Image Quality Criteria Research Articles

A novel free-breathing abdominal RAVE T2/T1 hybrid MRI sequence in patients with cystic fibrosis: Preliminary results

ObjectivesPatients with cystic fibrosis (CF) increasingly require imaging for the diagnosis of abdominal complications. We prospectively evaluated the image quality and signal-to-noise ratio (SNR) of a modern radial volumetric encoding (RAVE) T2/T1 hybrid sequence for abdominal magnetic resonance imaging (MRI). RAVET2/T1 is a three-dimensional radial sequence with fat saturation and blood flow suppression that acquires T2- and T1-weighted contrasts in one scan in an identical slice position during free-breathing. MethodsSixteen CF patients underwent axial T2 HASTE (1000 ms/93 ms TR/TE), T1 DIXON (6.8 ms/2.4 ms/4.8 ms TR/TE1/TE2), and RAVE T2/T1 hybrid sequence (1200 ms/1.7 ms/3.3 ms/4.9 ms/102 ms TR/TE1/TE2/TE3/TE4) of the upper abdomen at 1.5 Tesla. The SNR values in six different regions were assessed and compared using the Wilcoxon signed-rank test. The image quality criteria were rated on a 5-point Likert scale. ResultsIn all regions, the SNR was significantly higher in the T2 weighted aspect of the RAVE T2/T1 hybrid sequence compared to T2 HASTE (p < 0.05) and significantly lower in the T1 weighted in-phase aspect of the RAVE T2/T1 hybrid sequence compared to the T1 DIXON sequence (p < 0.05).Qualitatively the T2 weighted aspect of the RAVE T2/T1 hybrid sequence was rated significantly higher than the T2 HASTE in 6 of 7 categories (p < 0.05) and the T1 weighted in-phase aspect of the RAVE T2/T1 hybrid sequence was rated significantly higher than the T1 DIXON in 2 of 6 categories (p < 0.05). ConclusionsThe abdominal radial RAVE T2/T1 hybrid sequence provided higher image quality and SNR than the T2HASTEsequence. Together with increased robustness against motion artifacts, the RAVE T2/T1 hybrid sequence appears to be a good tool for abdominal imaging in CF patients.

KUALITAS CITRA SIDIK TULANG DENGAN PEMBERIAN DOSIS RADIOFARMAKA 99mTc-MDP BERDASARKAN BERAT BADAN DI INSTALASI KEDOKTERAN NUKLIR RS HASAN SADIKIN BANDUNG

A bone scan procedure using radiopharmaceutical 99mTc-MDP (Methylenediphosphonate) is an examination to help establish diagnostics in detecting metastases (spread) of cancer cells to bone. The dose given in this examination was according to the instructions from the Europe Association of Nuclear Medicine (EANM), the average dose given was 500MBq and also based on the guidelines in the attachment to the Regulation of the Head of Bapeten No. 17 of 2012 was 700 MBq. Justification for giving this dose will undoubtedly affect the acceptance of the exposure dose by the patient himself. On the other hand, giving a minimal dose sometimes creates doubts and concerns about the final result or poor image quality. Fulfillment of this image quality can be seen or calculated using ratio assessment. The results of this ratio assessment will show good image quality if it meets the target to background ratio results, which is the accumulation in the primary and background objects &gt; from 5:1. The result of the intensity ratio is obtained from the ROI (Region of Interest) in the lumbar spine divided by ROI (Region of Interest) in the non-bone (abdominal area), which is considered the background. Giving a low dose is expected so that the exposure dose is small, but the picture results are still fulfilled. Carried out the research by providing a dose of Bone scan examination based on the patient's body weight of 10 MBq/kg, then calculating the Target Background to Ratio on the results of scans in the lumbar region 1-5. Development of the research obtained the number of patients as many as 51 patients with three people removed from the study so that the number of patients who took data was 48 patients consisting of 38 women and ten men. The results of the calculation included 33 patients who met the ratio results (&gt;5), while 15 patients still did not meet the image quality criteria (&lt;5). The Conclusion of the ratio that meets the value &gt;5 is 69%, and those who do not meet the requirements are 31%. Keywords: Bone Scan, Ratio Results, 99mTc MDP

