RPS 113 - Advances in CT

Purpose:

To evaluate the technical performance of an ultra-high-resolution CT (UHRCT) system.

Methods and materials:

Standard procedures and phantoms were used to measure the modulation transfer function (MTF), slice sensitivity profile, uniformity, CT number accuracy, noise power spectrum (NPS), and low-contrast detectability of a UHRCT (Aquilion Precision, Canon Medical Systems) and compare them to a multidetector-row CT (MDCT, Aquilion ONE Genesis). The super-high-resolution (SHR) mode of the UHRCT uses detector elements (dels) with an effective size of 0.25x0.25 mm2 at the isocentre. The high-resolution (HR) mode bins two dels in the longitudinal direction. The normal-resolution (NR) mode bins 2x2 del, resulting in a del size equivalent to that of the MDCT system.

Results:

The UHRCT limiting-resolution MTF (10% MTF: 4.1 lp/mm) is twice as high as that in the NR mode and of the MDCT (10% MTF: 1.7 and 1.9 lp/mm, respectively). The slice sensitivity profile in the SHR mode (FWHM: 0.45 mm) is 40% narrower than that of the MDCT (0.77 mm). Uniformity and CT numbers are within the expected range for all UHRCT modes. Noise in the SHR and HR modes has a higher magnitude and higher frequency components than MDCT. Low-contrast detectability is lower for all UHRCT modes compared to MDCT, but a dose increase of about 14% for NR, and about 23% for HR and SHR, results in matching low contrast performance.

Conclusion:

HR and SHR modes of UHRCT result in twice the limiting spatial resolution of MDCT, requiring only a 23% increase in dose to achieve the same low-contrast detectability.

Limitations:

The use of linear metrics with non-linear reconstruction algorithms must be performed with care and only for the evaluation of the relative differences between techniques, as done here.

Ethics committee approval

n/a

Funding:

Funding received from Canon.

Purpose:

To improve the diagnostic workflow of bone imaging by providing organ-specific, context-sensitive images in whole-body photon-counting (PC) CT showing osseous structures at ultrahigh resolution (UHR) and soft tissue at standard resolution with reduced noise within a single image.

Methods and materials:

Acquisitions of the lumbar spine and pelvis were acquired in 5 patients and 10 forensic specimens using the UHR mode (pixel size 0.25 mm at the isocenter) of a PC CT prototype (SOMATOM CounT, Siemens Healthineers). Acquisitions are performed using 120 kV/300 mAs over a longitudinal scan range of 10 cm (CTDIvol 32cm=24 mGy, Deff=2.8 mSv). Images are reconstructed using sharp kernels (e.g. U70f) for bones and smoother kernels (e.g. D40f) for diagnostic tasks in soft tissue. Reconstruction of UHR data with low spatial resolution results in reduced noise compared to acquisitions with standard pixels sizes at the same dose [MedPhys 32(5):1321ff, 2005]. Bones, contrast agent, and soft tissue regions are automatically identified by decomposing the intrinsically acquired spectral data of the system. A context-sensitive image [Med. Phys. 45(10): 4541ff, 2018] is computed pixelwise using the sharp reconstruction for osseous structures and smooth reconstruction for soft tissue.

Results:

The resulting context-sensitive images simultaneously provide mutually exclusive image properties: low noise and moderate spatial resolution (7.4 lp/cm for D40f) in soft tissue and the highest spatial resolution in osseous structures (16.0 lp/cm for U70f).

Conclusion:

Context-sensitive images including UHR data improve clinical workflow by presenting mutually exclusive image properties within a single image and hence increased information content. This also may increase the rate of incidental findings.

Limitations:

The proposed method has only been applied to a limited number of patients and forensic specimens.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To evaluate the dependency of noise reduction of high-resolution whole-body photon-counting (PC) CT compared to conventional energy-integrating (EI) CT as a function of reconstruction kernel.

Methods and materials:

A semi-anthropomorphic liver phantom and forensic human specimens were measured using a prototype CT equipped with an EI and a PC detector (SOMATOM CounT, Siemens Healthineers). Images were acquired using the ultrahigh-resolution mode of the PC detector (PC-UHR, pixel size 0.25 mm at the isocenter) and the EI detector (0.6 mm at the isocenter). Reconstructions were performed using all available reconstruction kernels and resulted in different noise and spatial resolution levels. Theory predicts that PC-UHR images reconstructed to a lower spatial resolution result in reduced noise compared to EI at the same resolution and dose [MedPhys 32(5):1321ff, 2005]. Dose-normalised noise is evaluated and compared between both detector types for different kernels. Phantom results are verified in post mortem lung scans that typically require reconstructions using smooth and sharp kernels.

