how machine learning will influence clinical practice and the role of the medical imager therein. Medicine is one of the more recent scientific fields to see advancements in machine learning applications. Perhaps to be expected, as a specialty in which pattern recognition plays a major role, imaging is at the center of attention. While medicine is certainly not the first field to embrace machine learning, it represents a more complex environment for its implementation due to the heightened stakes and downstream effects relative to other scientific disciplines. Over the past few decades, perhaps more than any other medical specialty, the imaging community has had to adapt to a wide array of new technological advances and developments. These recent developments are especially apparent in the field of cardiac imaging, which has been propelled by disruptive innovations for a more accurate and detailed visualization of the cardiovascular system. However, the increasing complexity of technology requires the imager to become familiar with an expanding body of knowledge, such as new acquisition protocols, reconstructions, and post-processing techniques, all while addressing the demand for detailed structured medical reports. Given the recent traction of various cardiac CT applications and the increasing role of CT as a screening tool, the number of patient examinations and consequent pressure on the interpreter for timely evaluation has shown substantial growth. All things considered, the importance of cardiac CT in medical imaging along with its proven adaptation to technological improvements has placed cardiac imaging in a unique position to lead various medical disciplines into the era of artificial intelligence. The use of machine learning will allow imaging experts to assess large amounts of data and improve the efficiency of study evaluation while decreasing observer variability, time to diagnosis and treatment, and hospital costs. The number of development-phase machine learning algorithms used in cardiac CT research has been rapidly increasing, with multiple promising applications already on the horizon. The majority of current machine learning applications are centered on relatively straightforward but labor intensive procedures that address vessel segmentation and quantification techniques such as coronary calcium detection, left ventricular volumetric quantification, or large vessel anatomical measurements. The potential clinical use of machine learning for such applications may reduce reader variability and subjectivity while allowing for rapid interpretation of cardiac CT studies. Perhaps the most promising machine learning application approaching clinical implementation is fully automated coronary calcium scoring. Calcium scoring is the perfect example of a time consuming task with easy-to-interpret results that can easily be trusted to a computer algorithm. Research has shown that machine learning is able to accurately assess calcium scores from dedicated non-contrast CT acquisitions, cardiac CT angiography datasets, as well as non-gated chest CT studies. In the future, the focus of machine learning will likely shift from basic anatomical evaluations to more complex tasks, such as image based prognostication and risk prediction. Although machine learning offers many advantages to cardiac imagers, as reviewed by Shing et al. in the current issue of JCCT, potential limitations remain and should always be considered. Taking into account the complexity of medical imaging and the ultimate impact on patient care, machine learning algorithms should be created by machine learning specialists with relevant knowledge of medicine and a clear understanding of possible consequences. Indeed, it will take time to introduce artificial intelligence specialists to the medical field and vice versa; however, without extensive communication and understanding between these two highly specialized fields, the potential of machine learning algorithms is limited. Machine learning algorithms are highly specialized in solving complex problems and creating equally complex networks to solve them. These algorithms may have excellent accuracy in predicting outcomes based on a multitude of input features, however, they currently lack the capacity to provide context and causality for their predictions. Imaging physicians have a unique ability to put important information into the correct context and convey this knowledge to patients and other healthcare providers, taking into consideration the factual information as well as local circumstances and personal preferences of both the patient and treating physician. An example that highlights the potential of interpretability issues in another field of medicine has been reported by Caruana et al. and later discussed by Cabitza et al. The authors present cases in which the most optimal algorithm in terms of mathematics is not necessarily the best algorithm for addressing the targeted clinical problem, emphasizing that without substantial knowledge of the field, wrong decisions can easily be made when machine learning predictions are trusted without a second review. When using machine learning, it is

Myocardial tissue characterization by combining late gadolinium enhancement imaging and percent edema mapping: a novel T2 map-based MRI method in canine myocardial infarction

