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

Radiographer Education and Learning in Artificial Intelligence (REAL-AI): A survey of radiographers, radiologists, and students' knowledge of and attitude to education on AI

INTRODUCTION

In Autumn 2023, amendments to the Health and Care Professions Councils (HCPC) Standards of Proficiency for Radiographers were introduced requiring clinicians to demonstrate awareness of the principles of AI and deep learning technology, and its application to practice' (HCPC 2023; standard 12.25). With the rapid deployment of AI in departments, staff must be prepared to implement and utilise AI. AI readiness is crucial for adoption, with education as a key factor in overcoming fear and resistance. This survey aimed to assess the current understanding of AI among students and qualified staff in clinical practice.

METHODS

A survey targeting radiographers (diagnostic and therapeutic), radiologists and students was conducted to gather demographic data and assess awareness of AI in clinical practice. Hosted online via JISC, the survey included both closed and open-ended questions and was launched in March 2023 at the European Congress of Radiology (ECR).

RESULTS

A total of 136 responses were collected from participants across 25 countries and 5 continents. The majority were diagnostic radiographers 56.6 %, followed by students 27.2 %, dual-qualified 3.7 % and radiologists 2.9 %. Of the respondents, 30.1 % of respondents indicated that their highest level of qualification was a Bachelor's degree, 29.4 % stated that they are currently using AI in their role, whilst 27 % were unsure. Only 10.3 % had received formal AI training.

CONCLUSION

This study reveals significant gaps in training and understanding of AI among medical imaging staff. These findings will guide further research into AI education for medical imaging professionals.

IMPLICATIONS FOR PRACTICE

This paper lays foundations for future qualitative studies on the provision of AI education for medical imaging professionals, helping to prepare the workforce for the evolving role of AI in medical imaging.

Cardiac SABR: Image matching techniques for accurate treatment delivery

BACKGROUND

Ventricular tachycardia is an irregular heartbeat conventionally treated using invasive cardiac catheter ablation and medication. However, when standard treatments have been exhausted, cardiac SABR provides a final treatment option to this high-mortality condition. Complex diagnostic mapping and planning scans enable multi-disciplinary target delineation for a 25Gy single fraction. However, organs at risk (OAR) near the target make this treatment challenging to plan and deliver. Publications from cardiologists report the efficacy of cardiac SABR, however there is limited data on the treatment delivery and image matching of this complex procedure.

METHODS

Four specialist therapeutic radiographers experienced in cardiac SABR reviewed 40 CBCTs from 10 patients treated in the UK. Each therapeutic radiographer conducted five image matches: a manual match (manual), an automatic match to the heart structure (auto) and the auto match followed by manual adjustment to the PTV (PTV), all using three degrees of freedom (DoF) only. The auto and PTV matches were also repeated using 6DoF. Inter-observer variability was quantified using 95% limits of agreement from a modified Bland-Altman analysis.

RESULTS

The limits of agreement were smallest in the automatic matches suggesting the algorithm is reliable. A manual adjustment from the auto match to the PTV is clinically appropriate to optimise target coverage. The limits of agreement were smaller in the 6DoF PTV match 1.06 mm, 1.24 mm, 1.68 mm than the 3DoF PTV match 1.57 mm, 2.06 mm, 2.11 mm (lateral, vertical, longitudinal).

CONCLUSION

The 6DoF CBCT image match has less variability and therefore suggest using a 6DoF couch for treatment delivery.

IMPLICATIONS FOR PRACTICE

Cardiac SABR CBCT image matching at treatment delivery is complex, optimisation of CBCT acquisition parameters and therapeutic radiographer training is essential prior to implementation.

Radiation dose reduction and image quality in pediatric paranasal sinus CT: with automatic tube current modulation and iterative reconstruction technique

Our objective is to evaluate radiation dose and image quality in pediatric paranasal sinus computed tomography (CT) with automatic tube current modulation (ATCM) and sinogram-affirmed iterative reconstruction algorithm (SAFIRE). CT scans from 80 patients were divided into two groups: Group A [80 kVp, pitch 1.5, 40 mAs, the filtered back projection (FBP) algorithm] and Group B (70 kVp, pitch 3, ATCM with reference at 40 mAs, SAFIRE strengths 1-5). We have evaluated image quality and radiation dose. Group B demonstrated significantly lower volume computed tomography dose index, dose-length product, and effective dose than Group A (0.13 ± 0.03 vs. 1.57 ± 0.01 mGy, 2.27 ± 0.82 vs. 19.88 ± 2.01 mGy·cm, and 0.0081 ± 0.0017 vs. 0.079 ± 0.013 mSv, respectively; P < .001). Increasing SAFIRE strengths correlated with noise reduction and SNR enhancement. Group B's noise and SNRsoft at SAFIRE strength 5 were comparable with Group A. Images reconstructed with SAFIRE strength 5 in Group B exhibit comparable image quality with FBP in Group A.