Results:

Low-resolution images (below MTF10%=6.5 lp/cm, B40f kernel) show no significant difference in image noise between both detector types. PC-UHR reconstructed with a higher target MTF (MTF10%=9.3 lp/cm, B80f) achieves a noise reduction of up to 30% at the same dose and resolution as the EI detector. A noise evaluation in bronchioli typically performed at low spatial resolution does not benefit from acquisitions with small detector pixels while the visualisation of fine lung structures using sharp resolution kernels improves significantly using PC-UHR.

Conclusion:

Clinical applications requiring medium or high resolution will be improved by the usage of PC-UHR and the consequent noise reduction equivalent to a radiation dose reduction.

Limitations:

The effect has only been evaluated in phantoms and forensic specimens.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To reduce metal artefacts in photon-counting (PC) CT.

Methods and materials:

CT systems with PC detectors usually provide multiple energy bins and acquire spectral information by default. In the presence of metal artefacts, the bin images require metal artefact reduction. We linearly combined (LC) the bin images in two ways: a) to produce an image with reduced artefacts and b) to minimise noise. A corrected LC (CLC) image was obtained by applying frequency split (FS) normalised metal artefact reduction (NMAR), the gold standard MAR algorithm, to the LC with minimum noise, inserting the high-frequency components of the LC with reduced artefacts. Then, each bin was corrected with FSNMAR using the CLC as a prior and corresponding FS image. Images of forensic specimens containing metal implants were acquired at the SOMATOM CounT PC prototype CT system (Siemens Healthineers, Germany) at 140 kV with a tube current of 300 mAs and thresholds of 25, 45, 70, and 90 keV, resulting in 4 bin images. To quantify the artefact content standard deviations in 3 ROIs (metal artefact, soft tissue without artefacts, and bone), our new PC-FSNMAR were compared to conventional FSNMAR applied to each bin separately (binwise FSNMAR).

Results:

Compared to binwise FSNMAR, PC-FSNMAR shows several advantages: a reduced noise and artefact level of up to 70% close to the metal, fewer metal artefacts even in the low-energy bins due to the usage of an optimal prior image, and preserved spectral information in all bins.

Conclusion:

PC-FSNMAR significantly improves image quality compared to conventional binwise FSNMAR, in particular in regions close to the metal and for low-energy bins, which suffer most from metal artefacts.

Limitations:

n/a

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To guarantee optimal diagnostic image quality to dose trade-off in CT, acquisition and reconstruction parameters are optimised for standardised patient positions. For indications related to anatomical regions from the lower abdomen up to the head, arms should be placed outside the field-of-view. However, for a large group of patients, including trauma or restricted shoulder mobility cases, one or both arms can be fully immobilised. Then, arm-positioning-specific instructions do not apply, affecting dose and image quality. This could be overcome using trauma-specific CT protocols adapted for non-trivial arm positions. CT-manufacturers allow for complex dose modulation with non-trivial arm-positioning, but implementation and optimisation in clinical practice is rare. One of the reasons is the scarcity of commercially available anthropomorphic phantoms for CT-dosimetry with articulated arms. Our goal was to develop affordable and reproducible production methods of totally articulated arm extensions of the RANDO-phantom, widely used for CT QC in radiology and radiotherapy.

Methods and materials:

3D-modelling, 3D-printing, and moulding techniques were used to manufacture the phantom, using anthropomorphic bones and soft-tissue-like materials. After testing various 3D-printing, silicone, gels, and polyurethane materials, a nylon-aluminium mix aluminide and a custom polyurethane-rubber mix were selected, corresponding to average HUs of human hard bone and fat-muscle mix, respectively. Image quality, attenuation, and dose modulation properties of the phantom, attached to RANDO-phantom, were evaluated for trauma-CT protocols.

