Standard Image Quality Metrics Research Articles

The effect of head motion on brain age prediction using deep convolutional neural networks

Deep learning can be used effectively to predict participants' age from brain magnetic resonance imaging (MRI) data, and a growing body of evidence suggests that the difference between predicted and chronological age—referred to as brain-predicted age difference (brain-PAD)—is related to various neurological and neuropsychiatric disease states. A crucial aspect of the applicability of brain-PAD as a biomarker of individual brain health is whether and how brain-predicted age is affected by MR image artifacts commonly encountered in clinical settings. To investigate this issue, we trained and validated two different 3D convolutional neural network architectures (CNNs) from scratch and tested the models on a separate dataset consisting of motion-free and motion-corrupted T1-weighted MRI scans from the same participants, the quality of which were rated by neuroradiologists from a clinical diagnostic point of view. Our results revealed a systematic increase in brain-PAD with worsening image quality for both models. This effect was also observed for images that were deemed usable from a clinical perspective, with brains appearing older in medium than in good quality images. These findings were also supported by significant associations found between the brain-PAD and standard image quality metrics indicating larger brain-PAD for lower-quality images. Our results demonstrate a spurious effect of advanced brain aging as a result of head motion and underline the importance of controlling for image quality when using brain-predicted age based on structural neuroimaging data as a proxy measure for brain health.

Image Super-Resolution using Convolutional Neural Networks

Image super-resolution is the process of enhancing the resolution of an image, typically from a lower resolution input to a higher resolution output. This research aims to explore the application of convolutional neural networks (CNNs) for image super-resolution. Specifically, the study will focus on developing a deep learning model capable of generating high-resolution images from low-resolution inputs. Various CNN architectures, such as SRCNN (Super-Resolution Convolutional Neural Network) or SRGAN (Super-Resolution Generative Adversarial Network), will be investigated and compared for their effectiveness in producing visually pleasing and perceptually accurate high-resolution images. Additionally, techniques such as residual learning, attention mechanisms, and adversarial training may be incorporated to further improve the quality of super-resolved images. The performance of the proposed models will be evaluated using standard image quality metrics and subjective assessments. This research has practical applications in enhancing the visual quality of low-resolution images in fields such as medical imaging, surveillance, and entertainment

Enhanced range-doppler algorithm for SAR image formation

A Synthetic Aperture Radar (SAR) system is a powerful source of information due to its capability to operate day and night and in all weather conditions. It is essential for military and civilian purposes. Pulse-focusing is employed by the SAR system for achieving both long-range detection and fine-range resolution. The Range-Doppler Algorithm (RDA) is the most popular pulse-focusing technique for creating images from coherent SAR data. Pulse compression techniques for SAR signal processing targets seek to improve map resolution, lower peak power, and increase the Signal-to-Noise Ratio (SNR) of the detected object. This paper demonstrates an enhanced RDA algorithm that uses a modified pulse compression technique based on mismatched filter optimization where Quadratic Constrained Quadratic Program (QCQP) is used. The performance evaluation of the proposed method is performed using visual assessment and the standard image quality metrics peak-side-lobe-ratio (PSLR), impulse-response-width (IRW), and integrated-side-lobe-ratio (ISLR). Simulated and real SAR raw data are used for performance evaluation. The experimental results demonstrate the proposed method’s robustness, efficiency, and reliability in providing a higher-quality SAR image than the traditional RDA algorithm.

Accelerating High b-Value Diffusion-Weighted MRI Using a Convolutional Recurrent Neural Network (CRNN-DWI).

