Search

Magnetic resonance (MR) imaging visualises soft tissue contrast in exquisite detail without harmful ionising radiation. In this work, we provide a state-of-the-art review on the use of deep learning in MR image reconstruction from different image acquisition types involving compressed sensing techniques, parallel image acquisition and multi-contrast imaging. Publications with deep learning-based image reconstruction for MR imaging were identified from the literature (PubMed and Google Scholar), and a comprehensive description of each of the works was provided. A detailed comparison that highlights the differences, the data used and the performance of each of these works were also made. A discussion of the potential use cases for each of these methods is provided. The sparse image reconstruction methods were found to be most popular in using deep learning for improved performance, accelerating acquisitions by around 4-8 times. Multi-contrast image reconstruction methods rely on at least one pre-acquired image, but can achieve 16-fold, and even up to 32- to 50-fold acceleration depending on the set-up. Parallel imaging provides frameworks to be integrated in many of these methods for additional speed-up potential. The successful use of compressed sensing techniques and multi-contrast imaging with deep learning and parallel acquisition methods could yield significant MR acquisition speed-ups within clinical routines in the near future.

High resolution three-dimensional (3D) magnetic resonance (MR) images are well suited for automated cartilage segmentation in the human knee joint. However, volumetric scans such as 3D Double-Echo Steady-State (DESS) images are not routinely acquired in clinical practice which limits opportunities for reliable cartilage segmentation using (fully) automated algorithms. In this work, a method for generating synthetic 3D MR (syn3D-DESS) images with better contrast and higher spatial resolution from routine, low resolution, two-dimensional (2D) Turbo-Spin Echo (TSE) clinical knee scans is proposed.A UNet convolutional neural network is employed for synthesizing enhanced artificial MR images suitable for automated knee cartilage segmentation. Training of the model was performed on a large, publically available dataset from the OAI, consisting of 578 MR examinations of knee joints from 102 healthy individuals and patients with knee osteoarthritis.The generated synthetic images have higher spatial resolution and better tissue contrast than the original 2D TSE, which allow high quality automated 3D segmentations of the cartilage. The proposed approach was evaluated on a separate set of MR images from 88 subjects with manual cartilage segmentations. It provided a significant improvement in automated segmentation of knee cartilages when using the syn3D-DESS images compared to the original 2D TSE images.The proposed method can successfully synthesize 3D DESS images from 2D TSE images to provide images suitable for automated cartilage segmentation.

Combining high-resolution magnetic resonance imaging (MRI) with a linear accelerator (Linac) as a single MRI-Linac system provides the capability to monitor intra-fractional motion and anatomical changes during radiotherapy, which facilitates more accurate delivery of radiation dose to the tumour and less exposure to healthy tissue. The gradient non-linearity (GNL) induced distortions in MRI; however, hinder the implementation of MRI-Linac system in image-guided radiotherapy where highly accurate geometry and anatomy of the target tumour is indispensable. To correct the geometric distortions in MR images, in particular, for the 1 Tesla (T) MRI-Linac system, a deep fully-connected neural network was proposed to automatically learn the intricate relationship between the undistorted (theoretical) and distorted (real) space. A data set, consisting of spatial samples acquired by phantom measurement that covers both inside and outside the working diameter of spherical volume (DSV), was utilised for training the neural network; which offers the ability to describe subtle deviations of the GNL field within the entire region of interest (ROI). The performance of the proposed method was evaluated on MR images of a three-dimensional (3D) phantom and the pelvic region of an adult volunteer scanned in the 1T MRI-Linac system. The experimental results showed that the severe geometric distortions within the entire ROI had been successfully corrected with an error less than the pixel size. Also, the presented network is highly efficient, which achieved significant improvement in terms of computational efficiency compared to existing methods. The feasibility of the presented deep neural network for characterising the GNL field deviations in the 1T MRI-Linac system was demonstrated in this study, which shows promise in facilitating the MRI-Linac system to be routinely implemented in real-time MRI-guided radiotherapy.