Feasibility of a Prototype Image Reconstruction Algorithm for Motion Correction in Interventional Cone-Beam CT Scans

RATIONALE AND OBJECTIVES

Assess the feasibility of a prototype image reconstruction algorithm in correcting motion artifacts in cone-beam computed tomography (CBCT) scans of interventional instruments in the lung.

MATERIALS AND METHODS

First, phantom experiments were performed to assess the algorithm, using the Xsight lung phantom with custom inserts containing straight or curved catheters. During scanning, the inserts moved in a continuous sinusoidal or breath-hold mimicking pattern, with varying amplitudes and frequencies. Subsequently, the algorithm was applied to CBCT data from navigation bronchoscopy procedures. The algorithm's performance was assessed quantitatively via edge-sharpness measurements and qualitatively by three specialists.

RESULTS

In the phantom study, the algorithm improved sharpness in 13 out of 14 continuous sinusoidal motion and five out of seven breath-hold mimicking scans, with more significant effects at larger motion amplitudes. Analysis of 27 clinical scans showed that the motion corrected reconstructions had significantly sharper edges than standard reconstructions (2.81 (2.24-6.46) vs. 2.80 (2.16-4.75), p = 0.003). These results were consistent with the qualitative assessment, which showed higher scores in the sharpness of bronchoscope-tissue interface and catheter-tissue interface in the motion-corrected reconstructions. However, the tumor demarcation ratings were inconsistent between raters, and the overall image quality of the new reconstructions was rated lower.

CONCLUSION

Our findings suggest that applying the new prototype algorithm for motion correction in CBCT images is feasible. The algorithm improved the sharpness of medical instruments in CBCT scans obtained during diagnostic navigation bronchoscopy procedures, which was demonstrated both quantitatively and qualitatively.

CT abnormalities 3 and 12 months after hospitalization for COVID-19 and association with disease severity: A prospective cohort study

OBJECTIVES

To investigate changes in chest CT between 3 and 12 months and associations with disease severity in patients hospitalized for COVID-19 during the first wave in 2020.

MATERIALS AND METHODS

Longitudinal cohort study of patients hospitalized for COVID-19 in 2020. Chest CT was performed 3 and 12 months after admission. CT images were evaluated using a CT severity score (CSS) (0-12 scale) and recoded to an abbreviated version (0-3 scale). We analyzed determinants of the abbreviated CSS with multivariable mixed effects ordinal regression.

RESULTS

242 patients completed CT at 3 months, and 124 (mean age 62.3±13.3, 78 men) also at 12 months. Between 3 and 12 months (n = 124) CSS (0-12 scale) for ground-glass opacities (GGO) decreased from median 3 (25th-75th percentile: 0-12) at 3 months to 0.5 (0-12) at 12 months (p<0.001), but increased for parenchymal bands (p<0.001). In multivariable analysis of GGO, the odds ratio for more severe abbreviated CSS (0-3 scale) at 12 months was 0.11 (95%CI 0.11 0.05 to 0.21, p<0.001) compared to 3 months, for WHO severity category 5-7 (high-flow oxygen/non-invasive ventilation/ventilator) versus 3 (non-oxygen use) 37.16 (1.18 to 43.47, p = 0.032), and for age ≥60 compared to <60 years 4.8 (1.33 to 17.6, p = 0.016). Mosaicism was reduced at 12 compared to 3 months, OR 0.33 (95%CI 0.16 to 0.66, p = 0.002).

CONCLUSIONS

GGO and mosaicism decreased, while parenchymal bands increased from 3 to 12 months. Persistent GGO were associated with initial COVID-19 severity and age ≥60 years.

Evaluation of left ventricular and left atrial volumetric function from native MR multislice 4D flow magnitude data

OBJECTIVES

To assess the feasibility, precision, and accuracy of left ventricular (LV) and left atrial (LA) volumetric function evaluation from native magnetic resonance (MR) multislice 4D flow magnitude images.

MATERIALS & METHODS

In this prospective study, 60 subjects without signs or symptoms of heart failure underwent 3T native cardiac MR multislice 4D flow and bSSFP-cine realtime imaging. LV and LA volumetric function parameters were evaluated from 4D flow magnitude (4D flow-cine) and bSSFP-cine data using standard software to obtain end-diastolic volume (EDV), end-systolic volume (ESV), ejection-fraction (EF), stroke-volume (SV), LV muscle mass (LVM), LA maximum volume, LA minimum volume, and LA total ejection fraction (LATEF). Stroke volumes derived from both imaging methods were further compared to 4D pulmonary artery flow-derived net forward volumes (NFV). Methods were compared by correlation and Bland-Altman analysis.

RESULTS

Volumetric function parameters from 4D flow-cine and bSSFP-cine showed high to very high correlations (r = 0.83-0.98). SV, LA volumes and LATEF did not differ between methods. LV end-diastolic and end-systolic volumes were slightly underestimated (EDV: -2.9 ± 5.8 mL; ESV: -2.3 ± 3.8 mL), EF was slightly overestimated (EF: 0.9 ± 2.6%), and LV mass was considerably overestimated (LVM: 39.0 ± 11.4 g) by 4D flow-cine imaging. SVs from both methods correlated very highly with NFV (r = 0.91 in both cases) and did not differ from NFV.