To develop a novel convolutional recurrent neural network (CRNN-DWI) and apply it to reconstruct a highly undersampled (up to six-fold) multi-b-value, multi-direction diffusion-weighted imaging (DWI) dataset. A deep neural network that combines a convolutional neural network (CNN) and recurrent neural network (RNN) was first developed by using a set of diffusion images as input. The network was then used to reconstruct a DWI dataset consisting of 14 b-values, each with three diffusion directions. For comparison, the dataset was also reconstructed with zero-padding and 3D-CNN. The experiments were performed with undersampling rates (R) of 4 and 6. Standard image quality metrics (SSIM and PSNR) were employed to provide quantitative assessments of the reconstructed image quality. Additionally, an advanced non-Gaussian diffusion model was employed to fit the reconstructed images from the different approaches, thereby generating a set of diffusion parameter maps. These diffusion parameter maps from the different approaches were then compared using SSIM as a metric. Both the reconstructed diffusion images and diffusion parameter maps from CRNN-DWI were better than those from zero-padding or 3D-CNN. Specifically, the average SSIM and PSNR of CRNN-DWI were 0.750 ± 0.016 and 28.32 ± 0.69 (R = 4), and 0.675 ± 0.023 and 24.16 ± 0.77 (R = 6), respectively, both of which were substantially higher than those of zero-padding or 3D-CNN reconstructions. The diffusion parameter maps from CRNN-DWI also yielded higher SSIM values for R = 4 (>0.8) and for R = 6 (>0.7) than the other two approaches (for R = 4, <0.7, and for R = 6, <0.65). CRNN-DWI is a viable approach for reconstructing highly undersampled DWI data, providing opportunities to reduce the data acquisition burden.

Validation of deep learning techniques for quality augmentation in diffusion MRI for clinical studies

The objective of this study is to evaluate the efficacy of deep learning (DL) techniques in improving the quality of diffusion MRI (dMRI) data in clinical applications. The study aims to determine whether the use of artificial intelligence (AI) methods in medical images may result in the loss of critical clinical information and/or the appearance of false information. To assess this, the focus was on the angular resolution of dMRI and a clinical trial was conducted on migraine, specifically between episodic and chronic migraine patients. The number of gradient directions had an impact on white matter analysis results, with statistically significant differences between groups being drastically reduced when using 21 gradient directions instead of the original 61. Fourteen teams from different institutions were tasked to use DL to enhance three diffusion metrics (FA, AD and MD) calculated from data acquired with 21 gradient directions and a b-value of 1000 s/mm2. The goal was to produce results that were comparable to those calculated from 61 gradient directions. The results were evaluated using both standard image quality metrics and Tract-Based Spatial Statistics (TBSS) to compare episodic and chronic migraine patients. The study results suggest that while most DL techniques improved the ability to detect statistical differences between groups, they also led to an increase in false positive. The results showed that there was a constant growth rate of false positives linearly proportional to the new true positives, which highlights the risk of generalization of AI-based tasks when assessing diverse clinical cohorts and training using data from a single group. The methods also showed divergent performance when replicating the original distribution of the data and some exhibited significant bias. In conclusion, extreme caution should be exercised when using AI methods for harmonization or synthesis in clinical studies when processing heterogeneous data in clinical studies, as important information may be altered, even when global metrics such as structural similarity or peak signal-to-noise ratio appear to suggest otherwise.

Movement-related artefacts (MR-ART) dataset of matched motion-corrupted and clean structural MRI brain scans

Magnetic Resonance Imaging (MRI) provides a unique opportunity to investigate neural changes in healthy and clinical conditions. Its large inherent susceptibility to motion, however, often confounds the measurement. Approaches assessing, correcting, or preventing motion corruption of MRI measurements are under active development, and such efforts can greatly benefit from carefully controlled datasets. We present a unique dataset of structural brain MRI images collected from 148 healthy adults which includes both motion-free and motion-affected data acquired from the same participants. This matched dataset allows direct evaluation of motion artefacts, their impact on derived data, and testing approaches to correct for them. Our dataset further stands out by containing images with different levels of motion artefacts from the same participants, is enriched with expert scoring characterizing the image quality from a clinical point of view and is also complemented with standard image quality metrics obtained from MRIQC. The goal of the dataset is to raise awareness of the issue and provide a useful resource to assess and improve current motion correction approaches.

A novel family of edge preserving anisotropic filters

In this article, new anisotropic image filters are introduced and their performances are compared with existing isotropic and anisotropic filters. The developed filters were examined according to their image noise removing performances in terms of standard image quality metrics, as well as the edge preserving properties of the filtered images. Mathematical inferences of anisotropic filters are made based on the minimization of the Polyakov action energy integral. New anisotropic metrics are found by means of Finsler metrics that minimize the corresponding integral. The new metrics perform filtering by updating the image with anisotropic Laplace-Beltrami flow. After filtering, it was observed that the introduced metrics perform well against other anisotropic metrics. It is also observed that the developed New Randers, New Normalized Miron, and New Metric filters preserve edges better than other filters, making it a plausible noise removal tool prior to edge detection in image processing. The source codes of proposed filters are publicly available at https://github.com/HAYDARKILIC.