Active shape models (ASMs) have proved successful in automatic segmentation by using shape and appearance priors in a number of areas such as prostate segmentation, where accurate contouring is important in treatment planning for prostate cancer. The ASM approach however, is heavily reliant on a good initialisation for achieving high segmentation quality. This initialisation often requires algorithms with high computational complexity, such as three dimensional (3D) image registration. In this work, we present a fast, self-initialised ASM approach that simultaneously fits multiple objects hierarchically controlled by spatially weighted shape learning. Prominent objects are targeted initially and spatial weights are progressively adjusted so that the next (more difficult, less visible) object is simultaneously initialised using a series of weighted shape models. The scheme was validated and compared to a multi-atlas approach on 3D magnetic resonance (MR) images of 38 cancer patients and had the same (mean, median, inter-rater) Dice's similarity coefficients of (0.79, 0.81, 0.85), while having no registration error and a computational time of 12-15 min, nearly an order of magnitude faster than the multi-atlas approach.

To validate a fully automated scheme to extract biochemical information from the hip joint cartilages using MR T2 mapping images incorporating segmentation of co-registered three-dimensional Fast-Spin-Echo (3D-SPACE) images.Manual analyses of unilateral hip (3 Tesla) MR images of 24 asymptomatic volunteers were used to validate a 3D deformable model method for automated cartilage segmentation of SPACE scans, partitioning of the individual femoral and acetabular cartilage plates into clinically defined sub-regions and propagating these results to T2 maps to calculate region-wise T2 value statistics. Analyses were completed on a desktop computer (∼ 10 min per case).The mean voxel overlap between automated A and manual M segmentations of the cartilage volumes in the (clinically based) SPACE images was 73% (100 × 2|A∩M|/[|A|+|M|]). The automated and manual analyses demonstrated a relative difference error10% in the median "T2 average signal" for each cartilage plate. The automated and manual analyses showed consistent patterns between significant differences in T2 data across the hip cartilage sub-regions.The good agreement between the manual and automatic analyses of T2 values indicates the use of structural 3D-SPACE MR images with the proposed method provides a promising approach for automated quantitative T2 assessment of hip joint cartilages.

Breast mass detection and segmentation are difficult tasks due to the variation in size and shape of breast masses. Constructing classifiers for this problem is also challenging due to the fact that normal tissue regions overwhelmingly outnumber abnormal regions. In this paper, we propose a novel approach for detecting and segmenting breast masses in mammography based on multi-scale morphological filtering and a self-adaptive cascade of random forests (CasRFs). CasRFs can cope with severe class imbalance by adding layers to the cascade until a minimum number of false-positives (FPs) is reached. The approach achieves an average sensitivity of 0.94 with 1.99 FPs/image on INbreast and a sensitivity of 0.77 with 3.93 FPs/image on DDSM BCRP.

Magnetic resonance (MR) imaging is an important imaging modality for diagnostic medicine due to its ability to visualize the soft tissue of the human body in three dimensions (3D) without ionizing radiation. MR imaging however, typically relies on measurement techniques that do not exploit the geometry of discrete Fourier space, so called k-space, where the measurements are made. In this work, we firstly present a novel k-space tiling scheme that utilizes the finite geometry of discrete Fourier space via the discrete Fourier slice theorem. This produces a sampling of k-space that is pseudo-radial, to be more noise and patient-movement tolerant, and pseudo-random, further improving robustness to noise, whilst sampling with sufficient density near the central k-space region, where the majority of power of anatomical images lies. Secondly, we introduce a stable and iterative discrete reconstruction scheme for recovering images from their limited k-space measurements (in the form discrete slices) based on the maximum likelihood expectation maximization approach. We study the performance of the proposed approach using simulated MR measurements in k-space.