CONCLUSION

Native multislice 4D flow magnitude data allows precise evaluation of LV and LA volumetric parameters; however, apart from SV, LV volumetric parameters demonstrate bias and need to be referred to their respective normal values.

CLINICAL RELEVANCE STATEMENT

Volumetric function assessment from native multislice 4D flow magnitude images can be performed with routinely used clinical software, facilitating the application of 4D flow as a one-stop-shop functional cardiac MR exam, providing consistent, simultaneously acquired, volume and flow data.

KEY POINTS

• Native multislice 4D flow imaging allows evaluation of volumetric left ventricular and atrial function parameters. • Left ventricular and left atrial function parameters derived from native multislice 4D flow data correlate highly with corresponding standard cine-derived parameters. • Multislice 4D flow-derived volumetric stroke volume and net forward volume do not differ.

Image processing setting adaptions according to image dose and radiologist preference can improve image quality in computed radiography of the equine distal limb: A cadaveric study

Image processing (IP) in digital radiography has been steadily refined to improve image quality. Adaptable settings enable users to adjust systems to their specific requirements. This prospective, analytical study aimed to investigate the influence of different IP settings and dose reductions on image quality. Included were 20 cadaveric equine limb specimens distal to the metacarpophalangeal and metatarsophalangeal joints. Images were processed with the Dynamic Visualization II system (Fujifilm) using five different IP settings including multiobjective frequency processing, flexible noise control (FNC), and virtual grid processing (VGP). Seven criteria were assessed by three veterinary radiology Diplomates and one veterinary radiology resident in a blinded study using a scoring system. Algorithm comparison was performed using an absolute visual grading analysis. The rating of bone structures was improved by VGP at full dose (P < .05; AUCVGC  = 0.45). Überschwinger artifact perception was enhanced by VGP (P < .001; AUCVGC  = 0.66), whereas image noise perception was suppressed by FNC (P < .001; AUCVGC  = 0.29). The ratings of bone structures were improved by FNC at 50% dose (P < .05; AUCVGC  = 0.44), and 25% dose (P < .001; AUCVGC  = 0.32), and clinically acceptable image quality was maintained at 50% dose (mean rating 2.16; 95.8% ratings sufficient or better). The favored IP setting varied among observers, with higher agreement at lower dose levels. These findings supported using individualized IP settings based on the radiologist's preferences and situational image requirements, rather than using default settings.

Development and validation of core entrustable professional activities for abdominal radiology

OBJECTIVES

To develop and validate European entrustable professional activities (EPAs) for sub-specialised hepatobiliary and gastrointestinal (HB/GI) diagnostic imaging.

MATERIALS AND METHODS

Both European Society of Radiology and national curricula in HB/GI diagnostic radiology were thoroughly reviewed, resulting in preliminary EPAs drafted by a pilot group of expert radiologists in 2 different countries. Each EPA was fully described with 7 components (Specification/limitations; Potential risks of failing; Relevant domains of competence; Required experience, knowledge, skills, attitude and behaviour; Assessment information sources to assess progress and ground a summative entrustment decision; Entrustment for which level of supervision is to be reached; and Expiration date). The modified Delphi method with 3 Delphi rounds was chosen for validation. Content validity index (CVI) and median values were used for validation.

RESULTS

There were 15 preliminary EPAs, some of them divided according to 2 levels: resident and fellow level. The 37 members of the Delphi group were based in 2 different European countries with a background experience of 10 represented countries. Subsequent to the first Delphi round, 6 EPAs were accepted (CVI ≥ 0.8, median ≥ 4), 6 needed major revisions (CVI 0.7-0.79, median ≥ 4), 3 were rejected (CVI < 0.7) and 1 was added. After the second Delphi round, both the 6 revised EPAs and the additional one met the validation criteria (CVI ≥ 0.8, median ≥ 4). Finally, 13 EPAs were validated during the 3rd Delphi round with an agreement percentage of 95-100%.

CONCLUSION

This study creates and validates EPAs for sub-specialised HB/GI diagnostic imaging.

CRITICAL RELEVANCE STATEMENT

Thirteen EPAs for sub-specialised hepatobiliary and gastrointestinal diagnostic imaging were created with a strong methodology, and as a first example set in sub-specialised diagnostic imaging, they provide a template for others to be created.

KEY POINTS

• The competence-based teaching in medical studies has recently been reintroduced through EPAs. • Thirteen EPAs have been developed for hepatobiliary and gastrointestinal sub-specialised diagnostic imaging. • These EPAs were validated using a Delphi modified method and provide a template for other to be created.