A novel complex-valued convolutional neural network for medical image denoising

Several applications of complex-valued networks have been reported for computer vision tasks like image processing and classification. However, complex-valued convolutional neural networks are yet to be explored for medical image denoising. In this paper, a novel complex-valued convolutional neural network-based model, termed as CVMIDNet, is investigated for medical image denoising. Unlike traditional approaches, which learn clean images from noisy images, the proposed model utilizes residual learning, which learns noise from noisy images and then subtracts it from noisy images so as to obtain clean images. To assess the denoising performance of CVMIDNet, standard image quality metrics, namely, peak signal to noise ratio and the structural similarity index, have been used for five different additive white Gaussian noise levels in chest X-ray images. Chest X-ray denoising performance of CVMIDNet was compared with four recent state-of-the-art models, namely, Block-Matching and 3D (BM3D) filtering, DnCNN, Feature-guided Denoising Convolutional Neural Network (FDCNN), and deep CNN with residual learning. Additionally, it is also benchmarked against its real-valued counterpart, termed RVMIDNet. In all the reported performance investigations, CVMIDNet was found to be superior. For instance, for a Gaussian noise level of σ = 15, peak signal to noise ratio and structural similarity index values achieved by the CVMIDNet are 37.2010 and 0.9227, respectively, against the 36.2292 and 0.9086, 36.3203 and 0.9139, 35.0995 and 0.9005, 36.1830 and 0.8968, 34.2436 and 0.8874 achieved by BM3D filtering, DnCNN, RVMIDNet, FDCNN, and deep CNN with residual learning, respectively. Therefore, based on the presented investigations, it is concluded that CVMIDNet is a potential deep learning model for medical image denoising.

Efficient radiation dose reduction in whole-brain CT perfusion imaging using a 3D GAN: performance and clinical feasibility

Dose reduction in cerebral CT perfusion (CTP) imaging is desirable but is accompanied by an increase in noise that can compromise the image quality and the accuracy of image-based haemodynamic modelling used for clinical decision support in acute ischaemic stroke. The few reported methods aimed at denoising low-dose CTP images lack practicality by considering only small sections of the brain or being computationally expensive. Moreover, the prediction of infarct and penumbra size and location—the chief means of decision support for treatment options—from denoised data has not been explored using these approaches. In this work, we present the first application of a 3D generative adversarial network (3D GAN) for predicting normal-dose CTP data from low-dose CTP data. Feasibility of the approach was tested using real data from 30 acute ischaemic stroke patients in conjunction with low dose simulation. The 3D GAN model was applied to 643 voxel patches extracted from two different configurations of the CTP data—frame-based and stacked. The method led to whole-brain denoised data being generated for haemodynamic modelling within 90 s. Accuracy of the method was evaluated using standard image quality metrics and the extent to which the clinical content and lesion characteristics of the denoised CTP data were preserved. Results showed an average improvement of 5.15–5.32 dB PSNR and 0.025–0.033 structural similarity index (SSIM) for CTP images and 2.66–3.95 dB PSNR and 0.036–0.067 SSIM for functional maps at 50% and 25% of normal dose using GAN model in conjunction with a stacked data regime for image synthesis. Consequently, the average lesion volumetric error reduced significantly (p-value <0.05) by 18%–29% and dice coefficient improved significantly by 15%–22%. We conclude that GAN-based denoising is a promising practical approach for reducing radiation dose in CTP studies and improving lesion characterisation.

So you think you can DAS? A viewpoint on delay-and-sum beamforming

Delay-and-sum (DAS) is the most widespread digital beamformer in high-frame-rate ultrasound imaging. Its implementation is simple and compatible with real-time applications. In this viewpoint article, we describe the fundamentals of DAS beamforming. The underlying theory and numerical approach are detailed so that users can be aware of its functioning and limitations. In particular, we discuss the importance of the f-number and speed of sound on image quality, and propose one solution to set their values from a physical viewpoint. We suggest determining the f-number from the directivity of the transducer elements and the speed of sound from the phase dispersion of the delayed signals. Simplified Matlab codes are provided for the sake of clarity and openness. The effect of the f-number and speed of sound on the lateral resolution and contrast-to-noise ratio was investigated in vitro and in vivo. If not properly preset, these two factors had a substantial negative impact on standard metrics of image quality (namely CNR and FWHM). When beamforming with DAS in vitro or in vivo, it is recommended to optimize these parameters in order to use it wisely and prevent image degradation.

