Visual grading characteristics and ordinal regression analysis during optimisation of CT head examinations

Photon-counting CT allows better visualization of temporal bone structures in comparison with current generation multi-detector CT

PURPOSE

To compare photon-counting CT (PCCT) and multi-detector CT (MDCT) for visualization of temporal bone anatomic structures.

METHODS

Thirty-six exams of temporal bones without pathology were collected from consecutive patients on a MDCT, and another 35 exams on a PCCT scanner. Two radiologists independently scored visibility of 14 structures for the MDCT and PCCT dataset, using a 5-point Likert scale, with a 2-month wash-out period. For MDCT, the acquisition parameters were: 110 kV, 64 × 0.6 mm (slice thickness reconstructed to 0.4 mm), pitch 0.85, quality ref. mAs 150, and 1 s rotation time; for PCCT: 120 kV, 144 × 0.2 mm, pitch 0.35, IQ level 75, and 0.5 s rotation time. Patient doses were reported as dose length product values (DLP). Statistical analysis was done using the Mann-Whitney U test, visual grading characteristic (VGC) analysis, and ordinal regression.

RESULTS

Substantial agreement was found between readers (intraclass correlation coefficient 0.63 and 0.52 for MDCT and PCCT, resp.). All structures were scored higher for PCCT (p < 0.0001), except for Arnold's canal (p = 0.12). The area under the VGC curve was 0.76 (95% CI, 0.73-0.79), indicating a significantly better visualization on PCCT. Ordinal regression showed the odds for better visualization are 354 times higher (95% CI, 75-1673) in PCCT (p < 0.0001). Average (range) of DLP was 95 (79-127) mGy*cm for MDCT and 74 (50-95) mGy*cm for PCCT (p < 0.001).

CONCLUSION

PCCT provides a better depiction of temporal bone anatomy than MDCT, at a lower radiation dose.

CRITICAL RELEVANCE STATEMENT

PCCT provides a better depiction of temporal bone anatomy than MDCT, at a lower radiation dose.

KEY POINTS

1. PCCT allows high-resolution imaging of temporal bone structures. 2. Compared to MDCT, the visibility of normal temporal bone structures is scored better with PCCT. 3. PCCT allows to obtain high-quality CT images of the temporal bones at lower radiation doses than MDCT.

Comparison of image quality assessments between interventional radiographers and interventional radiologists using digital subtraction angiography

Purpose

The aim of our study was to compare the image quality assessments of vascular anatomy between interventional radiographers and interventional radiologists using digital subtraction angiography (DSA) runs acquired during an interventional radiology procedure.

Approach

Visual grading characteristics (VGC) analysis was used to assess image quality by comparing two groups of images, where one group consisted of procedures in which radiation dose was optimized (group A, n = 10 ) and one group where dose optimization was not performed (group B, n = 10 ). The radiation dose parameters were optimized based on theoretical and empirical evidence to achieve radiation dose reductions during uterine artery embolization procedures. The two observer groups comprised of interventional radiologists ( n = 4 ) and interventional radiographers ( n = 4 ). Each observer rated the image quality of 20 DSA runs using a five-point rating scale.

Results

The VGC analysis produced an area under the VGC curve (AUC VGC ) of 0.55 for interventional radiographers ( P = 0.61 ) and AUCVGC of 0.52 for interventional radiologists ( P = 0.83 ). The optimization of radiation dose parameters demonstrated a reduction in kerma-area product by 35% ( P = 0.026 , d = 0.5 ) and reference air kerma (Ka, r ) by 43% ( P = 0.042 , d = 0.5 ) between group A and group B.

Conclusions

VGC analysis indicated that the image quality assessments of interventional radiographers were comparable with interventional radiologists, where a reduction in radiation dose revealed no effect on both observer groups regarding their image quality assessment of vascular anatomy.

Assessment of image quality in photon-counting detector computed tomography of the wrist - An ex vivo study

PURPOSE

The aim of this study was to evaluate the effect of reconstruction parameters on image quality in wrist imaging using photon-counting detector CT (PCD-CT) and to compare the results with images from an energy-integrating detector CT (EID-CT).

METHODS

Twelve cadaveric wrist specimens were examined using a prototype PCD-CT and a clinical EID-CT using similar radiation dose. Reconstruction parameters were matched between scanners. Also, sharper reconstruction kernels, a larger matrix size, and smaller slice thicknesses were evaluated for PCD-CT. Image noise, contrast-to-noise ratio (CNR) and image sharpness in trabecular structures were quantitatively measured. Image quality with respect to the visibility of cortical and trabecular bone structures was assessed by six radiologists using visual grading methods.

RESULTS

Images obtained with PCD-CT had lower noise (42.6 ± 3.9 HU vs 75.1 ± 6.3 HU), higher CNR (38.9 ± 4.5 vs 19.0 ± 2.4) and higher trabecular sharpness (63.5 ± 6.0 vs 53.7 ± 8.5) than those obtained with EID-CT using similar scan and reconstruction parameters (p < 0.001). The image sharpness in trabecular structures was further improved by using sharper kernels, despite higher noise levels. Radiologists had a strong preference for PCD-CT images both in terms of spatial resolution and suitability for bone imaging. Visual grading analysis showed an improved visibility of cortical bone, trabeculae and nutritive canals (p < 0.005).