Evaluation of the Diagnostic Quality of Chest Radiographs

Introduction: With the advancement of conventional radiography to digital, digital radiography of the chest is commonly performed these days. The role and importance of diagnostic quality of radiographs are to help the radiologists and the clinicians in the diagnosis of diseases and their management, as well as to prevent the misdiagnosis of any pathology. This study was an attempt to quantify the diagnostic quality of chest radiographs by evaluating the quality of depiction of the anatomical details as well as other technical factors. Method: A total of 450 chest radiographs, produced at Tribhuvan University Teaching Hospital, were selected for study over two months period. Five image quality criteria i.e. anatomical coverage, rotation, adequate penetration, adequate inspiration, and scapula out of lung fields were evaluated and tabulated. Results: Out of the 450 radiographs taken for study, only 22.2% of the radiographs fulfilled all the image quality criteria, the rest 77.8% either lacked one or more of the quality criteria. Conclusion: Many technical factors affect the image quality of digital chest radiographs. However, the radiographer/technologist should always try to maintain adequate image quality of the radiographs.

METHODOLOGY FOR DETERMINING THE REQUIRED VALUE OF THE IMAGE QUALITY INDEX FOR DETECTION ALGORITHMS UNDER DISTORTION CONDITIONS

The article presents an improved technique for determining the required value of the image quality criterion for detection systems operating under distortion conditions. The calculated level of the quality indicator should be used to determine the fact of image distortion. On the basis of the proposed technique, a method of correlation detection in conditions of "smudge" and "defocusing" type distortions has been developed. The developed method of correlation detection is distinguished by the introduction of three stages: identification of distortions in the current image; formation of the convolution core based on the type and parameters of distortion; calculation of the number of iterations based on the required and current value of the image quality criterion. This made it possible to ensure the operation of the correlation detection method in conditions of distortion.

Assessment and comparison of image quality between two real-time sequences for dynamic MRI of distal joints at 3.0 Tesla.

Real-time sequences allow functional evaluation of various joint structures during a continuous motion and help understand the pathomechanics of underlying musculoskeletal diseases. To assess and compare the image quality of the two most frequently used real-time sequences for joint dynamic magnetic resonance imaging (MRI), acquired during finger and ankle joint motion. A real-time dynamic acquisition protocol, including radiofrequency (RF)-spoiled and balanced steady-state free precession (bSSFP) sequences, optimized for temporal resolution with similar spatial resolution, was performed using a 3.0-T MRI scanner on 10 fingers and 12 ankles from healthy individuals during active motion. Image quality criteria were evaluated on each time frame and compared between these two sequences. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were determined and compared from regions of interest placed on cortical bone, tendon, fat, and muscle. Visualization of anatomical structures and overall image quality appreciation were rated by two radiologists using a 0-10 grading scale. Mean CNR was significantly higher with bSSFP sequence compared to RF-spoiled sequence. The grading score was in the range of 5-9.3 and was significantly higher with RF-spoiled sequence for bone and joint evaluation and overall image appreciation on the two joints. The standard deviation for SNR, CNR, and grading score during motion was smaller with RF-spoiled sequence for both the joints. The inter-reader reliability was excellent (>0.75) for evaluating anatomical structures in both sequences. A RF-spoiled real-time sequence is recommended for the in vivo clinical evaluation of distal joints on a 3.0-T MRI scanner.

Ultra-Low-Dose Photon-Counting CT Imaging of the Paranasal Sinus With Tin Prefiltration: How Low Can We Go?