Accelerating cardiac cine MRI using a deep learning-based ESPIRiT reconstruction.

To propose a novel combined parallel imaging and deep learning-based reconstruction framework for robust reconstruction of highly accelerated 2D cardiac cine MRI data. We propose DL-ESPIRiT, an unrolled neural network architecture that utilizes an extended coil sensitivity model to address SENSE-related field-of-view (FOV) limitations in previously proposed deep learning-based reconstruction frameworks. Additionally, we propose a novel neural network design based on (2+1)D spatiotemporal convolutions to produce more accurate dynamic MRI reconstructions than conventional 3D convolutions. The network is trained on fully sampled 2D cardiac cine datasets collected from 11 healthy volunteers with IRB approval. DL-ESPIRiT is compared against a state-of-the-art parallel imaging and compressed sensing method known as -ESPIRiT. The reconstruction accuracy of both methods is evaluated on retrospectively undersampled datasets (R=12) with respect to standard image quality metrics as well as automatic deep learning-based segmentations of left ventricular volumes. Feasibility of DL-ESPIRiT is demonstrated on two prospectively undersampled datasets acquired in a single heartbeat per slice. The (2+1)D DL-ESPIRiT method produces higher fidelity image reconstructions when compared to -ESPIRiT reconstructions with respect to standard image quality metrics (P < .001). As a result of improved image quality, segmentations made from (2+1)D DL-ESPIRiT images are also more accurate than segmentations from -ESPIRiT images. DL-ESPIRiT synergistically combines a robust parallel imaging model and deep learning-based priors to produce high-fidelity reconstructions of retrospectively undersampled 2D cardiac cine data acquired with reduced FOV. Although a proof-of-concept is shown, further experiments are necessary to determine the efficacy of DL-ESPIRiT in prospectively undersampled data.

End-to-End Multitask Learning for Driver Gaze and Head Pose Estimation

The dead leaves pattern is very useful to obtain an SFR from a stochastic pattern and can be used to measure texture loss due to noise reduction or compression in images and video streams. In this paper, we present results from experiments that use the pattern and different analysis approaches to measure the dynamic range of a camera system as well as to describe the dependency of the SFR on object contrast and light intensity. The results can be used to improve the understanding of the performance of modern camera systems. These systems work adaptively and are scene aware but are not well described by standard image quality metrics.

Characterization of a prototype rapid kilovoltage x-ray image guidance system designed for a ring shape radiation therapy unit.