CONCLUSION

PCD-CT offers improved image quality regarding bone structures in the wrist relative to EID-CT systems, particularly when sharper reconstruction kernels, smaller slice thickness and a larger image matrix size are used.

Visual grading experiments and optimization in CBCT dental implantology imaging: preliminary application of integrated visual grading regression

This study uses a general formulation of integrated visual grading regression (IVGR) and applies it to cone beam computed tomography (CBCT) scan data related to anatomical landmarks for dental implantology. The aim was to assess and predict a minimum acceptable dose for diagnostic imaging and reporting. A skull phantom was imaged with a CBCT unit at various diagnostic exposures. Key anatomical landmarks within the images were independently reviewed by three trained observers. Each provided an overall image quality score. Statistical analysis was carried out to examine the acceptability of the images taken, using an IVGR analysis that was formulized as a three-stage protocol including defining an integrated score, development of an ordinal regression, and investigation of the possibility for dose reduction through estimated parameters. For a unit increase in the logarithm of radiation dose, the odds ratio that the integrated score for an image assessed by observers being rated in a higher category was 3.940 (95% confidence interval: 1.016-15.280). When assessed by the observers, the minimum dose required to achieve a 75% probability for an image to be classified as at least acceptable was 1346.91 mGy·cm² dose area product (DAP), a 31% reduction compared to the 1962 mGy·cm² DAP default dosage of the CBCT unit. The kappa values of the intra and inter-observer reliability indicated moderate agreements, while a discrepancy among observers was also identified because each, as expected, perceived visibility differently. The results of this work demonstrate the IVGR's predictive value of dose saving in the effort to reduce dose to patients while maintaining reportable diagnostic image quality.

Evaluation of Ultra-low-dose Paediatric Cone-beam Computed Tomography for Image-guided Radiotherapy

AIMS

In image-guided radiotherapy, daily cone-beam computed tomography (CBCT) is rarely applied to children due to concerns over imaging dose. Simulating low-dose CBCT can aid clinical protocol design by allowing visualisation of new scan protocols in patients without delivering additional dose. This work simulated ultra-low-dose CBCT and evaluated its use for paediatric image-guided radiotherapy by assessment of image registration accuracy and visual image quality.

MATERIALS AND METHODS

Ultra-low-dose CBCT was simulated by adding the appropriate amount of noise to projection images prior to reconstruction. This simulation was validated in phantoms before application to paediatric patient data. Scans from 20 patients acquired at our current clinical protocol (0.8 mGy) were simulated for a range of ultra-low doses (0.5, 0.4, 0.2 and 0.125 mGy) creating 100 scans in total. Automatic registration accuracy was assessed in all 100 scans. Inter-observer registration variation was next assessed for a subset of 40 scans (five scans at each simulated dose and 20 scans at the current clinical protocol). This subset was assessed for visual image quality by Likert scale grading of registration performance and visibility of target coverage, organs at risk, soft-tissue structures and bony anatomy.

RESULTS

Simulated and acquired phantom scans were in excellent agreement. For patient scans, bony atomy registration discrepancies for ultra-low-dose scans fell within 2 mm (translation) and 1° (rotation) compared with the current clinical protocol, with excellent inter-observer agreement. Soft-tissue registration showed large discrepancies. Bone visualisation and registration performance reached over 75% acceptability (rated 'well' or 'very well') down to the lowest doses. Soft-tissue visualisation did not reach this threshold for any dose.

CONCLUSION

Ultra-low-dose CBCT was accurately simulated and evaluated in patient data. Patient scans simulated down to 0.125 mGy were appropriate for bony anatomy set-up. The large dose reduction could allow for more frequent (e.g. daily) image guidance and, hence, more accurate set-up for paediatric radiotherapy.

Evaluation of unenhanced axial T1W and T2W liver MR images acquired from institutions within the Republic of Ireland and the Kingdom of Saudi Arabia

INTRODUCTION

This multi-site study evaluated two breath-hold sequences commonly utilised for liver MRI; non-enhanced T1W-3D-FS-GRE-TRA and T2W-2D-FSE-TRA sequences, using physical measurements of SNR and CNR, and observer perceptions' (Visual Grading Analysis: VGA).

METHODS

Liver MR image datasets (n = 168) from nine hospitals in the Kingdom of Saudi Arabia (KSA) and 11 hospitals in the Republic of Ireland were evaluated. Images were categorised into two groups per sequence, defined by slice thickness (T2W-2D-FSE, ≤5 mm vs ≥ 6 mm and T1W-3D-GRE-FS, ≤3 mm vs 4 mm). Images were evaluated using visual grading analysis VGA and physical measurements: SNR/CNR. Account was taken of varying patient sizes based on AP/transverse diameter measurements.