Results:

Attenuation of the phantom [bone: (562±336) HU, soft tissue: (56±24) HU] closely mimicked values of the human arm [compact-bone: (800±400) HU, fat-to-muscle: (-80:100) HU] and is stable within the 80-140kV range. CTDIvol dose of the thorax trauma-CT varied by 12% (0.3 mGy) for various arm positions (arms-up, down, and mixed).

Conclusion:

A reproducible method for the production of totally-articulated arm phantom for CT-imaging was developed to optimise new CT protocols with complex arm positions.

Limitations:

The simplified soft tissue anatomy.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To investigate the performances of two CT systems produced by the same manufacturer (Siemens Somatom Flash/Edge) with different detector technologies and different generations of iterative reconstruction (IR) algorithms (Safire and Admire).

Methods and materials:

A homemade phantom was scanned on Flash CT equipped with ultrafast ceramic (UFC) detector and Edge CT with a stellar detector. Images were reconstructed with filtered back-projection (FBP) and IR algorithms. Image quality was evaluated in terms of low-contrast detectability using a channelised hotelling observer (CHO). The performance on signal detection predicted by the CHO was compared to the outcomes of three expert readers for tuning the model-observer (MO). The noise reduction, image texture, and the effects of IR algorithms were evaluated in terms of the noise power spectrum (NPS).

Results:

The analysis with MO showed the best performance of Edge respect to Flash system for both FBP and IR algorithms. The better stellar detector efficiency improved the signal detection of the Edge versus the Flash system of about 20%. The IR algorithms further improved the detectability of the signal on both systems. The evaluated noise reduction due to the stellar detector was 57%. Admire IR preserved a more traditional CT image texture appearance due to lower NPS peak shift.

Conclusion:

The stellar detector implemented on Edge CT showed an overall greater improvement in low-contrast detectability performance compared to the CT scanner equipped with a conventional detector. Large differences in NPS were observed between Safire and Admire.

Limitations:

A very simple detection task, i.e. detecting a known symmetric object placed on a uniform background which is very far from the clinical condition.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To evaluate the dual-energy (DE) performance and spectral separation in a clinical, photon counting CT (PCCT) and compare it to dual-source CT (DSCT) DE imaging.

Methods and materials:

Semi-anthropomorphic phantoms (abdomen/thorax) extendable with fat rings (S/M/L) equipped with iodine vials (5-30 mg/mL) were measured in a clinical PCCT prototype (SOMATOM CounT, Siemens). The system is comprised of a PC detector with two energy bins, [20 keV, T] and [T, eU] with T=threshold and U=tube voltage. PCCT measurements were performed at all tube voltages (80-140 kV) and threshold settings (50-90 keV). Further measurements were performed using energy integrating DSCT (SOMATOM Flash, Siemens) using 100 kV/Sn 140 kV. Spectral separation was quantified as the relative contrast media ratio (RelCM) between the energy bins and low/high images. Image noise and the dose-normalised contrast-to-noise ratio (CNRD) were evaluated in the resulting virtual non-contrast and iodine images.

Results:

RelCM of the PC detector varied between 1.2 and 2.7 and increased with higher thresholds and higher tube voltage, and slightly decreased with larger phantoms. RelCM of the DSCT was found as 2.20 on average over all phantoms. The maximum CNRD in the iodine images was found for T=67/68/70/71 keV for 80/100/120/140 kV. This corresponds to CNRD values of 1.07/1.21/1.32/1.40 mGy-1/2 for the small phantom, while the EI DSCT achieves 1.67 mGy-1/2 in this case.

Conclusion:

Intrinsically acquired PC data are able to provide VNC and iodine images similar to conventional DSCT and allow for a retrospective spectral analysis. Quantitatively, the CNRD is 16% lower with PC than with dual-source dual-energy CT.

Limitations:

The presented experiments are limited to phantoms. A forensic study is pending.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

Computed tomography (CT) devices acquire and reconstruct volumetric data. In this condition, the 2D analysis of the target transfer function (TTF) and noise power spectrum (NPS) may cause an overestimation of the detectability index, in particular, when iterative reconstruction (IR) algorithms are used.

3D detectability index generalisations have been used to evaluate low-contrast CT performance.

Methods and materials:

A cylindrical water-filled phantom was equipped with a central cylindrical PMMA insert and then scanned changing CTDIvol, reconstruction algorithm, filter, and slice thickness; 2D TTF was evaluated.