This study aims to characterize the performance of a prototype rapid kilovoltage (kV) x-ray image guidance system onboard the newly released Halcyon 2.0 linear accelerator (Varian Medical Systems, Palo Alto, CA) by use of conventional and innovatively designed testing procedures. Basic imaging system performance tests and radiation dose measurements were performed for all eleven kV-cone beam computed tomography (CBCT) imaging protocols available on a preclinical Halcyon 2.0 LINAC. Both conventional CBCT reconstruction using the Feldkamp-Davis-Kress (FDK) algorithm and a novel, advanced iterative reconstruction (iCBCT) available on this platform were evaluated. Standard image quality metrics, including slice thickness accuracy, high-contrast resolution, low-contrast resolution, regional uniformity and noise, and CT Hounsfield unit (HU) number accuracy and linearity were evaluated using a manufacturer-supplied QUART phantom (GmbH, Zorneding, Germany) and an independent image quality phantom (Catphan 500, The Phantom Laboratory, New York, NY). Due to the simplified design of the QUART phantom, we developed surrogate and clinically feasible strategies for measuring slice thickness and high- and low-contrast resolution. Imaging dose delivered by these eleven protocols was measured using a computed tomography dose index phantom and pencil chamber with commonly accepted methods and procedures. A subset of measurements were repeated on a conventional C-arm LINAC (TrueBeam and Trilogy, Varian Medical System) for comparison. Clinical patient images of pelvic and abdominal regions are also presented for qualitative assessment as part of a feasibility study for clinical implementation. Image acquisition time was 17-42s on the Halcyon system compared with 60s on the C-arm LINAC systems. The kV imager projection offset, imaging and treatment isocenter coincidence and the couch three-dimensional match movement all achieved less than1mm mechanical accuracy. All major image quality metrics were within either the national guideline or vendor-recommended tolerances. The designed surrogate approach with the QUART phantom showed a range of 0.24-0.35cycles/mm for spatial resolution, a contrast-to-noise ratio (CNR) of 2-20 for FDK reconstruction and a tolerance of 0.5mm for slice thickness. Other metrics derived from the Catphan images obtained on the Halcyon and C-arm LINACs showed comparable values for the FDK reconstruction. The iterative reconstruction tended to reduce noise, as evidenced by a higher CNR ratio. The fast scan pelvis protocols for Halcyon resulted in 50% lower dose compared to the standard scans, and the thorax fast protocol similarly delivered 10% lower dose than the standard thoracic scan. Preliminary patient images indicated that rapid kV CBCT with breath-hold is feasible, with improved imaging quality compared to free-breathing scans. Independent and comprehensive characterization of the kV imaging guidance system on the Halcyon 2.0 system demonstrated acceptable image quality for clinical use. The imaging unit onboard the Halcyon meets vendor specifications and satisfies requirements for routine clinical use. The fast kV imaging system enables the potential for volumetric CBCT acquisition during a single breath-hold and the iterative reconstruction tends to reduce the noise therefore has the potential to improve the CNR for normal size patient.

A new quality assessment for Thangka image inpainting

SummaryThis paper presents an efficient metric for evaluation the effect of the inpainted Thangka images. In contrast to standard image quality metrics, the proposed one takes into account some constraints and characteristics related to the specific goals of inpainting techniques. The key is that we proposed a method to decompose the reference and distorted image and to compare the intensity of fuzzy edge for structure component and similarity for texture component. By comparative analyzing the experimental results with other metrics, the proposed image inpainting quality index delivers high consistency with subjective evaluation results, which will help the selecting algorithm in image restoration and could be also applied to most of inpainting image approaches.

Spatially adaptive total variation deblurring with split Bregman technique

In this study, the authors describe a modified non-blind and blind deconvolution model by introducing a regularisation parameter that incorporates the spatial image information. Indeed, they have used a weighted total variation term, where the weight is a spatially adaptive parameter based on the image gradient. The proposed models are solved by the split Bregman method. To handle adequately the discrete convolution transform in a moderate time, fast Fourier transform is used. Tests are conducted on several images, and for assessing the results, they define appropriate weighted versions of two standard image quality metrics. These new weighted metrics clearly highlight the advantage of the spatially adaptive approach.

Disparity refinement process based on RANSAC plane fitting for machine vision applications

This paper presents a new disparity map refinement process for stereo matching algorithm and the refinement stage that will be implemented by partitioning the place or mask image and re-projected to the preliminary disparity images. This process is to refine the noise and sparse of initial disparity map from weakly textured. The plane fitting algorithm is using Random Sample Consensus. Two well-known stereo matching algorithms have been tested on this framework with different filtering techniques applied at disparity refinement stage. The framework is evaluated on three Middlebury datasets. The experimental results show that the proposed framework produces better-quality and more accurate than normal flow state-of-the-art stereo matching algorithms. The performance evaluations are based on standard image quality metrics i.e. structural similarity index measure, peak signal-to-noise ratio and mean square error.

Medical image denoising using convolutional neural network: a residual learning approach

In medical imaging, denoising is very important for analysis of images, diagnosis and treatment of diseases. Currently, image denoising methods based on deep learning are effective, where the methods are however limited for the requirement of training sample size (i.e., not successful enough for small data size). Using small sample size, we design deep feed forward denoising convolutional neural networks by studying the model in deep framework, learning approach and regularization approach for medical image denoising. More specifically, we use residual learning as a learning approach and batch normalization as regularization in the deep model. Unlike most of the other image denoising approaches which directly learn the latent clean images, the residual learning approach learns the noise from the noisy images instead of the latent clean images where the denoised images are obtained by subtracting the learned residual from the noisy image. Moreover, batch normalization is integrated with residual learning to improve model learning accuracy and training time. We compute the quality of the reconstructed or denoised image in standard image quality metrics, peak signal to noise ratio and structural similarity and compare our model performance with some medical image denoising techniques. Experimental results reveal that our approach has better performance than some other methods.

