Research Presentation Session: Physics in Medical Imaging

RPS 313 - What lies in store for us in CT or MR imaging

February 28, 11:30 - 12:30 CET

7 min
Multi-spot source cone-beam CT toward quantitative image evaluations
Antonio Sarno, Naples / Italy
Author Block: A. Sarno1, P. Cardarelli2, P. Mauriello1, A. Minopoli3, M. C. MOLLO1, S. Pardi1, G. Paternò2, M. Pugliese1, R. de Asmundis1; 1Naples/IT, 2Ferrara/IT, 3Pisa/IT
Purpose: The purpose of this study was to investigate the potential of a cone beam CT (CBCT) scanner equipped with a multi-spot x-ray source for reduction of scatter and cone-beam artefacts toward quantitative evaluations.
Methods or Background: The simulated study mimicked a CBCT scanner with a multi-spot x-ray source. This configuration consists in several focal spots arranged in a linear array parallel to the rotation axis and permits to acquire projections in an unconventional scanning trajectory (ExoCT). A specific beam collimation is employed for scatter reduction and a controller operates the several spots in order to have a non-conventional trajectory of the source during the gantry rotation for tightening the FOV sampling. In-silico studies were performed to evaluate the conspicuity improvements of selected details and the impact on accuracy of estimates of materials’ attenuation coefficients/HU.
Results or Findings: In a test conducted using a Defrise phantom, which alternates PMMA slabs and 1-mm air gaps, ExoCT configuration demonstrated that the conspicuity of such gaps is kept constant across the reconstructed volume, differently from what happened in CBCT where it reduces moving along the axial direction. The use of an ExoCT configuration with 10 focal points parallel to the rotation axis permitted to reduce the scatter-to-primary ratio of 86%. On the other hand, three spots were sufficient to increase the estimation accuracy of the attenuation coefficients of a PMMA phantom up to 10%.
Conclusion: We investigated an innovative solution for updating the CBCT classical configuration toward quantitative evaluations of the images. Such a solution showed – unlike the results attainable with CBCT – to preserve image conspicuity over the entire reconstructed FOV and permitted to improve the accuracy in the estimates of the sample attenuation coefficients by 10%.
Limitations: The study was conducted in-silico making it a limitation.
Funding for this study: This work rose from qCT project funded by the Italian Ministry of University and Research (CUP E53D23012420006)
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: No ethical approval was required for this study.
7 min
Image domain deep-learning low-dose CT simulator
Sjoerd A. M. Tunissen, Nijmegen / Netherlands
Author Block: S. A. M. Tunissen1, N. Moriakov2, M. Mikerov1, E. J. Smit1, I. Sechopoulos1, J. Teuwen2; 1Nijmegen/NL, 2Amsterdam/NL
Purpose: The main goal of this study was to develop an image-domain method to generate low-dose CT (LDCT) scans from clinical-dose CT (CDCT) scans reconstructed with non-linear reconstruction methods, e.g., iterative reconstruction, without access to the reconstruction method.
Methods or Background: The method consists of three convolutional neural networks (CNNs), to (i) denoise the CDCT to obtain a noiseless estimate, (ii) estimate the standard deviation (σ) in each voxel of the LDCT, and, (iii) estimate the local noise power spectra (NPS) in the LDCT. White noise is transformed into LDCT noise, using the σ and NPS estimates, and added to the noiseless estimate to obtain the simulated LDCT. For training, paired brain LDCTs and CDCTs were used, reconstructed with iterative reconstruction (AIDR3D, Canon Medical Systems), divided into training/validation/test sets (251/25/50 cases). Each CNN was evaluated on the test set by (i) determining the decrease in standard deviation in the cerebrospinal fluid (CSF), (ii) comparing the estimated and actual σ of the LDCT noise inside the skull, and, (iii) comparing the NPS of the generated and actual LDCT noise.
Results or Findings: The denoising network decreased the σ by a median (IQR) factor of 1.71 (1.61-1.95). The median (IQR) difference between estimated and actual σ was +0.1 (-0.2-+0.3) HU. The median (IQR) error of the radially-averaged 2D NPS of the simulated and actual LDCT noise inside the skull was 13.9% (11.9%-15.9%).
