Single MR Image Research Articles

Adaptive wavelet-VNet for single-sample test time adaptation in medical image segmentation.

In medical image segmentation, a domain gap often exists between training and testing datasets due to different scanners or imaging protocols, which leads to performance degradation in deep learning-based segmentation models. Given the high cost of manual labeling and the need for privacy protection, it is often challenging to annotate the testing (target) domain data for model fine-tuning or to collect data from different domains to train domain generalization models. Therefore, using only unlabeled target domain data for test-time adaptation (TTA) presents a more practical but challenging solution. To improve the segmentation accuracy of deep learning-based models on unseen datasets, and especially to enhance the efficiency and stability of TTA for individual samples from heterogeneous domains. In this study, we proposed to dynamically adapt a wavelet-VNet (WaVNet) to unseen target domains with a hybrid objective function, based on each unlabeled test sample during the test time. We embedded multiscale wavelet coefficients into a V-Net encoder and adaptively adjusted the spatial and spectral features according to the input, and the model parameters were optimized by three loss functions. We integrated a shape-aware loss to focus on the foreground segmentations, a Refine loss to correct the incomplete and noisy segmentations caused by domain shifts, and an entropy loss to promote the global consistency of the segmentations. We evaluated the proposed method on multidomain liver and prostate segmentation datasets to assess its advantages over other TTA methods. For the source domain model training of the liver dataset, we used 15 3D MR image samples for training and 5 for validation. Correspondingly, for the prostate dataset, we used 22 3D MR image samples for training and 7 for validation. In the target domain, we used a single 3D MR image sample for adaptation and testing. The total number of testing samples is 60 in the liver dataset (for 3 different domains) and 116 in the prostate dataset (for 6 different domains). The proposed method showed the highest segmentation accuracy among all methods, achieving a mean(±SD) Dice coefficient (DSC) of 78.10±5.23% and a mean 95th Hausdorff distance (HD95) of 15.52±5.84mm on the liver dataset; and a mean DSC of 80.02±3.89% and a mean HD95 of 9.18±3.47mm on the prostate dataset. The DSC is 11.67% (in absolute terms) and 15.27% higher than that of the baseline (no adaptation) method, for the liver and the prostate datasets, respectively. The proposed adaptive WaVNet enhanced the image segmentation accuracy from unseen domains during the test time via unsupervised learning and multi-objective optimization. It can benefit clinical applications where data are scarce or with changing data distributions, including online adaptive radiotherapy. The code will be released at: https://github.com/sanny1226/WaVNet.

Single slice MR image estimation of 3D supraspinatus intramuscular fatty infiltration in older adults: relevance for clinical practice and research.

(1) To compare older adults stratified by supraspinatus tendon tear status (STT status)-no tear (Intact), partial-thickness (PT) tear, full-thickness (FT) tear-by 3D Dixon fat fraction (3D-FF); 2D fat fraction (2D-FF); and 2D Goutallier grade (2D-GG) at the Y-shaped view, and 1.4cm and 2.8cm medial to the Y-shaped view. Stratified by STT status to determine (2) correlation of 3D-FF with 2D-FF and 2D-GG and (3) inter-rater reliability at and medial to the Y-shaped view. Forty-five volunteers ≥ 60years recruited prospectively received shoulder MRI. 3D-FF and 2D-FF were measured on 6-point-Dixon MRI by three trainees. Goutallier grade was assessed on T1-weighted MRI by three fellowship-trained diagnostic radiologists. Descriptive, reliability, and correlation analyses were performed. Groups showed no difference in age. The FT group showed higher (p < 0.05) mean 3D-FF (14.09% ± 10.99%), mean 2D-FF (1.4cm medial to Y-shaped view, 14.91% ± 12.11%; 2.8cm medial to Y-shaped view, 13.32% ± 9.48%), and mean 2D-GG (Y-shaped view, 1.71 ± 0.78; 1.4cm medial to Y-shaped view, 1.71 ± 0.69; 2.8cm medial to Y-shaped view, 1.71 ± 0.72), relative to Intact/PT groups. 3D-FF showed strong correlation with 2D-FF among all groups/all analyses (rho, 0.80-0.98; p < 0.001). 3D-FF showed strong correlation with 2D-GG for all FT group analyses (rho, 0.85-0.91; p < 0.05). 3D-FF showed moderate-to-strong correlation considering all Intact/PT group analyses (rho, 0.51-0.79; p < 0.50). Dixon fat fraction showed excellent reliability for all groups (≥ 0.884, intraclass correlation coefficient). Goutallier grade showed excellent reliability for FT group (0.771, weighted Fleiss's kappa) but poor (0.294) and fair (0.502) for Intact and PT groups, respectively. Single slice MR image estimation of 3D supraspinatus intramuscular fatty infiltration has merit for continued use in clinical populations requiring potential rotator-cuff-repair surgery. However, Dixon fat fraction should be prioritized for use in research over Goutallier grade due to superior reliability.