WE‐FG‐207B‐11: Objective Image Characterization of Spectral CT with a Dual‐Layer Detector

Purpose:To obtain objective reference data for the spectral performance on a dual‐layer detector CT platform (IQon, Philips) and compare virtual monoenergetic to conventional CT images.Methods:Scanning was performed using the hospital's clinical adult body protocol: helical acquisition at 120kVp, with CTDIvol=15mGy. Multiple modules (591, 515, 528) of a CATPHAN 600 phantom and a 20 cm diameter cylindrical water phantom were scanned. No modifications to the standard protocol were necessary to enable spectral imaging. Both conventional and virtual monoenergetic images were generated from acquired data. Noise characteristics were assessed through Noise Power Spectra (NPS) and pixel standard deviation from water phantom images. Spatial resolution was evaluated using Modulation Transfer Functions (MTF) of a tungsten wire as well as resolution bars. Low‐contrast detectability was studied using contrast‐to‐noise ratio (CNR) of a low contrast object.Results:MTF curves of monoenergetic and conventional images were almost identical. MTF 50%, 10%, and 5% levels for monoenergetic images agreed with conventional images within 0.05lp/cm. These observations were verified by the resolution bars, which were clearly resolved at 7lp/cm but started blurring at 8lp/cm for this protocol in both conventional and 70 keV images. NPS curves indicated that, compared to conventional images, the noise power distribution of 70 keV monoenergetic images is similar (i.e. noise texture is similar) but exhibit a low frequency peak at keVs higher and lower than 70 keV. Standard deviation measurements show monoenergetic images have lower noise except at 40 keV where it is slightly higher. CNR of monoenergetic images is mostly flat across keV values and is superior to that of conventional images.Conclusion:Values for standard image quality metrics are the same or better for monoenergetic images compared to conventional images. Results indicate virtual monoenergetic images can be used without any loss in image quality or noise penalties relative to conventional images.This study was performed as part of a research agreement among Philips Healthcare, University Hospitals of Cleveland, and Case Western Reserve University.

Monte-Carlo simulation of a slot-scanning X-ray imaging system

We present a method for simulating slot-scanning X-ray imaging using the general-purpose Monte Carlo simulation package PENELOPE and penEasy Imaging. Different phantoms can be defined with the PENGEOM package, which defines bodies as combinations of volumes limited by quadric surfaces. The source-detector geometry, the position of the object, the collimator, the X-ray tube properties, the detector material and the pixel dimensions are defined. The output of the time-delay integration detector is simulated using sequential slot outputs derived from penEasy Imaging. The simulations are validated using tungsten and aluminium test objects, which are both simulated and imaged. The simulations are compared to the X-ray images using standard image quality metrics. The MTF, NPS and DQE curves show that the real and simulated X-ray images are comparable in terms of spatial resolution, noise and frequency information. The implementation can be modified to suit alterations in the system being simulated.

Strong reflector-based beamforming in ultrasound medical imaging

This paper investigates the use of sparse priors in creating original two-dimensional beamforming methods for ultrasound imaging. The proposed approaches detect the strong reflectors from the scanned medium based on the well known Bayesian Information Criteria used in statistical modeling. Moreover, they allow a parametric selection of the level of speckle in the final beamformed image. These methods are applied on simulated data and on recorded experimental data. Their performance is evaluated considering the standard image quality metrics: contrast ratio (CR), contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR). A comparison is made with the classical delay-and-sum and minimum variance beamforming methods to confirm the ability of the proposed methods to precisely detect the number and the position of the strong reflectors in a sparse medium and to accurately reduce the speckle and highly enhance the contrast in a non-sparse medium.We confirm that our methods improve the contrast of the final image for both simulated and experimental data. In all experiments, the proposed approaches tend to preserve the speckle, which can be of major interest in clinical examinations, as it can contain useful information. In sparse mediums we achieve a highly improvement in contrast compared with the classical methods.