Generative Adversarial Networks Can Create High Quality Artificial Prostate Cancer Magnetic Resonance Images

The recent integration of open-source data with machine learning models, especially in the medical field, has opened new doors to studying disease progression and/or regression. However, the ability to use medical data for machine learning approaches is limited by the specificity of data for a particular medical condition. In this context, the most recent technologies, like generative adversarial networks (GANs), are being looked upon as a potential way to generate high-quality synthetic data that preserve the clinical variability of a condition. However, despite some success, GAN model usage remains largely minimal when depicting the heterogeneity of a disease such as prostate cancer. Previous studies from our group members have focused on automating the quantitative multi-parametric magnetic resonance imaging (mpMRI) using habitat risk scoring (HRS) maps on the prostate cancer patients in the BLaStM trial. In the current study, we aimed to use the images from the BLaStM trial and other sources to train the GAN models, generate synthetic images, and validate their quality. In this context, we used T2-weighted prostate MRI images as training data for Single Natural Image GANs (SinGANs) to make a generative model. A deep learning semantic segmentation pipeline trained the model to segment the prostate boundary on 2D MRI slices. Synthetic images with a high-level segmentation boundary of the prostate were filtered and used in the quality control assessment by participating scientists with varying degrees of experience (more than ten years, one year, or no experience) to work with MRI images. Results showed that the most experienced participating group correctly identified conventional vs. synthetic images with 67% accuracy, the group with one year of experience correctly identified the images with 58% accuracy, and the group with no prior experience reached 50% accuracy. Nearly half (47%) of the synthetic images were mistakenly evaluated as conventional. Interestingly, in a blinded quality assessment, a board-certified radiologist did not significantly differentiate between conventional and synthetic images in the context of the mean quality of synthetic and conventional images. Furthermore, to validate the usability of the generated synthetic images from prostate cancer MRIs, we subjected these to anomaly detection along with the original images. Importantly, the success rate of anomaly detection for quality control-approved synthetic data in phase one corresponded to that of the conventional images. In sum, this study shows promise that high-quality synthetic images from MRIs can be generated using GANs. Such an AI model may contribute significantly to various clinical applications which involve supervised machine-learning approaches.

Comparison of lesion detection and conspicuity between narrow-angle and wide-angle digital breast tomosynthesis for dense and non-dense breasts

Digital breast tomosynthesis (DBT) has been shown to improve both sensitivity and specificity for breast cancer detection compared to full-field digital mammography. However, its performance could be limited for patients with dense breasts. Clinical DBT systems vary in their system designs, one of which is the acquisition angular range (AR), which leads to varied performance for different imaging tasks. In this study, we aim to compare DBT systems with different AR. We used a previously validated cascaded linear system model to investigate the dependence of in-plane breast structural noise (BSN) and detectability of masses on AR. We conducted a pilot clinical study to compare the lesion conspicuity between clinical DBT systems with the narrowest and the widest AR. Patients called back for diagnostic imaging on suspicious findings were imaged with both narrow-angle (NA) and wide-angle (WA) DBT. We analyzed the BSN for clinical images using noise power spectrum (NPS) analysis. A 5-point Likert scale was used in the reader study to compare the lesion conspicuity. Our theoretical calculation results show that increasing AR leads to reduced BSN and improved mass detectability. The NPS analysis on clinical images shows the lowest BSN for WA DBT. The WA DBT provides better lesion conspicuity for masses and asymmetries and shows a greater advantage for non-microcalcification lesions in dense breasts. The NA DBT provides better characterizations for microcalcifications. The WA DBT can downgrade false-positive findings seen on NA DBT. In conclusion, WA DBT could improve the detection of masses and asymmetries for patients with dense breasts.

Quality assessment of anatomical MRI images from generative adversarial networks: Human assessment and image quality metrics

BACKGROUND

Generative Adversarial Networks (GANs) can synthesize brain images from image or noise input. So far, the gold standard for assessing the quality of the generated images has been human expert ratings. However, due to limitations of human assessment in terms of cost, scalability, and the limited sensitivity of the human eye to more subtle statistical relationships, a more automated approach towards evaluating GANs is required.

NEW METHOD

We investigated to what extent visual quality can be assessed using image quality metrics and we used group analysis and spatial independent components analysis to verify that the GAN reproduces multivariate statistical relationships found in real data. Reference human data was obtained by recruiting neuroimaging experts to assess real Magnetic Resonance (MR) images and images generated by a GAN. Image quality was manipulated by exporting images at different stages of GAN training.

RESULTS

Experts were sensitive to changes in image quality as evidenced by ratings and reaction times, and the generated images reproduced group effects (age, gender) and spatial correlations moderately well. We also surveyed a number of image quality metrics. Overall, Fréchet Inception Distance (FID), Maximum Mean Discrepancy (MMD) and Naturalness Image Quality Evaluator (NIQE) showed sensitivity to image quality and good correspondence with the human data, especially for lower-quality images (i.e., images from early stages of GAN training). However, only a Deep Quality Assessment (QA) model trained on human ratings was able to reproduce the subtle differences between higher-quality images.

CONCLUSIONS

We recommend a combination of group analyses, spatial correlation analyses, and both distortion metrics (FID, MMD, NIQE) and perceptual models (Deep QA) for a comprehensive evaluation and comparison of brain images produced by GANs.