In this study, we compared photon-counting detector computed tomography (PCD-CT) and energy-integrating detector computed tomography (EID-CT) for ultra-low-dose paranasal sinus examinations with employed tin prefiltration. The goal of our investigation was to define the most dose-effective scan protocols for diagnostic assessment of midface trauma, preoperative sinonasal anatomy, and acute rhinosinusitis. Five cadaveric heads were examined with the standard-resolution scan mode of both CT systems using a tube potential of 100 kV and tin prefiltration for 7 dose-equivalent scan protocols (CTDI vol = 4.16-0.15 mGy) and 2 additional ultra-low-dose protocols exclusively feasible on the PCD-CT scanner (0.10 and 0.08 mGy). After applying comparable iterative reconstruction algorithms, image quality was subjectively assessed by 4 radiologists. The intraclass correlation coefficient was calculated to estimate the agreement among readers. Image noise was quantified in standardized regions of interest to establish an additional quantitative criterion of image quality. The most dose-effective scan protocols for diagnostic imaging of midface trauma (PCD-CT: 1.24 mGy; EID-CT: 2.05 mGy), preoperative sinonasal anatomy (PCD-CT: 0.20 mGy; EID-CT: 0.40 mGy), and acute rhinosinusitis (PCD-CT: 0.08 mGy; EID-CT: 0.15 mGy) required less radiation exposure on the PCD-CT system ( P &lt; 0.050). Despite higher image noise, ultra-low-dose PCD-CT studies (0.08 and 0.10 mGy) were considered suitable for inflammation-focused imaging, offering lower-dose penalties than EID-CT studies. Interobserver reliability for subjective image quality was excellent (intraclass correlation coefficient, 0.90; 95% confidence interval, 0.88-0.93; P &lt; 0.001). In paranasal sinus imaging with tin prefiltration, the PCD-CT allowed for superior image quality compared with high-end EID-CT. Assessment of paranasal sinuses with an ultra-low radiation exposure of 0.08 mGy was deemed adequate, suggesting substantial dose reduction potential for clinical routine, for example, in the diagnostic workup of patients with rhinosinusitis.

Image Quality Evaluation in Dual-Energy CT of the Chest, Abdomen, and Pelvis in Obese Patients With Deep Learning Image Reconstruction.

The aim of this study was to evaluate image quality in vascular and oncologic dual-energy computed tomography (CT) imaging studies performed with a deep learning (DL)-based image reconstruction algorithm in patients with body mass index of ≥30. Vascular and multiphase oncologic staging dual-energy CT examinations were evaluated. Two image reconstruction algorithms were applied to the dual-energy CT data sets: standard of care Adaptive Statistical Iterative Reconstruction (ASiR-V) and TrueFidelity DL image reconstruction at 2 levels (medium and high). Subjective quality criteria were independently evaluated by 4 abdominal radiologists, and interreader agreement was assessed. Signal-to-noise ratio (SNR) and contrast-to-noise ratio were compared between image reconstruction methods. Forty-eight patients were included in this study, and the mean patient body mass index was 39.5 (SD, 7.36). TrueFidelity-High (DL-High) and TrueFidelity-Medium (DL-Med) image reconstructions showed statistically significant higher Likert scores compared with ASiR-V across all subjective image quality criteria ( P < 0.001 for DL-High vs ASiR-V; P < 0.05 for DL-Med vs ASiR-V), and SNRs for aorta and liver were significantly higher for DL-High versus ASiR-V ( P < 0.001). Contrast-to-noise ratio for aorta and SNR for aorta and liver were significantly higher for DL-Med versus ASiR-V ( P < 0.05). TrueFidelity DL image reconstruction provides improved image quality compared with ASiR-V in dual-energy CTs obtained in obese patients.

Ferumoxtran-10-enhanced 3-T Magnetic Resonance Angiography of Pelvic Arteries: Initial Experience

BackgroundPatients with renal impairment cannot undergo angiography because iodine and gadolinium contrast agents are contraindicated. Iron-containing ultrasmall superparamagnetic iron oxide particles, such as ferumoxtran-10, are not contraindicated in these patients. Thus, patients with renal failure can still undergo angiography with ferumoxtran-10. ObjectiveTo evaluate the visibility of pelvic vessels with magnetic resonance angiography (MRA) using ferumoxtran-10 as contrast agent. Design, setting, and participantsThree hundred and eighty-one patients diagnosed with primary or recurrent prostate cancer underwent pelvic ferumoxtran-10 MRA. Eleven anatomical pelvic-vessel segments per patient were evaluated using qualitative and quantitative criteria for image quality (IQ), vessel visibility (VV), and the contrast-to-noise ratio (CNR). InterventionFerumoxtran-10-enhaced MRA. Outcome measurements and statistical analysisIQ, VV, and CNR were assessed on a 5-point scale for each data set/vessel segment (very poor, poor, moderate, good, and excellent). Results and limitationsIQ was good to excellent for 98.2% of the data sets and VV was good to excellent for 97.7% of all vessel segments. The mean CNR for all segments was 88.13 (standard deviation 4.22). Contrast bolus imaging cannot be performed with this technique, so it is impossible to visualize the arterial or venous phase separately. The timing of contrast administration is also a limitation, with MRA performed 1 d after contrast infusion. ConclusionsFerumoxtran-10 MRA showed excellent image quality and visibility for pelvic vessels. In addition, the homogeneity of the intraluminal contrast was superior. Patients with preterminal or terminal renal function can benefit from ferumoxtran-10 MRA if visualization of their pelvic vessels is required. Patient summaryMagnetic resonance imaging of blood vessels using a contrast agent called ferumoxtran-10 is a promising technique for patients with impaired kidney function, as it provides high-quality visualization of blood vessels in the pelvis.