Traditional shape learning of medical image data has been implemented via Principal Component Analysis (PCA). These PCA based statistical shape models batch process all shapes at once to generate a fixed model of shape variation as principal components, which may require significant computation resources for large number of shapes. This paper applies incremental PCA (IPCA) on a dataset of 728 surfaces (derived from magnetic resonance imaging examinations displaying the articulating bones of the knee joint) that can efficiently adapt to changes in training sets. After comparing the compactness and the accuracy of shape reconstruction of both batch PCA and IPCA models, our results show that IPCA produces a model comparable to batch PCA in terms of compactness and applicability to shape reconstruction, while requiring considerably shorter processing time and computer memory for computation.

Accurate localization of the prostate and its surrounding tissue is essential in the treatment of prostate cancer. This paper presents a novel approach to fully automatically segment the prostate, including its seminal vesicles, within a few minutes of a magnetic resonance (MR) scan acquired without an endorectal coil. Such MR images are important in external beam radiation therapy, where using an endorectal coil is highly undesirable. The segmentation is obtained using a deformable model that is trained on-the-fly so that it is specific to the patient's scan. This case specific deformable model consists of a patient specific initialized triangulated surface and image feature model that are trained during its initialization. The image feature model is used to deform the initialized surface by template matching image features (via normalized cross-correlation) to the features of the scan. The resulting deformations are regularized over the surface via well established simple surface smoothing algorithms, which is then made anatomically valid via an optimized shape model. Mean and median Dice's similarity coefficients (DSCs) of 0.85 and 0.87 were achieved when segmenting 3T MR clinical scans of 50 patients. The median DSC result was equal to the inter-rater DSC and had a mean absolute surface error of 1.85 mm. The approach is showed to perform well near the apex and seminal vesicles of the prostate. © 1982-2012 IEEE.

Automated segmentation of intervertebral discs (IVDs) from magnetic resonance imaging has the potential to enhance the efficiencies of radiological investigations of large clinical and research imaging datasets. This work presents an automated method for localization and 3D segmentation of IVDs that is applied to magnetic resonance imaging of the thoraco-lumbar spine as part of the segmentation challenge at the 3rd MICCAI Workshop & Challenge on Computational Methods and Clinical Applications for Spine Imaging - MICCAI–CSI2015. Our initialization method involves multi-atlas registration and a hierarchical conditional shape regression for localization of all imaged lumbar and thoracic discs, and active shape model based 3D segmentation. Comparisons between manual (ground truth) and automated segmentation of 105 disc volumes (T11/T12 - L5/S1) revealed a mean Dice score of \(0.896\,{\pm }\,0.024\) and mean absolute square distance of \(0.642\,{\pm }\,0.169\) mm. Our automated segmentation approach provided accurate segmentation of IVDs from turbo spine echo images which are highly competitive with leading state-of-the-art 3D segmentation techniques.

We present a statistical shape model approach for automated segmentation of the proximal humerus and scapula with subsequent bone-cartilage interface (BCI) extraction from 3D magnetic resonance (MR) images of the shoulder region. Manual and automated bone segmentations from shoulder MR examinations from 25 healthy subjects acquired using steady-state free precession sequences were compared with the Dice similarity coefficient (DSC). The mean DSC scores between the manual and automated segmentations of the humerus and scapula bone volumes surrounding the BCI region were 0.926 ± 0.050 and 0.837 ± 0.059, respectively. The mean DSC values obtained for BCI extraction were 0.806 ± 0.133 for the humerus and 0.795 ± 0.117 for the scapula. The current model-based approach successfully provided automated bone segmentation and BCI extraction from MR images of the shoulder. In future work, this framework appears to provide a promising avenue for automated segmentation and quantitative analysis of cartilage in the glenohumeral joint.