3D TTF was investigated using a central PMMA spherical insert in the same cylindrical water phantom.

2D and 3D NPS were calculated in a homogeneous phantom. Furthermore, FFT and NPS results were combined in non-pre-whitening (NPW) model to obtain the 3D detectability index.

Results:

3D TTFs show a spherical symmetry for low CTDI level. NPS analysis shows that the variation between 2D and 3D NPS is around 10% in terms of variance for FBP and iterative noise-based reconstruction; the difference increases significantly for model-based iterative reconstruction (from 20% to 180%). The 2D NPS and 3D NPS ratio are flat for FBP, while it depends on spatial frequency for model-based IR; it is higher for lower frequencies. A comparison between 3D and 2D detectability indexes shows a reduction of 15 and 35% for FBP and IR, respectively.

Conclusion:

The 3D generalisation of the detectability index shows a lower performance for a model-based IR algorithm when compared with the one obtained with standard 2D metrics.

Limitations:

A comparison with a human observer has to be performed in order to check absolute performances.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To analyse the main sources of textural radiomic features (RFs) variability in the different steps of radiomic workflow to quantify the effect on trial results and to evaluate possible recommendations for its reduction.

Methods and materials:

Radiomic workflow consists of several steps (imaging acquisition and reconstruction, segmentation, and features extraction) and for each of them, potential sources of variability were analysed.

The analyses were performed on Catphan® acquisitions or patients’ images for parameters not significant on the phantom. A wide set of scanner manufacturers and models were considered, involving different centres.

The software used for segmentation and features extraction were IntelliSpace Portal 8, 3DSlicer, and IBEX.

Results:

In the imaging acquisition step, repeatability, inter-scanner reproducibility, and tube voltage caused high RFs variability, with a mean relative standard deviation of 30% (range 0%-800%), while the workload was demonstrated to not affect RFs values strongly.

Regarding imaging reconstruction, the most crucial parameters were the algorithm and kernel, with RFs mean variations of 50% and 20%, respectively (maximum 600% and 400%).

The inter-reader variation in contouring was the overall largest source of variability, with a mean value of 60% (maximum 1000%).

In the RFs extraction step, the inter-slice resampling seemed not to be a useful solution and the choice of the feature category parameters was the most critical point to standardising in the radiomic workflow.

Conclusion:

A phantom study is preparatory to determine an optimal workflow that maximises RFs predictivity on patients. Once the main variability sources are identified, these can be limited or removed through a reasoned and shared standardisation. Nevertheless, a compromise between RFs stability and predictivity must be evaluated and tailored on each specific trial.

Limitations:

The conclusions derive from choices made for the analysis.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

Iterative reconstruction (IR) techniques allow for radiation dose savings through noise reduction in CT image processing. This effect on imaging texture and on radiomic features (RFs) was analysed.

Methods and materials:

The analyses were performed on Catphan® acquisitions on 3 scanners of different vendors (Siemens Healthineers, Philips Healthcare, and GE Healthcare) reconstructed with several combinations of an algorithm (filtered back projection (FBP) or IR), IR blending, and kernel.

RFs were extracted from the same cylindric region using IBEX.

The following variability sources were considered as relative discrepancy (RD) or relative standard deviation (RSD) of RFs values: FBP and IR algorithms with a fixed kernel, different blending, and different kernels with the same algorithm.

Results:

The RD between FBP and IR was feature, vendor, and kernel dependent. The Siemens scanner showed the highest variability (up to 2,469%), while the Philips scanner the lowest (up to 73%).

The effect of blending values was strongly vendor dependent with RSDs higher than the variability due to FBP-IR reconstruction.

Considering similar kernels as in clinical situations, the greatest effect can be seen in the Philips scanner (up to 70%).

The RFs trend as a function of the different kernels suggests differences among kernels of the 3 vendors but also highlights the physical meaning that lies under some RFs definitions.

Conclusion:

The imaging reconstruction step introduces a not negligible variability in the radiomic workflow.

A second reconstruction, other than the diagnostic one, could be proposed to reduce the variability affecting patient studies, fixing the most similar algorithm and kernel among the different vendors. Nevertheless, a residual difference which affects inter-scanner reproducibility is inevitable and must be taken into account.

Limitations:

The use of a uniform target within the phantom.

Ethics committee approval

n/a

Funding:

No funding was received for this work.