Evaluation of axial gradient Echo spiral MRI of the spine at 1.5 T

Axial gradient echo T2*-weighed MRI of the spine is a valuable diagnostic tool with several advantages over axial T2-weighted TSE MRI, but it suffers from a low signal-to-noise ratio (SNR) and inconsistent image quality. This work investigates the potential of spiral MRI to reduce artifacts and produce improved SNR and image quality in axial T2*-weighted gradient echo MRI of the spine of pediatric patients. For the purposes of image quality evaluation, 15 pediatric patients were recruited among those scheduled for a routine spine or brain exam at 1.5 T. Pediatric spine images were rated by three pediatric neuroradiologists on a subjective scale of 1–5 using four image quality criteria. Image quality scores were evaluated using non-parametric Wilcoxon signed-rank testing and a mixed effects logistic regression model. Significant differences were found in the image quality scores in favor of spiral MRI. The odds of spiral images receiving an overall image quality score higher than 3 was 16.3 times greater than the odds of Cartesian images receiving a score higher than 3 (p < 0.001, 95% CI of 4.6 to 86) as calculated using a mixed effects logistic regression model. A quantitative comparison was also performed on a single volunteer to illustrate the SNR benefit of spiral MRI. In conclusion, spiral MRI was found to provide equal or better image quality than Cartesian MRI in axial T2*-weighted gradient echo MRI in the spine of a small cohort of pediatric patients at 1.5 T.

An assessment of image reject rates for digital radiography in Saudi Arabia: A cross-sectional study

Repeat radiographs are necessary to obtain the required diagnostic information, which increases the patient's radiation dose and detracts from the “as low as reasonably achievable” (ALARA) principle. The aim of this study was to analyse the reject rate for digital radiography, causes of repetition, the most commonly repeated types of examinations for fixed and portable radiography and reject rate variations between different technologists. The data were collected from major medical imaging departments in Saudi Arabia (n = 27,238 x-ray images). Of the 27,238 images acquired, 2,682 were rejected, resulting in an overall reject rate of 8.96%. Remarkably, when rejected images of individual anatomical areas were evaluated, some areas were observed to have reject rates as high as 20% and 66% for fixed and portable digital radiography, respectively. For fixed and portable radiography, pelvic and lumbar spine procedures had the highest reject rates, respectively. Additionally, the projections with the highest reject rates for fixed and portable radiography were the shoulder axial and lumbar spine cross-table lateral projections, respectively. This study recommends initiating more consistent criteria for image quality along with a feedback scheme within the medical imaging departments to guide internal education programs.

A survey of deep learning approaches to image restoration

In this paper, we present an extensive review on deep learning methods for image restoration tasks. Deep learning techniques, led by convolutional neural networks, have received a great deal of attention in almost all areas of image processing, especially in image classification. However, image restoration is a fundamental and challenging topic and plays significant roles in image processing, understanding and representation. It typically addresses image deblurring, denoising, dehazing and super-resolution. There are substantial differences in the approaches and mechanisms in deep learning methods for image restoration. Discriminative learning based methods are able to deal with issues of learning a restoration mapping function effectively, while optimisation models based methods can further enhance the performance with certain learning constraints. In this paper, we offer a comparative study of deep learning techniques in image denoising, deblurring, dehazing, and super-resolution, and summarise the principles involved in these tasks from various supervised deep network architectures, residual or skip connection and receptive field to unsupervised autoencoder mechanisms. Image quality criteria are also reviewed and their roles in image restoration are assessed. Based on our analysis, we further present an efficient network for deblurring and a couple of multi-objective training functions for super-resolution restoration tasks. The proposed methods are compared extensively with the state-of-the-art methods with both quantitative and qualitative analyses. Finally, we point out potential challenges and directions for future research.

