Rician noise removal by non-Local Means filtering for low signal-to-noise ratio MRI: applications to DT-MRI

Left Ventricle Detection from Cardiac Magnetic Resonance Relaxometry Images Using Visual Transformer

Left Ventricle (LV) detection from Cardiac Magnetic Resonance (CMR) imaging is a fundamental step, preliminary to myocardium segmentation and characterization. This paper focuses on the application of a Visual Transformer (ViT), a novel neural network architecture, to automatically detect LV from CMR relaxometry sequences. We implemented an object detector based on the ViT model to identify LV from CMR multi-echo T2* sequences. We evaluated performances differentiated by slice location according to the American Heart Association model using 5-fold cross-validation and on an independent dataset of CMR T2*, T2, and T1 acquisitions. To the best of our knowledge, this is the first attempt to localize LV from relaxometry sequences and the first application of ViT for LV detection. We collected an Intersection over Union (IoU) index of 0.68 and a Correct Identification Rate (CIR) of blood pool centroid of 0.99, comparable with other state-of-the-art methods. IoU and CIR values were significantly lower in apical slices. No significant differences in performances were assessed on independent T2* dataset (IoU = 0.68, p = 0.405; CIR = 0.94, p = 0.066). Performances were significantly worse on the T2 and T1 independent datasets (T2: IoU = 0.62, CIR = 0.95; T1: IoU = 0.67, CIR = 0.98), but still encouraging considering the different types of acquisition. This study confirms the feasibility of the application of ViT architectures in LV detection and defines a benchmark for relaxometry imaging.

Level-Set Method for Image Analysis of Schlemm's Canal and Trabecular Meshwork

Purpose

To evaluate different segmentation methods in analyzing Schlemm's canal (SC) and the trabecular meshwork (TM) in ultrasound biomicroscopy (UBM) images.

Methods

Twenty-six healthy volunteers were recruited. The intraocular pressure (IOP) was measured while study subjects blew a trumpet. Images were obtained at different IOPs by 50-MHz UBM. ImageJ software and three segmentation methods—K-means, fuzzy C-means, and level set—were applied to segment the UBM images. The quantitative analysis of the TM-SC region was based on the segmentation results. The relative error and the interclass correlation coefficient (ICC) were used to quantify the accuracy and the repeatability of measurements. Pearson correlation analysis was conducted to evaluate the associations between the IOP and the TM and SC geometric measurements.

Results

A total of 104 UBM images were obtained. Among them, 84 were adequately clear to be segmented. The level-set method results had a higher similarity to ImageJ results than the other two methods. The ICC values of the level-set method were 0.97, 0.95, 0.9, and 0.57, respectively. Pearson correlation coefficients for the IOP to the SC area, SC perimeter, SC length, and TM width were −0.91, −0.72, −0.66, and −0.61 (P < 0.0001), respectively.

Conclusions

The level-set method showed better accuracy than the other two methods. Compared with manual methods, it can achieve similar precision, better repeatability, and greater efficiency. Therefore, the level-set method can be used for reliable UBM image segmentation.

Translational Relevance

The level-set method can be used to analyze TM and SC region in UBM images semiautomatically.

Predictive prosthetic socket design: part 1-population-based evaluation of transtibial prosthetic sockets by FEA-driven surrogate modelling

It has been proposed that finite element analysis can complement clinical decision making for the appropriate design and manufacture of prosthetic sockets for amputees. However, clinical translation has not been achieved, in part due to lengthy solver times and the complexity involved in model development. In this study, a parametric model was created, informed by variation in (i) population-driven residuum shape morphology, (ii) soft tissue compliance and (iii) prosthetic socket design. A Kriging surrogate model was fitted to the response of the analyses across the design space enabling prediction for new residual limb morphologies and socket designs. It was predicted that morphological variability and prosthetic socket design had a substantial effect on socket-limb interfacial pressure and shear conditions as well as sub-dermal soft tissue strains. These relationships were investigated with a higher resolution of anatomical, surgical and design variability than previously reported, with a reduction in computational expense of six orders of magnitude. This enabled real-time predictions (1.6 ms) with error vs the analytical solutions of < 4 kPa in pressure at residuum tip, and < 3% in soft tissue strain. As such, this framework represents a substantial step towards implementation of finite element analysis in the prosthetics clinic.