RESULTS

Physical image quality measurements (SNR/CNR) returned no significant findings across Irish and KSA hospitals, for both sequences, despite variations in acquisition parameters. Statistically significant differences were found for some scoring criteria based on the observers' perceptions including spleen parenchyma, and spatial resolution for the non-enhanced T1W-3D-FS-GRE-TRA images, with a preference for images acquired using thin slices (≤3 mm). In addition, statistically significant difference was found for the scoring criteria motion artefact for the axial T2W-2D-FSE-TRA images, with a preference for images acquired using thick slices (≥5 mm). Negligible correlation was noted between SNR/CNR and measured abdominal AP/transverse diameters.

CONCLUSION

Whilst variations in sequences rendered no statistical differences in SNR/CNR findings, significant differences in observer image criteria scores was noted. The importance of both physical measurements and observers' perceptions evaluation methods for quality assessment of MR images was demonstrated and optimisation of liver sequence parameters is warranted.

Visual grading analysis of digital neonatal chest phantom X-ray images: Impact of detector type, dose and image processing on image quality

OBJECTIVES

To evaluate the impact of digital detector, dose level and post-processing on neonatal chest phantom X-ray image quality (IQ).

METHODS

A neonatal phantom was imaged using four different detectors: a CR powder phosphor (PIP), a CR needle phosphor (NIP) and two wireless CsI DR detectors (DXD and DRX). Five different dose levels were studied for each detector and two post-processing algorithms evaluated for each vendor. Three paediatric radiologists scored the images using European quality criteria plus additional questions on vascular lines, noise and disease simulation. Visual grading characteristics and ordinal regression statistics were used to evaluate the effect of detector type, post-processing and dose on VGA score (VGAS).

RESULTS

No significant differences were found between the NIP, DXD and CRX detectors (p>0.05) whereas the PIP detector had significantly lower VGAS (p< 0.0001). Processing did not influence VGAS (p=0.819). Increasing dose resulted in significantly higher VGAS (p<0.0001). Visual grading analysis (VGA) identified a detector air kerma/image (DAK/image) of ~2.4 μGy as an ideal working point for NIP, DXD and DRX detectors.

CONCLUSIONS

VGAS tracked IQ differences between detectors and dose levels but not image post-processing changes. VGA showed a DAK/image value above which perceived IQ did not improve, potentially useful for commissioning.

KEY POINTS

• A VGA study detects IQ differences between detectors and dose levels. • The NIP detector matched the VGAS of the CsI DR detectors. • VGA data are useful in setting initial detector air kerma level. • Differences in NNPS were consistent with changes in VGAS.

Methods for the analysis of ordinal response data in medical image quality assessment

The assessment of image quality in medical imaging often requires observers to rate images for some metric or detectability task. These subjective results are used in optimization, radiation dose reduction or system comparison studies and may be compared to objective measures from a computer vision algorithm performing the same task. One popular scoring approach is to use a Likert scale, then assign consecutive numbers to the categories. The mean of these response values is then taken and used for comparison with the objective or second subjective response. Agreement is often assessed using correlation coefficients. We highlight a number of weaknesses in this common approach, including inappropriate analyses of ordinal data and the inability to properly account for correlations caused by repeated images or observers. We suggest alternative data collection and analysis techniques such as amendments to the scale and multilevel proportional odds models. We detail the suitability of each approach depending upon the data structure and demonstrate each method using a medical imaging example. Whilst others have raised some of these issues, we evaluated the entire study from data collection to analysis, suggested sources for software and further reading, and provided a checklist plus flowchart for use with any ordinal data. We hope that raised awareness of the limitations of the current approaches will encourage greater method consideration and the utilization of a more appropriate analysis. More accurate comparisons between measures in medical imaging will lead to a more robust contribution to the imaging literature and ultimately improved patient care.

VGC ANALYZER: A SOFTWARE FOR STATISTICAL ANALYSIS OF FULLY CROSSED MULTIPLE-READER MULTIPLE-CASE VISUAL GRADING CHARACTERISTICS STUDIES

Visual grading characteristics (VGC) analysis is a non-parametric rank-invariant method for analysis of visual grading data. In VGC analysis, image quality ratings for two different conditions are compared by producing a VGC curve, similar to how the ratings for normal and abnormal cases in receiver operating characteristic (ROC) analysis are used to create an ROC curve. The use of established ROC software for the analysis of VGC data has therefore previously been proposed. However, the ROC analysis is based on the assumption of independence between normal and abnormal cases. In VGC analysis, this independence cannot always be assumed, e.g. if the ratings are based on the same patients imaged under both conditions. A dedicated software intended for analysis of VGC studies, which takes possible dependencies between ratings into account in the statistical analysis of a VGC study, has therefore been developed. The software-VGC Analyzer-determines the area under the VGC curve and its uncertainty using non-parametric resampling techniques. This article gives an introduction to VGC Analyzer, describes the types of analyses that can be performed and instructs the user about the input and output data.