Unified Model for Children's Brain Image Segmentation with Co-Registration Framework Guided by Longitudinal MRI.

Accurate segmentation of brain structures is crucial for analyzing longitudinal changes in children's brains. However, existing methods are mostly based on models established at a single time-point due to difficulty in obtaining annotated data and dynamic variation of tissue intensity. The main problem with such approaches is that, when conducting longitudinal analysis, images from different time points are segmented by different models, leading to significant variation in estimating development trends. In this paper, we propose a novel unified model with co-registration framework to segment children's brain images covering neonates to preschoolers, which is formulated as two stages. First, to overcome the shortage of annotated data, we propose building gold-standard segmentation with co-registration framework guided by longitudinal data. Second, we construct a unified segmentation model tailored to brain images at 0-6 years old through the introduction of a convolutional network (named SE-VB-Net), which combines our previously proposed VB-Net with Squeeze-and-Excitation (SE) block. Moreover, different from existing methods that only require both T1- and T2-weighted MR images as inputs, our designed model also allows a single T1-weighted MR image as input. The proposed method is evaluated on the main dataset (320 longitudinal subjects with average 2 time-points) and two external datasets (10 cases with 6-month-old and 40 cases with 20-45 weeks, respectively). Results demonstrate that our proposed method achieves a high performance (>92%), even over a single time-point. This means that it is suitable for brain image analysis with large appearance variation, and largely broadens the application scenarios.

Deep Learning Algorithm Trained on Cervical Magnetic Resonance Imaging to Predict Outcomes of Transforaminal Epidural Steroid Injection for Radicular Pain from Cervical Foraminal Stenosis.

A convolutional neural network (CNN) is one of the representative deep learning (DL) model that is especially useful for image recognition and classification. In the current study, using cervical axial magnetic resonance imaging (MRI) data obtained prior to transforaminal epidural steroid injection (TFESI), we developed a CNN model to predict the therapeutic outcome of cervical TFESI in patients with cervical foraminal stenosis. We retrospectively recruited 288 patients with cervical foraminal stenosis who received cervical TFESI due to cervical radicular pain. We collected single T2-axial spine MR image obtained from each patient. The image showing narrowest width of the neural foramen in the level at which TFESI was performed was used for input data. A "favor outcome" was defined as a ≥ 50% reduction in the NRS score at 2 months post-TFESI vs the pretreatment NRS score. A "poor outcome" was defined as a < 50% reduction in the NRS score at 2 months post-TFESI vs the pretreatment score. The area under the curve of our developed model for predicting therapeutic outcome of cervical TFESI in patients with cervical spinal stenosis was 0.801. We showed that a CNN model trained using cervical axial MRI could be helpful for predicting therapeutic outcome after cervical TFESI in patients with cervical foraminal stenosis.

Quantifying Brain Iron in Hereditary Hemochromatosis Using R2* and Susceptibility Mapping.

Brain iron dyshomeostasis is increasingly recognized as an important contributor to neurodegeneration. Hereditary hemochromatosis is the most commonly inherited disorder of systemic iron overload. Although there is an increasing interest in excessive brain iron deposition, there is a paucity of evidence showing changes in brain iron exceeding that in healthy controls. Quantitative susceptibility mapping and R2* mapping are established MR imaging techniques that we used to noninvasively quantify brain iron in subjects with hereditary hemochromatosis. Fifty-two patients with hereditary hemochromatosis and 47 age- and sex-matched healthy controls were imaged using a multiecho gradient-echo sequence at 3T. Quantitative susceptibility mapping and R2* data were generated, and regions within the deep gray matter were manually segmented. Mean susceptibility and R2* relaxation rates were calculated for each region, and iron content was compared between the groups. We noted elevated iron levels in patients with hereditary hemochromatosis compared with healthy controls using both R2* and QSM methods in the caudate nucleus, putamen, pulvinar thalamus, red nucleus, and dentate nucleus. Additionally, the substantia nigra showed increased susceptibility while the thalamus showed an increased R2* relaxation rate compared with healthy controls, respectively. Both quantitative susceptibility mapping and R2* showed abnormal levels of brain iron in subjects with hereditary hemochromatosis compared with controls. Quantitative susceptibility mapping and R2* can be acquired in a single MR imaging sequence and are complementary in quantifying deep gray matter iron.