Image quality enhancement of CT hepatic portal venography using dual energy blending with computer determined parameters

BACKGROUND

Previous studies have shown that using some post-processing methods, such as nonlinear-blending and linear blending techniques, has potential to improve dual-energy computed (DECT) image quality.

OBJECTIVE

To improve DECT image quality of hepatic portal venography (CTPV) using a new non-linear blending method with computer-determined parameters, and to compare the results to additional linear and non-linear blending techniques.

METHODS

DECT images of 60 patients who were clinically diagnosed with liver cirrhosis were selected and studied. Dual-energy scanning (80 kVp and Sn140 kVp) of CTPV was utilized in the portal venous phase through a dual-source CT scanner. For image processing, four protocols were utilized including linear blending with a weighing factor of 0.3 (protocol A) and 1.0 (protocol B), non-linear blending with fixed blending width of 200 HU and set blending center of 150HU (protocol C), and computer-based blending (protocol D). Several image quality indicators, including signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR) and contrast of hepatic portal vein and hepatic parenchyma, were evaluated using the paired-sample t-test. A 5-grade scale scoring system was also utilized for subjective analysis.

RESULTS

SNR of protocols A-D were 9.1±2.1, 12.1±3.0, 11.6±2.8 and 14.4±3.2, respectively. CNR of protocols A-D were 4.6±1.3, 8.0±2.3, 7.0±2.0 and 9.8±2.4, respectively. The contrast of protocols A-D were 37.7±11.6, 91.9±21.0, 66.2±19.0 and 107.7±21.3, respectively. The differences between protocol D and other three protocols were significant (P < 0.01). In subjective evaluation, the modes of protocols A, B, C, and D were rated poor, good, generally acceptable, and excellent, respectively.

CONCLUSION

The non-linear blending technique of protocol D with computer-determined blending parameters can help improve imaging quality of CTPV and contribute to a diagnosis of liver disease.

The effect of different statistical approaches on image quality data obtained from radiological examinations

INTRODUCTION

Selection of optimal image acquisition protocols in medical imaging remains a grey area, the superimposed use of the Likert scale in radiological image quality evaluations creates an additional challenge for the statistical analysis of image quality data. Using a simulation study, we have trialled a novel approach to analysing radiological image quality Likert scale data.

METHODS

A simulation study was undertaken where simulated datasets were generated based on the distribution of Likert scale values according to varying image acquisition protocols from a real dataset. Simulated Likert scale values were pooled in four different ways; the mean, median, mode and the summation of patient Likert scale values of which the total was assigned a categorical Likert scale value. Estimates of bias, MAPE and RMSPE were then calculated for all four pooling approaches to determine which method most accurately represented an expert's opinion.

RESULTS

When compared to an expert's opinion, the method of summation and categorisation of Likert scale values was most accurate 49 times out of the 114 (43.0%) tests. The mean 28 times out of 114 (24.6%), the median 23 times out of 114 (20.2%) and the mode 17 times out of 114 (14.9%).

CONCLUSION

We conclude that our method of summation and categorisation of Likert scale values is most often the best representation of the simulated data compared to the expert's opinion.

IMPLICATIONS FOR PRACTICE

There is scope to reproduce this simulation study with multiple observers to reflect clinical reality more accurately with the dynamic nature of multiple observers. This also prompts future investigation into other anatomical areas, to see if the same methods produce similar results.

EVALUATION OF VGC ANALYZER BY COMPARISON WITH GOLD STANDARD ROC SOFTWARE AND ANALYSIS OF SIMULATED VISUAL GRADING DATA

The purpose of the present work was to evaluate the use of resampling statistical methods for analysis of visual grading data-implemented in the software VGC Analyzer-by comparing the reanalyzed results from previously performed visual grading studies with the results calculated by gold standard receiver operating characteristic (ROC) methodology, Obuchowski-Rockette (OR)-Dorfman-Berbaum-Metz (DBM) multiple-readers and multiple-case (MRMC) and by analysis of simulated visual grading data where the true distribution was presumed to be known. The reanalysis was performed on two multiple-reader studies with non-paired data and paired data, respectively. The simulation study was performed by simulating a large number of visual grading characteristics (VGC) studies and by analyzing the statistical distribution of null hypothesis (H0) rejection rate. The comparison with OR-DBM MRMC showed good agreement when analyzing non-paired data for both fixed-reader and random-reader settings for the calculated area under the curve values and the confidence intervals (CIs). For paired data analysis, VGC Analyzer showed significantly lower CIs compared with the ROC software. This effect was also illustrated by the simulation study, where the VGC Analyzer, in general, showed good accuracy for simulated studies with stable statistical basis. For simulated studies with unstable statistics, the accuracy in the H0 rejection rate decreased. The present study has shown that resampling methodology can be used to accurately perform the statistical analysis of a VGC study, although the resampling technique used makes the method sensitive to small data sets.