Synthesis of freeform optical surfaces using neural networks

Subject of study. Developed methods for synthesis of freeform illumination optics (ellipsoid, Monge-Ampere, string, and flow line methods) do not contain a correct description of the transformation of input and output wavefronts determined by the wave aberration function, do not consider the image quality criteria, and are not suitable for the calculation of imaging systems. The most common methods for synthesis of freeform elements of imaging systems (simultaneous multiple surface and ray-mapping methods) are also not ubiquitous because they are aimed at realization of a specific system configuration and, at least, require modification to solve a specific task. The largest complication in the synthesis stages of systems of imaging optics is the determination of the initial system scheme. Design of a method for synthesis of optical systems that does not require a starting point is a highly relevant objective of applied optics. A method allowing the synthesis of freeform optical surfaces for imaging systems using an artificial neural network is presented in this paper. Method. A neural network was built upon radial-basis functions and uses the geometric mapping method to generate the coordinates of the desired optical surface during backward ray tracing. The neural network was trained on different numbers of optical systems, which were automatically generated with variations in structural parameters and positions of the object and image. Main results. The method was demonstrated to operate effectively using more than 17,000 training systems at the ray set dimension of M=4096. The model of a flat-hyperbolic lens calculated using this method has a wave aberration of W&lt;5×10−2 for wavelengths within the diffraction tolerance. Practical significance. The developed method can be used to design high-performance imaging systems.

THEORETICAL FUNDAMENTALS OF THE CALCULATING THREE-MIRROR DECENTER ANASTIGMATS METHODOLOGY

The article is devoted to the theory of calculating mirror systems with anastigmatic properties, namely, the area of research in terms of developing methods for parametric calculation of dimensions and aberration correction. The such systems can correct three third-order aberrations. Mirror anastigmats allow developing the angular field of view of devices while maintaining a high numerical aperture, which allows them to be used in optoelectronic equipment operating in a wide spectral range. Complete absence of chromatic aberrations, high resolution, permissible wave criteria for image quality provide excellent opportunities for using mirror anastigmatic systems. General methodological approaches have been developed that can be applied to the creation of detailed engineering and technical methods for calculating a group of mirror anastigmatic systems. A serious drawback of reflective optics is center without central screening, which degrades image quality. To eliminate it, rotations or displacements of the mirrors are intro-duced, but non-elementary aberrations of even orders appear, which must be corrected. The creation of compositions with decentered catoptric elements requires further development of the calculation and methodological base. Mathematical solutions to the problem of creating basic models of non-centered mirror systems are presented. Accurate formulas are obtained for the calculation of real rays from the conditions of astigmatism and coma correction for the given angles of incidence of the chief ray on the mirror surfaces and the «oblique» thickness , which determines their relative position. Based on the proposed formulas, a new method for parametric calculation of decentered mirror systems has been created, which allows one to compose algorithms and design both basic models and complex mirror systems from off-axis mirrors. The development of new algorithms for two- and three-mirror decenter lenses will increase the accumulated potential of computational optics. The scope of the proposed technique can be expanded in terms of the number of components.

Towards an image quality criterion to optimize Digital image correlation. Use of an analytical model to optimize acquisition conditions

Digital image correlation (DIC) is expanding in many fields. In particular, it requires that the acquisition conditions and speckle be adapted to each application. A new methodology allowing to define the quality of the images in a fast and precise way is proposed. An analytical model based on mean intensity gradient of speckle pattern criterion and the standard deviation of displacement on images has been proposed. The model can be used on the whole image or by creating subsets in the image. Standard deviation maps of displacement are obtained. The methodology makes it possible to save a considerable amount of time compared to the full DIC calculation to optimize the image acquisition conditions, the speckle and the DIC parameters.

Sorted L1 regularization method for damage detection based on electrical impedance tomography.