Graphics processing units-accelerated adaptive nonlocal means filter for denoising three-dimensional Monte Carlo photon transport simulations

The Monte Carlo (MC) method is widely recognized as the gold standard for modeling light propagation inside turbid media. Due to the stochastic nature of this method, MC simulations suffer from inherent stochastic noise. Launching large numbers of photons can reduce noise but results in significantly greater computation times, even with graphics processing units (GPU)-based acceleration. We develop a GPU-accelerated adaptive nonlocal means (ANLM) filter to denoise MC simulation outputs. This filter can effectively suppress the spatially varying stochastic noise present in low-photon MC simulations and improve the image signal-to-noise ratio (SNR) by over 5 dB. This is equivalent to the SNR improvement of running nearly 3.5  ×   more photons. We validate this denoising approach using both homogeneous and heterogeneous domains at various photon counts. The ability to preserve rapid optical fluence changes is also demonstrated using domains with inclusions. We demonstrate that this GPU-ANLM filter can shorten simulation runtimes in most photon counts and domain settings even combined with our highly accelerated GPU MC simulations. We also compare this GPU-ANLM filter with the CPU version and report a threefold to fourfold speedup. The developed GPU-ANLM filter not only can enhance three-dimensional MC photon simulation results but also be a valuable tool for noise reduction in other volumetric images such as MRI and CT scans.

Evaluation of deep learning methods for parotid gland segmentation from CT images

The segmentation of organs at risk is a crucial and time-consuming step in radiotherapy planning. Good automatic methods can significantly reduce the time clinicians have to spend on this task. Due to its variability in shape and low contrast to surrounding structures, segmenting the parotid gland is challenging. Motivated by the recent success of deep learning, we study the use of two-dimensional (2-D), 2-D ensemble, and three-dimensional (3-D) U-Nets for segmentation. The mean Dice similarity to ground truth is ∼0.83 for all three models. A patch-based approach for class balancing seems promising for false-positive reduction. The 2-D ensemble and 3-D U-Net are applied to the test data of the 2015 MICCAI challenge on head and neck autosegmentation. Both deep learning methods generalize well onto independent data (Dice 0.865 and 0.88) and are superior to a selection of model- and atlas-based methods with respect to the Dice coefficient. Since appropriate reference annotations are essential for training but often difficult and expensive to obtain, it is important to know how many samples are needed for training. We evaluate the performance after training with different-sized training sets and observe no significant increase in the Dice coefficient for more than 250 training cases.

Geometric and dosimetric evaluations of atlas-based segmentation methods of MR images in the head and neck region