Optimisation of Varian TrueBeam head, thorax and pelvis CBCT based on patient size

Purpose:

The aim of this study was to optimise patient dose and image quality of Varian TrueBeam cone beam computed tomography (CBCT) pelvis, thorax and head and neck (H&N) images based on patient size.

Methods:

An elliptical phantom of small, medium and large size was designed representative of a local population of pelvis, thorax and H&N patients. The phantom was used to establish the relationship between image noise, CT and CBCT exposure settings. Using this insight, clinical images were optimised in phases and the image quality graded qualitatively by radiographers. At each phase, the time required to match the images was recorded from the record and verify system.

Results:

Average patient diameter was a suitable metric to categorise patient size. Phantom measurements showed the power relationship between noise and CBCT exposure settings of value −0·15, −0·35 and −0·43 for thorax, pelvis and H&N, respectively. These quantitative phantom measurements provided confidence that phased variation of ~±20% in mAs should result in clinically usable images. Qualitative assessment of almost 2000 images reduced the exposure settings in H&N images by −50%, thorax images by up to −66% and pelvis images by up to −80%. These optimised CBCT settings did not affect the time required to match images.

Findings:

Varian TrueBeam CBCT mAs settings have been optimised for dose and image quality based on patient size for three treatment sites: pelvis, thorax and H&N. Quantitative phantom measurements provided insight into the magnitude of change to implement clinically. The final optimised exposure settings were determined from radiographer qualitative image assessment.

Optimisation of tube voltage range (kVp) for AP abdomen, pelvis and spine imaging of average patients with a digital radiography (DR) imaging system using a computer simulator

Objectives:

To investigate via computer simulation, an optimised tube voltage (kVp) range for caesium iodide (CsI)-based digital radiography (DR) of the abdomen, pelvis and lumbar spine.

Methods:

Software capable of simulating abdomen, pelvis and spine radiographs was used. Five evaluators graded clinical image criteria in images of 20 patients at tube voltages ranging from 60 to 120 kVp in 10 kVp increments. These criteria were scored blindly against the same patient reconstructed at a specific reference kVp. Linear mixed effects analysis was used to evaluate image scores for each criterion and test for statistical significance.

Results:

Score was dependent on tube voltage and image criteria; both were statistically significant. All criteria for all anatomies scored very poorly at 60 kVp. Scores for abdomen, pelvis and spine imaging peaked at 70, 70 and 100 kVp, respectively, but other kVp values were not significantly poorer.

Conclusions:

Results indicate optimum tube voltages of 70 kVp for abdomen and pelvis (with an optimum range 70–120 kVp), and 100 kVp (optimum range 80–120 kVp) for lumbar spine.

Advances in knowledge:

There are no recommendations for optimised tube voltage parameters for DR abdomen, pelvis or lumbar spine imaging. This study has investigated and recommended an optimal tube voltage range.

Improvement of image quality applying iterative scatter correction for grid-less skeletal radiography in trauma room setting

BACKGROUND

Iterative reconstruction is well established for CT. Plain radiography also takes advantage of iterative algorithms to reduce scatter radiation and improve image quality. First applications have been described for bedside chest X-ray. A recent experimental approach also provided proof of principle for skeletal imaging.

PURPOSE

To examine clinical applicability of iterative scatter correction for skeletal imaging in the trauma setting.

MATERIAL AND METHODS

In this retrospective single-center study, 209 grid-less radiographs were routinely acquired in the trauma room for 12 months, with imaging of the chest (n = 31), knee (n = 111), pelvis (n = 14), shoulder (n = 24), and other regions close to the trunk (n = 29). Radiographs were postprocessed with iterative scatter correction, doubling the number of images. The radiographs were then independently evaluated by three radiologists and three surgeons. A five-step rating scale and visual grading characteristics analysis were used. The area under the VGC curve (AUC_VGC) quantified differences in image quality.

RESULTS

Images with iterative scatter correction were generally rated significantly better (AUC_VGC = 0.59, P < 0.01). This included both radiologists (AUC_VGC = 0.61, P < 0.01) and surgeons (AUC_VGC = 0.56, P < 0.01). The image-improving effect was significant for all body regions; in detail: chest (AUC_VGC = 0.64, P < 0.01), knee (AUC_VGC = 0.61, P < 0.01), pelvis (AUC_VGC = 0.60, P = 0.01), shoulder (AUC_VGC = 0.59, P = 0.02), and others close to the trunk (AUC_VGC = 0.59, P < 0.01).

CONCLUSION

Iterative scatter correction improves the image quality of grid-less skeletal radiography in the clinical setting for a wide range of body regions. Therefore, iterative scatter correction may be the future method of choice for free exposure imaging when an anti-scatter grid is omitted due to high risk of tube-detector misalignment.

Use of a computer simulator to investigate optimized tube voltage for chest imaging of average patients with a digital radiography (DR) imaging system

Objective:

The aim of this study was to investigate via computer simulation a proposed improvement to clinical practice by deriving an optimized tube voltage (kVp) range for digital radiography (DR) chest imaging.