BackgroundAssessing the extent of ischemic and reperfusion-associated myocardial injuries remains challenging with current magnetic resonance imaging (MRI) techniques. Our aim was to develop a tissue characterization mapping (TCM) technique by combining late gadolinium enhancement (LGE) with our novel percent edema mapping (PEM) approach to enable the classification of tissue represented by MRI voxels as healthy, myocardial edema (ME), necrosis, myocardial hemorrhage (MH), or scar.MethodsSix dogs underwent closed-chest myocardial infarct (MI) generation. Serial MRI scans were performed post-MI on days 3, 4, 6, 14, and 56, including T2 mapping and LGE. Dogs were sacrificed on day 4 (n = 4, acute MI) or day 56 (n = 2, chronic MI). TCMs were generated based on a voxel classification algorithm taking into account signal intensity from LGE and T2-based estimation of ME. TCM-based MI and MH were validated with post mortem triphenyl tetrazolium chloride (TTC) staining. Pearson’s correlation and Bland-Altman analyses were performed.ResultsThe MI, ME, and MH measured by TCM were 13.4% [25th–75th percentile 1.6–28.8], 28.1% [2.1–37.5] and 4.3% [1.0–11.3], respectively. TCM measured higher MH and MI compared to TTC (p = 0.0033 and p = 0.0007, respectively). MH size was linearly correlated with MI size by both MRI (r = 0.9528, p < 0.0001) and TTC (r = 0.9625, p < 0.0001). MH quantification demonstrated good agreement between TCM and TTC (r = 0.8766, p < 0.0001, 2.4% overestimation by TCM). A similar correlation was observed for MI size (r = 0.9429, p < 0.0001, 6.1% overestimation by TCM).ConclusionsPreliminary results suggest that the TCM method is feasible for the in vivo localization and quantification of various MI-related tissue components.

Diagnostic accuracy of low and high tube voltage coronary CT angiography using an X-ray tube potential-tailored contrast medium injection protocol

ObjectivesTo compare the diagnostic accuracy between low-kilovolt peak (kVp) (≤ 100) and high-kVp (> 100) third-generation dual-source coronary CT angiography (CCTA) using a kVp-tailored contrast media injection protocol.MethodsOne hundred twenty patients (mean age = 62.6 years, BMI = 29.0 kg/m2) who underwent catheter angiography and CCTA with automated kVp selection were separated into two cohorts (each n = 60, mean kVp = 84 and 117). Contrast media dose was tailored to the kVp level: 70 = 40 ml, 80 = 50 ml, 90 = 60 ml, 100 = 70 ml, 110 = 80 ml, and 120 = 90 ml. Contrast-to-noise ratio (CNR) was measured. Two observers evaluated image quality and the presence of significant coronary stenosis (> 50% luminal narrowing).ResultsDiagnostic accuracy (sensitivity/specificity) with ≤ 100 vs. > 100 kVp CCTA was comparable: per patient = 93.9/92.6% vs. 90.9/92.6%, per vessel = 91.5/97.8% vs. 94.0/96.8%, and per segment = 90.0/96.7% vs. 90.7/95.2% (all P > 0.64). CNR was similar (P > 0.18) in the low-kVp vs. high-kVp group (12.0 vs. 11.1), as ws subjective image quality (P = 0.38). Contrast media requirements were reduced by 38.1% in the low- vs. high-kVp cohort (53.6 vs. 86.6 ml, P < 0.001) and radiation dose by 59.6% (4.3 vs. 10.6 mSv, P < 0.001).ConclusionsAutomated tube voltage selection with a tailored contrast media injection protocol allows CCTA to be performed at ≤ 100 kVp with substantial dose reductions and equivalent diagnostic accuracy for coronary stenosis detection compared to acquisitions at > 100 kVp.Key points• Low-kVp coronary CT angiography (CCTA) enables reduced contrast and radiation dose.• Diagnostic accuracy is comparable between ≤ 100 and > 100 kVp CCTA.• Image quality is similar for low- and high-kVp CCTA.• Low-kVp image acquisition is facilitated by automated tube voltage selection.• Tailoring contrast injection protocols to the automatically selected kVp-level is feasible.