Carbon fiber reinforced polymers (CFRPs) are composite materials in which carbon provides strength and stiffness, whereas polymers provide cohesiveness and toughness. The electrical impedance of CFRP laminates is changed due to different kinds of damages. Electrical impedance tomography (EIT) has significant advantages such as non-intrusion, portability, low cost, and quick response and has widely been used as a nondestructive testing method. Therefore, EIT has great potential in structural health monitoring of CFRPs. Regularization can solve the ill-posed inverse problem of EIT. However, conventional regularization algorithms have their own limitations, such as over-smoothness of reconstructed edges and unstable solution caused by measurement noise. In addition, the anisotropic property of CFRPs also affects the image quality based on traditional methods. In this paper, the sorted L1-norm regularization is proposed. It promotes grouping highly correlated variables while encouraging sparsity by using more effective penalty terms. The sharp edges between different materials can be obtained, and the obtained solution is more stable. The image quality of different objects, especially the image quality of multi-targets, can be significantly improved with this new method. In addition, the sorted L1 norm can generate adaptive regularization parameters without empirical selection. The new regularization problem is solved by the alternating direction method of multipliers. Both experimental and simulation results demonstrate that the sorted L1 norm improves the quality of reconstructed images under various noise levels. The proposed method is comprehensively evaluated with three image quality criteria by numerical simulation quantitatively.

The effect of tin prefiltration on extremity cone-beam CT imaging with a twin robotic X-ray system

IntroductionWhile tin prefiltration is established in various CT applications, its value in extremity cone-beam CT relative to optimized spectra has not been thoroughly assessed thus far. This study aims to investigate the effect of tin filters in extremity cone-beam CT with a twin-robotic X-ray system. MethodsWrist, elbow and ankle joints of two cadaveric specimens were examined in a laboratory setup with different combinations of prefiltration (copper, tin), tube voltage and current–time product. Image quality was assessed subjectively by five radiologists with Fleiss’ kappa being computed to measure interrater agreement. To provide a semiquantitative criterion for image quality, contrast-to-noise ratios (CNR) were compared for standardized regions of interest. Volume CT dose indices were calculated for a 16 cm polymethylmethacrylate phantom. ResultsRadiation dose ranged from 17.4 mGy in the clinical standard protocol without tin filter to as low as 0.7 mGy with tin prefiltration. Image quality ratings and CNR for tin-filtered scans with 100 kV were lower than for 80 kV studies with copper prefiltration despite higher dose (11.2 and 5.6 vs. 4.5 mGy; p < 0.001). No difference was ascertained between 100 kV scans with tin filtration and 60 kV copper-filtered scans with 75% dose reduction (subjective: p = 0.101; CNR: p = 0.706). Fleiss’ kappa of 0.597 (95% confidence interval 0.567–0.626; p < 0.001) indicated moderate interrater agreement. ConclusionConsiderable dose reduction is feasible with tin prefiltration, however, the twin-robotic X-ray system's low-dose potential for extremity 3D imaging is maximized with a dedicated low-kilovolt scan protocol in situations without extensive beam-hardening artifacts. Implications for practiceLow-kilovolt imaging with copper prefiltration provides a superior trade-off between dose reduction and image quality compared to tin-filtered cone-beam CT scan protocols with higher tube voltage.

Harmonisation of imaging protocols, radiation doses and image quality in gastrointestinal fluoroscopy examinations - multicentre study.

Paediatric gastrointestinal fluoroscopy examinations can impart varying amounts of radiation for the same patient size and exam type. To investigate the variability of imaging protocol, radiation dose and image quality in paediatric fluoroscopy examinations in order to provide recommendations for the harmonisation and optimisation of local practices. Five paediatric radiology departments performing fluoroscopically-guided contrast enema, micturating cystourethrography and upper gastrointestinal tract examinations participated in this study. Information on imaging protocols and radiation doses was retrospectively collected for more than 2,400 examinations. Image quality was analysed on clinical and phantom images. Patient doses showed great variability among centers with up to a factor of 5 for similar fluoroscopy times. The five departments had imaging protocols with major differences in fluoroscopy dose regulation curves and additional filtration. Image quality analysis on phantoms and patients images showed no major improvement in contrast, spatial resolution or noise when increasing the radiation dose. Age-based diagnostic reference levels using both dose area product and fluoroscopy time were proposed per procedure type. Disparities between centers and no correlation of radiation dose with image quality criteria create margins for optimisation. These results highlight the need for guidelines on fluoroscopy image quality and dose reference levels in paediatric gastrointestinal examinations to harmonise practices and optimise patient dose.