Methods:

A digitally reconstructed radiograph algorithm was used which was capable of simulating DR chest radiographs containing clinically relevant anatomy. Five experienced image evaluators graded clinical image criteria, i.e. overall quality, rib, lung, hilar, spine, diaphragm and lung nodule in images of 20 patients at tube voltages across the diagnostic energy range. These criteria were scored against corresponding images of the same patient reconstructed at a specific reference kVp. Evaluators were blinded to kVp. Evaluator score for each criterion was modelled with a linear mixed effects algorithm and compared with the score for the reference image.

Results:

Score was dependent on tube voltage and image criteria in a statistically significant manner for both. Overall quality, hilar, diaphragm and spine criteria performed poorly at low and high tube voltages, peaking at 80–100 kVp. Lung and lung nodule demonstrated little variation. Rib demonstrated superiority at low kVp.

Conclusion:

A virtual clinical trial has been performed with simulated chest DR images. Results indicate mid-range tube voltages of 80–100 kVp are optimum for average adults.

Advances in knowledge:

There are currently no specific recommendations for optimized tube voltage parameters for DR chest imaging. This study, validated with images containing realistic anatomical noise, has investigated and recommended an optimal tube voltage range.

Generative Adversarial Networks for the Creation of Realistic Artificial Brain Magnetic Resonance Images

Even as medical data sets become more publicly accessible, most are restricted to specific medical conditions. Thus, data collection for machine learning approaches remains challenging, and synthetic data augmentation, such as generative adversarial networks (GAN), may overcome this hurdle. In the present quality control study, deep convolutional GAN (DCGAN)–based human brain magnetic resonance (MR) images were validated by blinded radiologists. In total, 96 T1-weighted brain images from 30 healthy individuals and 33 patients with cerebrovascular accident were included. A training data set was generated from the T1-weighted images and DCGAN was applied to generate additional artificial brain images. The likelihood that images were DCGAN-created versus acquired was evaluated by 5 radiologists (2 neuroradiologists [NRs], vs 3 non-neuroradiologists [NNRs]) in a binary fashion to identify real vs created images. Images were selected randomly from the data set (variation of created images, 40%–60%). None of the investigated images was rated as unknown. Of the created images, the NRs rated 45% and 71% as real magnetic resonance imaging images (NNRs, 24%, 40%, and 44%). In contradistinction, 44% and 70% of the real images were rated as generated images by NRs (NNRs, 10%, 17%, and 27%). The accuracy for the NRs was 0.55 and 0.30 (NNRs, 0.83, 0.72, and 0.64). DCGAN-created brain MR images are similar enough to acquired MR images so as to be indistinguishable in some cases. Such an artificial intelligence algorithm may contribute to synthetic data augmentation for “data-hungry” technologies, such as supervised machine learning approaches, in various clinical applications.

Optimization of cone beam computed tomography image quality in implant dentistry

This study was conducted to optimize the cone beam computed tomography image quality in implant dentistry using both clinical and quantitative image quality evaluation with measurement of the radiation dose. A natural bone human skull phantom and an image quality phantom were used to evaluate the images produced after changing the exposure parameters (kVp and mA). A 10 × 5 cm2 field of view was selected for average adult. Five scans were taken with varying kVp (70-90 kVp) first at fixed 4 mA. After assessment of the scans and selecting the best kVp, nine scans were taken with 2-12 mA, and the kVp was fixed at the optimal value. A clinical assessment of the implant-related anatomical landmarks was done in random order by two blinded examiners. Quantitative image quality was assessed for noise/uniformity, artifact added value, contrast-to-noise ratio, spatial resolution, and geometrical distortion. A dosimetry index phantom and thimble ion chamber were used to measure the absorbed dose for each scan setting. The anatomical landmarks of the maxilla had good image quality at all kVp settings. To produce good quality images, the mandibular landmarks demanded higher exposure parameters than the maxillary landmarks. The quantitative image quality values were acceptable at all selected exposure settings. Changing the exposure parameters does not necessarily produce higher image quality outcomes but does affect the radiation dose to the patient. The image quality could be optimized for implant treatment planning at lower exposure settings and dose than the default settings.

Quantitative Measurements Versus Receiver Operating Characteristics and Visual Grading Regression in CT Images Reconstructed with Iterative Reconstruction: A Phantom Study

RATIONALE AND OBJECTIVES

This study aimed to evaluate the correlation of quantitative measurements with visual grading regression (VGR) and receiver operating characteristics (ROC) analysis in computed tomography (CT) images reconstructed with iterative reconstruction.

MATERIALS AND METHODS

CT scans on a liver phantom were performed on CT scanners from GE, Philips, and Toshiba at three dose levels. Images were reconstructed with filtered back projection (FBP) and hybrid iterative techniques (ASiR, iDose, and AIDR 3D of different strengths). Images were visually assessed by five readers using a four- and five-grade ordinal scale for liver low contrast lesions and for 10 image quality criteria. The results were analyzed with ROC and VGR. Standard deviation, signal-to-noise ratios, and contrast-to-noise ratios were measured in the images.