Owing to its excellent soft-tissue contrast, magnetic resonance (MR) imaging has found an increased application in radiation therapy (RT). By harnessing these properties for treatment planning, automated segmentation methods can alleviate the manual workload burden to the clinical workflow. We investigated atlas-based segmentation methods of organs at risk (OARs) in the head and neck (H&N) region using one approach that selected the most similar atlas from a library of segmented images and two multi-atlas approaches. The latter were based on weighted majority voting and an iterative atlas-fusion approach called STEPS. We built the atlas library from pre-treatment T1-weighted MR images of 12 patients with manual contours of the parotids, spinal cord and mandible, delineated by a clinician. Following a leave-one-out cross-validation strategy, we measured the geometric accuracy by calculating Dice similarity coefficients (DSC), standard and 95% Hausdorff distances (HD and HD95), and the mean surface distance (MSD), whereby the manual contours served as the gold standard. To benchmark the algorithm, we determined the inter-observer variability (IOV) between three observers. To investigate the dosimetric effect of segmentation inaccuracies, we implemented an auto-planning strategy within the treatment planning system Monaco (Elekta AB, Stockholm, Sweden). For each set of auto-segmented OARs, we generated a plan for a 9-beam step and shoot intensity modulated RT treatment, designed according to our institution's clinical H&N protocol. Superimposing the dose distributions on the gold standard OARs, we calculated dose differences to OARs caused by delineation differences between auto-segmented and gold standard OARs. We investigated the correlations between geometric and dosimetric differences. The mean DSC was larger than 0.8 and the mean MSD smaller than 2 mm for the multi-atlas approaches, resulting in a geometric accuracy comparable to previously published results and within the range of the IOV. While dosimetric differences could be as large as 23% of the clinical goal, treatment plans fulfilled all imposed clinical goals for the gold standard OARs. Correlations between geometric and dosimetric measures were low with R2  <  0.5. The geometric accuracy and the ability to achieve clinically acceptable treatment plans indicate the suitability of using atlas-based contours for RT treatment planning purposes. The low correlations between geometric and dosimetric measures suggest that geometric measures alone are not sufficient to predict the dosimetric impact of segmentation inaccuracies on treatment planning for the data utilised in this study.

Geometric and dosimetric evaluations of atlas-based segmentation methods of MR images in the head and neck region

Owing to its excellent soft-tissue contrast, magnetic resonance (MR) imaging has found an increased application in radiation therapy (RT). By harnessing these properties for treatment planning, automated segmentation methods can alleviate the manual workload burden to the clinical workflow. We investigated atlas-based segmentation methods of organs at risk (OARs) in the head and neck (H&N) region using one approach that selected the most similar atlas from a library of segmented images and two multi-atlas approaches. The latter were based on weighted majority voting and an iterative atlas-fusion approach called STEPS. We built the atlas library from pre-treatment T1-weighted MR images of 12 patients with manual contours of the parotids, spinal cord and mandible, delineated by a clinician. Following a leave-one-out cross-validation strategy, we measured the geometric accuracy by calculating Dice similarity coefficients (DSC), standard and 95% Hausdorff distances (HD and HD95), and the mean surface distance (MSD), whereby the manual contours served as the gold standard. To benchmark the algorithm, we determined the inter-observer variability (IOV) between three observers. To investigate the dosimetric effect of segmentation inaccuracies, we implemented an auto-planning strategy within the treatment planning system Monaco (Elekta AB, Stockholm, Sweden). For each set of auto-segmented OARs, we generated a plan for a 9-beam step and shoot intensity modulated RT treatment, designed according to our institution's clinical H&N protocol. Superimposing the dose distributions on the gold standard OARs, we calculated dose differences to OARs caused by delineation differences between auto-segmented and gold standard OARs. We investigated the correlations between geometric and dosimetric differences. The mean DSC was larger than 0.8 and the mean MSD smaller than 2 mm for the multi-atlas approaches, resulting in a geometric accuracy comparable to previously published results and within the range of the IOV. While dosimetric differences could be as large as 23% of the clinical goal, treatment plans fulfilled all imposed clinical goals for the gold standard OARs. Correlations between geometric and dosimetric measures were low with R²  <  0.5. The geometric accuracy and the ability to achieve clinically acceptable treatment plans indicate the suitability of using atlas-based contours for RT treatment planning purposes. The low correlations between geometric and dosimetric measures suggest that geometric measures alone are not sufficient to predict the dosimetric impact of segmentation inaccuracies on treatment planning for the data utilised in this study.