Detection and removal of dust artifacts in retinal images via sparse-based inpainting

Dust particle artifacts are present in all imaging modalities but have more adverse consequences in medical images like retinal images. They could be mistaken as small lesions, such as microaneurysms. We propose a method for detecting and accurately segmenting dust artifacts in retinal images based on multi-scale template-matching on several input images and an iterative segmentation via an inpainting approach. The inpainting is done through dictionary learning and sparse-based representation. The artifact segmentation is refined by comparing the original image to the initial restoration. On average, 90% of the dust artifacts were detected in the test images, with state-of-theart restoration results. All detected artifacts were accurately segmented and removed. Even the most challenging artifacts located on top of blood vessels were removed. Thus, ensuring the continuity of the retinal structures. The proposed method successfully detects and removes dust artifacts in retinal images, which could be used to avoid false-positive lesion detections or as an image quality criterion. An implementation of the proposed algorithm can be accessed and executed through a Code Ocean compute capsule

Modular deep neural networks for automatic quality control of retinal optical coherence tomography scans

Retinal optical coherence tomography (OCT) with intraretinal layer segmentation is increasingly used not only in ophthalmology but also for neurological diseases such as multiple sclerosis (MS). Signal quality influences segmentation results, and high-quality OCT images are needed for accurate segmentation and quantification of subtle intraretinal layer changes. Among others, OCT image quality depends on the ability to focus, patient compliance and operator skills. Current criteria for OCT quality define acceptable image quality, but depend on manual rating by experienced graders and are time consuming and subjective. In this paper, we propose and validate a standardized, grader-independent, real-time feedback system for automatic quality assessment of retinal OCT images. We defined image quality criteria for scan centering, signal quality and image completeness based on published quality criteria and typical artifacts identified by experienced graders when inspecting OCT images. We then trained modular neural networks on OCT data with manual quality grading to analyze image quality features. Quality analysis by a combination of these trained networks generates a comprehensive quality report containing quantitative results. We validated the approach against quality assessment according to the OSCAR-IB criteria by an experienced grader. Here, 100 OCT files with volume, circular and radial scans, centered on optic nerve head and macula, were analyzed and classified. A specificity of 0.96, a sensitivity of 0.97 and an accuracy of 0.97 as well as a Matthews correlation coefficient of 0.93 indicate a high rate of correct classification. Our method shows promising results in comparison to manual OCT grading and may be useful for real-time image quality analysis or analysis of large data sets, supporting standardized application of image quality criteria.

Accelerating Whole-Body Diffusion-weighted MRI with Deep Learning–based Denoising Image Filters

PurposeTo use deep learning to improve the image quality of subsampled images (number of acquisitions = 1 [NOA1]) to reduce whole-body diffusion-weighted MRI (WBDWI) acquisition times.Materials and MethodsBoth retrospective and prospective patient groups were used to develop a deep learning–based denoising image filter (DNIF) model. For initial model training and validation, 17 patients with metastatic prostate cancer with acquired WBDWI NOA1 and NOA9 images (acquisition period, 2015–2017) were retrospectively included. An additional 22 prospective patients with advanced prostate cancer, myeloma, and advanced breast cancer were used for model testing (2019), and the radiologic quality of DNIF-processed NOA1 (NOA1-DNIF) images were compared with NOA1 images and clinical NOA16 images by using a three-point Likert scale (good, average, or poor; statistical significance was calculated by using a Wilcoxon signed ranked test). The model was also retrained and tested in 28 patients with malignant pleural mesothelioma (MPM) who underwent lung MRI (2015–2017) to demonstrate feasibility in other body regions.ResultsThe model visually improved the quality of NOA1 images in all test patients, with the majority of NOA1-DNIF and NOA16 images being graded as either “average” or “good” across all image-quality criteria. From validation data, the mean apparent diffusion coefficient (ADC) values within NOA1-DNIF images of bone disease deviated from those within NOA9 images by an average of 1.9% (range, 1.1%–2.6%). The model was also successfully applied in the context of MPM; the mean ADCs from NOA1-DNIF images of MPM deviated from those measured by using clinical-standard images (NOA12) by 3.7% (range, 0.2%–10.6%).ConclusionClinical-standard images were generated from subsampled images by using a DNIF.Keywords: Image Postprocessing, MR-Diffusion-weighted Imaging, Neural Networks, Oncology, Whole-Body Imaging, Supervised Learning, MR-Functional Imaging, Metastases, Prostate, Lung Supplemental material is available for this article. Published under a CC BY 4.0 license.