Conclusion: The proposed method allows for generation of LDCT from CDCT scans without access to the reconstruction algorithm and works fully in the image domain. Making LDCTs more available for testing new applications without patient radiation.
Limitations: The CNNs of the pipeline need to be retrained for different reconstruction methods. The method is tested only on brain scans, however it is not anatomy specific.
Funding for this study: The funding for this study was funded by the Health Holland, Canon Medical Systems.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: Not applicable for this study.
7 min
Anthropomophic 3D printed lung vessel phantoms combined with lung nodules for thorax CT imaging clinical performance evaluation
Irene Hernandez-Giron, Dublin / Ireland
Author Block: I. Hernandez-Giron1, P. McHale1, A. Gaffney1, B. Snow1, J. Egan1, C. D'Helft1, R. Byrne1, W. Veldkamp2, J. den Harder3; 1Dublin/IE, 2Leiden/NL, 3Amsterdam/NL
Purpose: This study aimed to design and manufacture 3D-printed low cost customised anthropomorphic phantoms for thorax CT imaging evaluation, including lung vessels and disease (lung parenchyma and nodules).
Methods or Background: Two identical mirrored models of the lung vessel tree (vessels diameters between 1 cm and 1 mm) were generated inside elliptical inserts (10x15x6 cm) using in-house code (Matlab). By design, small phantom sections with vessels were closed to trap the 3D-printed powder in to mimic diseased parenchyma. Lung nodules with different geometries (spiculared, spherical, ovoids, rugous) were designed (Meshmixer), with 3-15 mm diameter range. A thorax elliptical model (30x20x6 cm) with holes to insert the lungs and the spine, was manufactured with PMMA using CNC. The lung inserts were 3D printed with TPU (HP, MJF technique), the spine with allumide (SLS) and the nodules with different 3D printers and materials (PA12-MJF, PA12-BlueSint-SLS, PRUSA Tough Resin). The thorax phantom was scanned in a Siemens Somaton Edge Plus CT (thorax protocol) for a range of kV and the attenuation of the materials measured with selected ROIs and histograms .
Results or Findings: The attenuation of the phantom materials was, for 120 kV: thorax-PMMA (118+-4)HU, spine-allumide (785+-10)HU, lung vessels-TPU (80+-10HU), parenchyma-(raw-TPUpowder) (-680+-30)HU. For the lung nodules it was: PA12-Bluesint (-90+-20HU), PA12-MJM (-32+-10HU) and Prusa_ThoughResin(123+-10HU). The 3D printed materials showed in general a slight increase in HU with increasing kV. The total cost of this in-house anthropomorphic phantom was around 500 euros.
Conclusion: A low cost anthropomorphic thorax phantom, containing realistic lung vessel distributions and lung nodules was designed and manufactured with tissue-equivalent materials using 3D printing, to be used in task-based clinical assessment of image quality in CT.
Limitations: The accuracy and reproducibility of the 3D-printed lungs and nodules will be stablised in the future.
Funding for this study: The lung models were created with in-house software created during the NWO funded CLUES project (Pr. Nr. 13592, 2015-2019, Wouter Veldkamp project leader). The in-house code was expanded on during the project Through the eyes of AI: safe and optimal integration of Artificial Intelligence in the Radiology Workflow (Pr.Nr 17392, 2019-2022), funded by NWO-Veni Personal grant programma awarded to I. Hernandez-Giron
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: This study did not use any patient data and did not require Ethics committee aproval
7 min
Development of a breast perfusion phantom including in-line optical spectroscopy measurements for the validation and testing of dynamic contrast-enhanced dedicated breast CT
Liselot Goris, Hengelo / Netherlands
Author Block: L. Goris1, S. Manohar2, I. Sechopoulos3; 1Hengelo/NL, 2Enschede/NL, 3Nijmegen/NL
Purpose: To develop a dynamic perfusion breast phantom to validate the time-dependent quantitative iodine concentration estimates in dynamic contrast-enhanced dedicated breast CT (DCE-bCT).