Deformable models incorporating shape priors have proved to be a successful approach in segmenting anatomical regions and specific structures in medical images. This paper introduces weighted shape priors for deformable models in the context of 3D magnetic resonance (MR) image segmentation of the bony elements of the human hip joint. The fully automated approach allows the focusing of the shape model energy to a priori selected anatomical structures or regions of clinical interest by preferentially ordering the shape representation (or eigen-modes) within this type of model to the highly weighted areas. This focused shape model improves accuracy of the shape constraints in those regions compared to standard approaches. The proposed method achieved femoral head and acetabular bone segmentation mean absolute surface distance errors of 0.55±0.18mm and 0.75±0.20mm respectively in 35 3D unilateral MR datasets from 25 subjects acquired at 3T with different limited field of views for individual bony components of the hip joint.

This paper presents a validation study on the application of a novel interslice interpolation technique for musculoskeletal structure segmentation of articulated joints and muscles on human magnetic resonance imaging data. The interpolation technique is based on morphological shape-based interpolation combined with intensity based voxel classification. Shape-based interpolation in the absence of the original intensity image has been investigated intensively. However, in some applications of medical image analysis, the intensity image of the slice to be interpolated is available. For example, when manual segmentation is conducted on selected slices, the segmentation on those unselected slices can be obtained by interpolation. We proposed a two- step interpolation method to utilize both the shape information in the manual segmentation and local intensity information in the image. The method was tested on segmentations of knee, hip and shoulder joint bones and hamstring muscles. The results were compared with two existing interpolation methods. Based on the calculated Dice similarity coefficient and normalized error rate, the proposed method outperformed the other two methods.

The accurate segmentation of the bone from Magnetic Resonance (MR) images of the hip is important for clinical studies and drug trials into conditions like Osteoarthritis. This paper presents an automatic segmentation scheme that utilises a deformable model robust to different field of views by training shape priors from partial and full bone surfaces. The deformable model with these priors were used to segment the hip joint within 16 unilateral 3T MR images having different field of views, so that parts of the model outside the image could be ignored fully without affecting the accuracy of the segmentation within the image. Mean and median Dice's Similarity Coefficients of 0.91 & 0.92 for the femur and 0.86 & 0.88 for one half of the pelvis were obtained using a leave-one-out approach.

This paper presents a novel approach to automatically segment the prostate (including seminal vesicles) using a surface that is actively deformed via shape and gray level models. The surface deformation process utilises the results of a multi-atlas registration approach, where training images are matched to the case image via non-rigid registration. Normalised mutual information is then used to measure the similarity between each image in the training set and the case image. The set of training images with a similarity greater than a threshold is then used to build the initialisation and the gray level model of the segmentation process. This case specific gray level model is used to deform the initial surface to more closely match the prostate boundary via normalised cross-correlation based template matching of gray level profiles. Mean and median Dice's Similarity Coefficients of 0.849 and 0.855, as well as a mean surface error of 2.11 mm, were achieved when segmenting 3T Magnetic Resonance clinical scans of fifty patients.

International audience; Ghosts are digital images that contain highly constrained patterns of signed pixel values. The pixels are located so as to create zero-sums when discrete projections are taken across the image at a pre-determined set of angles. Ghosts can be applied to create image/anti-image pairs. An image that is entangled with its anti-image can be used to achieve forward error-correction in redundant data transmission schemes. Ghosts can also be used to help reconstruct images from asymmetric sets of real, noisy tomographic projection data. Minimal ghosts do these tasks most efficiently. We present here new methods to construct minimal ghost images that employ just $2N$ pixels to obtain zero-sum projections for $N$ angles. Construction of an Nth order ghost had previously required $O(2^N)$ pixels.

Our group have been developing methods for MRI-alone prostate cancer radiation therapy treatment planning. To assist with clinical validation of the workflow we are investigating a cloud platform solution for research purposes. Benefits of cloud computing can include increased scalability, performance and extensibility while reducing total cost of ownership. In this paper we demonstrate the generation of DICOM-RT directories containing an automatic average atlas based electron density image and fast pelvic organ contouring from whole pelvis MR scans.