Semiautomated biventricular segmentation in three-dimensional echocardiography by coupled deformable surfaces

With the advancement of three-dimensional (3-D) real-time echocardiography in recent years, automatic creation of patient specific geometric models is becoming feasible and important in clinical decision making. However, the vast majority of echocardiographic segmentation methods presented in the literature focus on the left ventricle (LV) endocardial border, leaving segmentation of the right ventricle (RV) a largely unexplored problem, despite the increasing recognition of the RV's role in cardiovascular disease. We present a method for coupled segmentation of the endo- and epicardial borders of both the LV and RV in 3-D ultrasound images. To solve the segmentation problem, we propose an extension of a successful state-estimation segmentation framework with a geometrical representation of coupled surfaces, as well as the introduction of myocardial incompressibility to regularize the segmentation. The method was validated against manual measurements and segmentations in images of 16 patients. Mean absolute distances of [Formula: see text], [Formula: see text], and [Formula: see text] between the proposed and reference segmentations were observed for the LV endocardium, RV endocardium, and LV epicardium surfaces, respectively. The method was computationally efficient, with a computation time of [Formula: see text].

Automated quality assessment in three-dimensional breast ultrasound images

Automated three-dimensional breast ultrasound (ABUS) is a valuable adjunct to x-ray mammography for breast cancer screening of women with dense breasts. High image quality is essential for proper diagnostics and computer-aided detection. We propose an automated image quality assessment system for ABUS images that detects artifacts at the time of acquisition. Therefore, we study three aspects that can corrupt ABUS images: the nipple position relative to the rest of the breast, the shadow caused by the nipple, and the shape of the breast contour on the image. Image processing and machine learning algorithms are combined to detect these artifacts based on 368 clinical ABUS images that have been rated manually by two experienced clinicians. At a specificity of 0.99, 55% of the images that were rated as low quality are detected by the proposed algorithms. The areas under the ROC curves of the single classifiers are 0.99 for the nipple position, 0.84 for the nipple shadow, and 0.89 for the breast contour shape. The proposed algorithms work fast and reliably, which makes them adequate for online evaluation of image quality during acquisition. The presented concept may be extended to further image modalities and quality aspects.

Robust Spatio-Temporal Registration of 4D Cardiac Ultrasound Sequences

Registration of multiple 3D ultrasound sectors in order to provide an extended field of view is important for the appreciation of larger anatomical structures at high spatial and temporal resolution. In this paper, we present a method for fully automatic spatio-temporal registration between two partially overlapping 3D ultrasound sequences. The temporal alignment is solved by aligning the normalized cross correlation-over-time curves of the sequences. For the spatial alignment, corresponding 3D Scale Invariant Feature Transform (SIFT) features are extracted from all frames of both sequences independently of the temporal alignment. A rigid transform is then calculated by least squares minimization in combination with random sample consensus. The method is applied to 16 echocardiographic sequences of the left and right ventricles and evaluated against manually annotated temporal events and spatial anatomical landmarks. The mean distances between manually identified landmarks in the left and right ventricles after automatic registration were (mean ± SD) 4.3 ± 1.2 mm compared to a reference error of 2.8 ± 0.6 mm with manual registration. For the temporal alignment, the absolute errors in valvular event times were 14.4 ± 11.6 ms for Aortic Valve (AV) opening, 18.6 ± 16.0 ms for AV closing, and 34.6 ± 26.4 ms for mitral valve opening, compared to a mean inter-frame time of 29 ms.

Using fMRI non-local means denoising to uncover activation in sub-cortical structures at 1.5 T for guided HARDI tractography

In recent years, there has been ever-increasing interest in combining functional magnetic resonance imaging (fMRI) and diffusion magnetic resonance imaging (dMRI) for better understanding the link between cortical activity and connectivity, respectively. However, it is challenging to detect and validate fMRI activity in key sub-cortical areas such as the thalamus, given that they are prone to susceptibility artifacts due to the partial volume effects (PVE) of surrounding tissues (GM/WM interface). This is especially true on relatively low-field clinical MR systems (e.g., 1.5 T). We propose to overcome this limitation by using a spatial denoising technique used in structural MRI and more recently in diffusion MRI called non-local means (NLM) denoising, which uses a patch-based approach to suppress the noise locally. To test this, we measured fMRI in 20 healthy subjects performing three block-based tasks : eyes-open closed (EOC) and left/right finger tapping (FTL, FTR). Overall, we found that NLM yielded more thalamic activity compared to traditional denoising methods. In order to validate our pipeline, we also investigated known structural connectivity going through the thalamus using HARDI tractography: the optic radiations, related to the EOC task, and the cortico-spinal tract (CST) for FTL and FTR. To do so, we reconstructed the tracts using functionally based thalamic and cortical ROIs to initiates seeds of tractography in a two-level coarse-to-fine fashion. We applied this method at the single subject level, which allowed us to see the structural connections underlying fMRI thalamic activity. In summary, we propose a new fMRI processing pipeline which uses a recent spatial denoising technique (NLM) to successfully detect sub-cortical activity which was validated using an advanced dMRI seeding strategy in single subjects at 1.5 T.