Methods or Background: Perfusion functionality is added to a 3D-printed anatomic breast phantom with tumor mimic, by incorporating a pumping system, including two programmable pumps for input of water and contrast agent, tubing, and an output reservoir. Potassium iodide is used as the contrast agent since its light absorbance was found to be concentration-dependent. The in-line spectroscopy setup includes a light source (400-600 nm), to illuminate the contrast inlet and outlet leading into and leaving the breast phantom respectively, and photodetectors to measure the transmitted light. To evaluate precision, nine different iodine concentrations (0.5-7.6 mg I/mL) were tested three times and the coefficient of variation (CoV) was calculated. For accuracy, measurements of six concentrations were used to fit a calibration curve, and the errors of the remaining three were calculated. The division of concentrations used for fitting and testing was repeated three times to determine the average error. A time-varying iodine profile, including wash-in and wash-out, was pumped through the phantom and simultaneously measured with the optical and DCE-bCT systems.
Results or Findings: The in-line spectroscopy measurements had a CoV of 0.29% and an average error of 0.13%. The optical system detected the time-varying iodine profiles, with the recorded timepoints matching those visible in the transmission curves.
Conclusion: A combination of a 3D-printed phantom, a pumping system, and real-time spectroscopy seems a feasible approach for physical simulations of DCE processes in a breast for validation of DCE-bCT.
Limitations: Iodine-concentration estimates through the tumor mimic are not yet possible.
Funding for this study: ERC Grant No.864929
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: Not applicable
7 min
Model-based noise correction of CT perfusion blood flow measurements based on digital perfusion phantoms
Neha Vats, Heidelberg / Germany
Author Block: N. Vats, P. Mayer, M. Klauß, H-U. Kauczor, W. Stiller, S. Skornitzke; Heidelberg/DE
Purpose: Image noise can negatively influence the accuracy of CT perfusion blood flow (BF) measurements, the aim of this study was to evaluate model-based noise correction using digital perfusion phantoms.
Methods or Background: Digital perfusion phantoms were simulated by forward convolution of impulse response function with arterial input function averaged over 59 patients, considering ground-truth BF (GTBF) of 5-420ml/100 ml/min with a temporal sampling of 1.5s. To investigate the impact of noise on perfusion measurements, 2x576 random samples of Gaussian noise (standard deviation=25HU) were introduced. These noise-added phantoms were evaluated on a commercially available workstation (syngo.via; Siemens Healthineers) using deconvolution to measure noise-impacted BF. Phantoms were iteratively generated and evaluated to measure BF accurately by approximating GTBF from noise-impacted BF. To this end, noise correction was estimated from differences between measured BF and noise-impacted BF. The absolute difference between BF and GTBF were calculated, as well as the contrast-to-noise ratio (CNR). For validation, parenchyma and tumors of 14 pancreatic adenocarcinoma patients were evaluated.
Results or Findings: The measured noise-impacted BF and noise-corrected BF were 140.3±111.7 ml/100ml/min, and 131.9±125.9 ml/100ml/min, respectively, whereas GTBF was 131.3±127.7 ml/100ml/min. After correction, the average absolute difference in BF measurements decreased significantly from 18.8 to 3.6 ml/100ml/min and CNR improved from 2.52 to 2.66. For patients, BF converged from 148.3±50.8 ml/100ml/min to 155.0±91.5 ml/100ml/min for parenchyma, and 45.8±20.3 ml/100ml/min to 13.3±18.7 ml/100ml/min for tumour tissue following correction.
Conclusion: Convergence of BF in simulated and patient data signifies efficacy of developed algorithm in improving measurement accuracy by correcting negative influence of image noise. With further refinement, algorithm holds the potential to standardise perfusion measurements, enhancing comparability across patients, imaging centres, and equipment vendors, thus contributing to more accurate diagnosis and treatment planning.
Limitations: For patient dataset, ground-truth were unavailable, and BF in some tumours approached to zero with correction.
Funding for this study: This study was funded by the BMBF: grant 031L0163.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The research protocol received approval from the ethics committee at University Hospital Heidelberg and was conducted in accordance with the ethical standards of the World Medical Association (Declaration of Helsinki).