Beam-hardening in 70-kV Coronary CT angiography: Artifact reduction using an advanced post-processing algorithm

PURPOSE To investigate the effect of an iterative beam-hardening correction algorithm (iBHC) on artifact reduction and image quality in coronary CT angiography (cCTA) with low tube voltage. MATERIAL AND METHODS Thirty-six patients (17 male, mean age, 57.3 ± 14.5 years) were prospectively enrolled in this IRB-approved study and underwent 70-kV cCTA using a third-generation dual-source CT scanner. Images were reconstructed using a standard algorithm (Bv36) both with and without the iBHC technique. Several region-of-interest (ROI) measurements were performed in the inferior wall of the left ventricle (LV), an area prone to beam-hardening, as well as other myocardial regions. Coronary contrast-to-noise (CNR) and signal-to-noise ratios (SNR) were calculated. Two radiologists assessed subjective image quality. RESULTS The iBHC algorithm generally increased myocardial attenuation in all ROIs (P < 0.566); however, the increase was significantly more distinct in beam-hardening prone areas such as the inferior LV (increase, +13.9 HU, +18.6%, P < 0.001), compared to the remaining myocardium (increase, +4.4 HU, +4.5%, P < 0.003). While no significant difference was found for image noise (P < 0.092), greater CNR and SNR values for the left main coronary artery (increase, +20.7% and +17.3%, respectively) were found using the iBHC algorithm (both with P < 0.001). Subjective image quality was comparable between both image series (P = 0.217). CONCLUSION The iBHC post-processing algorithm leads to significantly reduced beam-hardening while providing improved objective and equivalent subjective image quality in 70-kV cCTA.

Cardiac Dual-Energy CT Applications and Clinical Impact

Purpose of ReviewTo provide an updated review on the applications of dual-energy CT (DECT) for cardiac and coronary imaging and the clinical impact therein. This review also attempts to provide guidance for the appropriate clinical use of DECT of the heart.Recent FindingsSimultaneous image acquisition at different kilovolt (kV) levels using DECT techniques offers a wide range of post-processing capabilities for cardiac imaging. Initially used for research purposes, this emerging technique is slowly being adopted into clinical routine for both coronary and myocardial imaging. Former limitations have been addressed in the development of DECT hardware and software. In addition to heavily validated analytical DECT approaches, various investigational techniques are evolving that may provide clinical value in the future.ConclusionDECT leads to more comprehensive and accurate diagnosis of myocardial and coronary disease by providing several ancillary analyses in addition to conventional CT images. However, the application of DECT should continue to be refined and further validated in future studies.

Coronary CT-Derived Fractional Flow Reserve

Purpose of ReviewCoronary artery disease (CAD) is the leading cause of mortality in the United States and is accountable for a significant portion of overall healthcare costs. Non-invasive imaging plays a major role in the modern workup of CAD. This article will educate the reader on the foundations of computed tomography-derived fractional flow reserve (CT-FFR) and provide guidance for its appropriate clinical use.Recent FindingsCoronary computed tomography angiography (CCTA) has high sensitivity and negative predictive value to non-invasively rule out CAD. However, discrimination of ischemia-inducing lesions based on macroscopic anatomy derived from either CCTA or the gold standard for the detection of anatomic stenoses, invasive coronary angiography, is suboptimal. Invasive pressure wire-guided estimation of FFR across coronary stenoses yields reliable functional information regarding the effect of a lesion on myocardial blood supply. Recently, non-invasive methods have attempted to calculate FFR from CCTA datasets. CT-FFR allows for higher specificity compared to CCTA alone, while preserving the high sensitivity and negative predictive value of CCTA. Whereas off-site solutions for CT-FFR calculation have been heavily validated and are clinically available, other techniques that can be performed on-site have recently evolved and are under current investigation.SummaryNon-invasive CT-FFR has facilitated the reliable assessment of the hemodynamic significance of coronary artery stenosis, potentially increasing the specificity of the modality while maintaining its excellent sensitivity.