Characterizing growth patterns in longitudinal MRI using image contrast

Understanding the growth patterns of the early brain is crucial to the study of neuro-development. In the early stages of brain growth, a rapid sequence of biophysical and chemical processes take place. A crucial component of these processes, known as myelination, consists of the formation of a myelin sheath around a nerve fiber, enabling the effective transmission of neural impulses. As the brain undergoes myelination, there is a subsequent change in the contrast between gray matter and white matter as observed in MR scans. In this work, gray-white matter contrast is proposed as an effective measure of appearance which is relatively invariant to location, scanner type, and scanning conditions. To validate this, contrast is computed over various cortical regions for an adult human phantom. MR (Magnetic Resonance) images of the phantom were repeatedly generated using different scanners, and at different locations. Contrast displays less variability over changing conditions of scan compared to intensity-based measures, demonstrating that it is less dependent than intensity on external factors. Additionally, contrast is used to analyze longitudinal MR scans of the early brain, belonging to healthy controls and Down's Syndrome (DS) patients. Kernel regression is used to model subject-specific trajectories of contrast changing with time. Trajectories of contrast changing with time, as well as time-based biomarkers extracted from contrast modeling, show large differences between groups. The preliminary applications of contrast based analysis indicate its future potential to reveal new information not covered by conventional volumetric or deformation-based analysis, particularly for distinguishing between normal and abnormal growth patterns.

Quantitative error analysis for computer assisted navigation: a feasibility study

PURPOSE

The benefit of computer-assisted navigation depends on the registration process, at which patient features are correlated to some preoperative imagery. The operator-induced uncertainty in localizing patient features-the user localization error (ULE)-is unknown and most likely dominating the application accuracy. This initial feasibility study aims at providing first data for ULE with a research navigation system.

METHODS

Active optical navigation was done in CT-images of a plastic skull, an anatomic specimen (both with implanted fiducials), and a volunteer with anatomical landmarks exclusively. Each object was registered ten times with 3, 5, 7, and 9 registration points. Measurements were taken at 10 (anatomic specimen and volunteer) and 11 targets (plastic skull). The active NDI Polaris system was used under ideal working conditions (tracking accuracy 0.23 mm root-mean-square, RMS; probe tip calibration was 0.18 mm RMS). Variances of tracking along the principal directions were measured as 0.18 mm(2), 0.32 mm(2), and 0.42 mm(2). ULE was calculated from predicted application accuracy with isotropic and anisotropic models and from experimental variances, respectively.

RESULTS

The ULE was determined from the variances as 0.45 mm (plastic skull), 0.60 mm (anatomic specimen), and 4.96 mm (volunteer). The predicted application accuracy did not yield consistent values for the ULE.

CONCLUSIONS

Quantitative data of application accuracy could be tested against prediction models with iso- and anisotropic noise models and revealed some discrepancies. This could potentially be due to the facts that navigation and one prediction model wrongly assume isotropic noise (tracking is anisotropic), while the anisotropic noise prediction model assumes an anisotropic registration strategy (registration is isotropic in typical navigation systems). The ULE data are presumably the first quantitative values for the precision of localizing anatomical landmarks and implanted fiducials. Submillimetric localization is possible for implanted screws; anatomic landmarks are not suitable for high-precision clinical navigation.