7 min
An update on iMet-MRI: a European project aimed at improving metrology for quantitative MRI
Matt G Hall, London / Ukraine
Author Block: C. McGrath1, C. Clarkson2, A. McDowell2, E. Cooke2, M. Causevic3, M. Cashmore2, P. Tofts4, A. Manzin5, M. G. Hall2; 1Belfast/UK, 2London/UK, 3Sarajevo/BA, 4Brighton/UK, 5Torino/IT
Purpose: Conventional clinical MRI is designed to produce single-use images for use by individual human radiologists. Because conventional MRI does not contain quantitative information, comparisons of images between different scanners and timepoints is extremely challenging. Quantitative MRI (qMRI), in contrast, aims to makes measurements of physical parameters. As such it offers huge potential for improved reproducibility and to detect diffusion changes in tissue which are difficult or impossible to detect in conventional images. Effective measurements require uncertainties to be understood and the performance of individual systems to be quantified. The iMet-MRI project aims to provide a solid foundation for qMRI to support clinical translation. This is an update on its progress.
Methods or Background: Metrologically characterised materials suitable for quantitative T1 and T2 imaging, diffusion, fat fraction, and iron content. Samples with multiple target parameter values have been produced. We have developed two sets of scan protocols: one maximising measurement quality, another for clinically feasible timescales. Analysis codes have been developed for all measurands and analysis is in progress. We have also developed a detailed digital phantom which has allowed simulation-based investigation of sensitivity to different measurement inaccuracies.
Results or Findings: A modular phantom designed and manufactured, phantoms validated, and scan protocols developed. Data has been received from core MRI systems and is being analysed, with additional acquisition continuing.
Conclusion: The project is now approximately 75% complete, and we are on target to deliver all our objectives. Experience from protocol implementation is being collated ahead of the production of a good practice guide for MRI QA. The software tools will also be made available to allow open reproducibility of all our results.
Limitations: The project's focus is the performance of the scanner, not the interaction of scanner and patient.
Funding for this study: This project (iMet-MRI) has received funding from the EMPIR programme co-financed by the Participating States and from the European Union's Horizon 2020 research and innovation programme.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: Not Applicable for this study.
7 min
Magic angle Magnetic resonance imaging (maMRI): first in vivo study using a unique low field scanner: what can we see?
Dimitri Amiras, London / United Kingdom
Author Block: D. Amiras, K. Chappell, J. McGinley, H. Lanz, C. Gupte, M. Ristic; London/UK
Purpose: Conventional knee MRI (cMRI) imaging is sensitive and specific at identifying acute injuries in soft tissues due the presence of blood or fluid. However, due to the inherent low MRI signal in collagen, cMRI does not show a significant contrast between the scarred lax tissues with poor collagen alignment and normal tissues. It is observed that well aligned collagen fibres produce increased signal, at short TEs, when imaged at the Magic Angle (MA≈55°) to the main magnetic field (B0). Our aim is to exploit this effect.

The aim of our study was to investigate whether a novel low-field magnet rotating about 2 axes could exploit the magic angle phenomenon to generate images of soft tissues in the knees of healthy volunteers (maMRI).
Methods or Background: A bespoke low-field system was developed to exploit the magic angle effect. This comprises a 0.15T open magnet with B0 being parallel to the magnet poles. The magnet can be rotated about two orthogonal axis to produce arbitrary orientation of B0 to the stationary patient.
The overall magnet assembly provides an accessible gap sufficient to comfortably accommodate extremity imaging. The scanner was also capable of performing cMRI to a standard comparable to other low-field MRI. A pre-clinical study with ethical approval for imaging healthy volunteers imaged subjects, maMRI sequences were performed on each subject in addition to cMRI.
Results or Findings: Our results highlight the potential benefit of maMRI - healthy ligaments and tendons could potentially be differentiated from those that are degenerative and those with chronic injuries. This is of considerable interest in informing clinical decision-making and can provide valuable information for the prevention of injury.
Conclusion: Our results highlight the potential benefit of magic maMRI - healthy ligaments and tendons could potentially be differentiated from those that are degenerative and those with chronic injuries. This is of considerable interest in informing clinical decision-making and can provide valuable information for prevention of injury.
Limitations: Not applicable for this study.
Funding for this study: This study was funded with the Wellcome Trust Innovator Award.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Local ethics committee approved this study.

This session will not be streamed, nor will it be available on-demand!