RESULTS

All data were compared to FBP. The results of the quantitative measurements were improved for all algorithms. ROC analysis showed improved lesion detection with ASiR and AIDR and decreased lesion detection with iDose. VGR found improved noise properties for all algorithms, increased sharpness with iDose and AIDR, and decreased artifacts from the spine with AIDR, whereas iDose increased the artifacts from the spine. The contrast in the spine decreased with ASiR and iDose.

CONCLUSIONS

Improved quantitative measurements in images reconstructed with iterative reconstruction compared to FBP are not equivalent to improved diagnostic image accuracy.

Comprehensive assessment of patient image quality and radiation dose in latest generation cardiac x-ray equipment for percutaneous coronary interventions

This study aimed to determine whether a reduction in radiation dose was found for percutaneous coronary interventional (PCI) patients using a cardiac interventional x-ray system with state-of-the-art image enhancement and x-ray optimization, compared to the current generation x-ray system, and to determine the corresponding impact on clinical image quality. Patient procedure dose area product (DAP) and fluoroscopy duration of 131 PCI patient cases from each x-ray system were compared using a Wilcoxon test on median values. Significant reductions in patient dose ([Formula: see text]) were found for the new system with no significant change in fluoroscopy duration ([Formula: see text]); procedure DAP reduced by 64%, fluoroscopy DAP by 51%, and "cine" acquisition DAP by 76%. The image quality of 15 patient angiograms from each x-ray system (30 total) was scored by 75 clinical professionals on a continuous scale for the ability to determine the presence and severity of stenotic lesions; image quality scores were analyzed using a two-sample [Formula: see text]-test. Image quality was reduced by 9% ([Formula: see text]) for the new x-ray system. This demonstrates a substantial reduction in patient dose, from acquisition more than fluoroscopy imaging, with slightly reduced image quality, for the new x-ray system compared to the current generation system.

Can image enhancement allow radiation dose to be reduced whilst maintaining the perceived diagnostic image quality required for coronary angiography?

OBJECTIVE

The aim of this research was to quantify the reduction in radiation dose facilitated by image processing alone for percutaneous coronary intervention (PCI) patient angiograms, without reducing the perceived image quality required to confidently make a diagnosis.

METHODS

Incremental amounts of image noise were added to five PCI angiograms, simulating the angiogram as having been acquired at corresponding lower dose levels (10-89% dose reduction). 16 observers with relevant experience scored the image quality of these angiograms in 3 states-with no image processing and with 2 different modern image processing algorithms applied. These algorithms are used on state-of-the-art and previous generation cardiac interventional X-ray systems. Ordinal regression allowing for random effects and the delta method were used to quantify the dose reduction possible by the processing algorithms, for equivalent image quality scores.

RESULTS

Observers rated the quality of the images processed with the state-of-the-art and previous generation image processing with a 24.9% and 15.6% dose reduction, respectively, as equivalent in quality to the unenhanced images. The dose reduction facilitated by the state-of-the-art image processing relative to previous generation processing was 10.3%.

CONCLUSION

Results demonstrate that statistically significant dose reduction can be facilitated with no loss in perceived image quality using modern image enhancement; the most recent processing algorithm was more effective in preserving image quality at lower doses. Advances in knowledge: Image enhancement was shown to maintain perceived image quality in coronary angiography at a reduced level of radiation dose using computer software to produce synthetic images from real angiograms simulating a reduction in dose.

Impact of latest generation cardiac interventional X-ray equipment on patient image quality and radiation dose for trans-catheter aortic valve implantations

OBJECTIVES

This study aimed to determine the impact on radiation dose and image quality of a new cardiac interventional X-ray system for trans-catheter aortic valve implantation (TAVI) patients compared to the previously-used cardiac X-ray system.

METHODS

Patient dose and image data were retrospectively collected from a Philips AlluraClarity (new) and Siemens Axion Artis (reference) X-ray system. Patient dose area product (DAP) and fluoroscopy duration of 41 patient cases from each X-ray system were compared using a Wilcoxon test. Ten patient aortograms from each X-ray system were scored by 32 observers on a continuous scale to assess the clinical image quality at the given phase of the TAVI procedure. Scores were dichotomised by acceptability and analysed using a Chi-squared test.

RESULTS

Significant reductions in patient dose (p << 0.001) were found for the new system with no significant change in fluoroscopy duration (p = 0.052); procedure DAP reduced by 55%, fluoroscopy DAP by 48% and "cine" acquisition DAP by 61%. There was no significant difference between image quality scores of the two X-ray systems (p = 0.06).

CONCLUSIONS

The new cardiac X-ray system demonstrated a very significant reduction in patient dose with no loss of clinical image quality. Advances in Knowledge: The huge growth of TAVI may impact on the radiation exposure of cardiac patients and particularly on operators including anaesthetists; cumulative exposure of interventional cardiologists performing high volume TAVI over 30-40 years may be harmful. The Phillips Clarity upgrade including improved image enhancement and optimised X-ray settings significantly reduced radiation without reducing clinically acceptable image quality.