Retrospective Tuning of MRI Contrast From a Single T1-Weighted Image

While versatile soft tissue contrasts are achievable in MRI, contrast attainable from every MRI scan is predetermined by imaging protocol. Retrospective tuning of MRI contrast provides an opportunity to presumably change imaging parameter values without actual data acquisition. In this study, we present a new paradigm to obtain various contrasts from a single T1-weighted image. In this way, MRI data acquired at different medical centers can be normalized and used for large-scaled radiomics analysis.A novel framework is proposed for retrospective contrast tuning, which combines deep learning-based quantitative MRI with Bloch equations. Theoretically, retrospective change of MRI contrast is achievable by applying Bloch equations on tissue relaxation parametric maps. However, in practice, quantitative MRI is seldom acquired due to long scan time. Leveraging from the capability of deep learning, we propose a quantitative MRI approach to extract tissue relaxation parametric maps from a single MR image without extra data acquisition. Using deep neural networks, we predict quantitative tissue relaxation parametric maps (T1 map, proton density map) and radiofrequency field map (B1 map) from a single T1-weighted image. Subsequently, various MRI contrasts are obtained from estimated parametric maps with the application of Bloch equations. The principle is validated on knee MRI. A total of 1,344 slice images from 56 subjects were retrospectively collected. For every subject, T1 map was measured from four variable flip angles T1-weighted images and a B1 map (obtained using the actual flip angle method), and proton density map was calculated from T1-weighted image and T1 map. These maps are used as the ground truth. In our study, every quantitative parametric map is predicted from a single T1 weighted image using a self-attention convolutional neural network equipped with unique shortcuts and attention mechanism (to make efficient use of non-local information). All the images are utilized for training and testing with the six-fold cross validation strategy adopted. Given estimated parametric maps, MR images with different contrasts are generated with the application of Bloch equations. While a wide spectrum of contrasts can be obtained with various imaging parameter values, the result is only validated at certain contrasts (corresponding to variable flip angles).High accuracy has been achieved in quantitative parametric maps and qualitative images. From a T1-weighted image, quantitative T1 map, proton density map and B1 map are predicted with an averaged correlation coefficient of 0.95∼0.99 and L1 error of 0.02∼0.09; and T1-weighted images corresponding to alternative flip angles are obtained with an averaged correlation coefficient of 0.97∼0.99 and L1 error of 0.04∼0.09.A new data-driven strategy is proposed for retrospective MRI contrast tuning from a single T1-weighted image, which requires no additional data acquisition.

Creation and Validation of a Shorthand Knee MRI Bone Age Assessment Tool as an Alternative for Skeletal Maturity Assessment (203)

Objectives:In managing pediatric knee conditions, an accurate bone age assessment is often critical for diagnostic, prognostic, and treatment purposes. Historically, the Greulich and Pyle Atlas (hand atlas) has been the gold standard bone age assessment tool. In 2013, a shorthand bone age assessment tool based on this atlas (hand shorthand) was established as a simpler and more efficient alternative. Recently, a knee MRI bone age atlas (MRI atlas) was created potentially to circumvent the need for a left hand radiograph. Our objective is to create a shorthand version of the magnetic resonance imaging atlas.Methods:A shorthand bone age method (Figure 1) was created utilizing the previously published MRI atlas, which utilizes several criteria that are visualized across a series of images. The MRI shorthand draws on the most characteristic criteria for each age that is best observed on a single MR image. For validation, we performed a retrospective assessment of skeletally immature patients that had a knee MRI and left hand radiograph within four weeks. Four readers who were familiar with the hand atlas, hand shorthand, MRI atlas, and MRI shorthand read each of the images in a blinded fashion. Inter- and intra-observer reliability was evaluated using intraclass correlation coefficient (ICC), variability among observers was evaluated using percent agreement.Results:26 patients with a mean age of 13.6 years (range 9.0-16.9) met the inclusion criteria. The intra- and inter-observer reliability of all four assessment tools was excellent (ICC ≥ 0.8, p<0.001) (Table 1). When comparing the MRI shorthand to the MRI atlas, there was excellent agreement (ICC = 0.974), whereas the hand shorthand compared to the hand atlas had good agreement (ICC = 0.765). The MRI shorthand also had perfect agreement in 58% of reads among all four readers and 96% of reads had agreement within 1 year, whereas the hand shorthand had perfect agreement in 32% of reads and 77% agreement within 1 year (Table 2).Conclusions:The MRI shorthand is a simple and efficient means of assessing skeletal maturity of adolescent patients with a knee MRI. This bone age assessment technique has inter-observer and intra-observer reliability equivalent or better than the standard means utilizing a left hand radiograph.

3D dense convolutional neural network for fast and accurate single MR image super-resolution

Super-resolution (SR) MR image reconstruction has shown to be a very promising direction to improve the spatial resolution of low-resolution (LR) MR images. In this paper, we presented a novel MR image SR method based on a dense convolutional neural network (DDSR), and its enhanced version called EDDSR. There are three major innovations: first, we re-designed dense modules to extract hierarchical features directly from LR images and propagate the extracted feature maps through dense connections. Therefore, unlike other CNN-based SR MR techniques that upsample LR patches in the initial phase, our methods take the original LR images or patches as input. This effectively reduces computational complexity and speeds up SR reconstruction. Second, a final deconvolution filter in our model automatically learns filters to fuse and upscale all hierarchical feature maps to generate HR MR images. Using this, EDDSR can perform SR reconstructions at different upscale factors using a single model with one stride fixed deconvolution operation. Third, to further improve SR reconstruction accuracy, we exploited a geometric self-ensemble strategy. Experimental results on three benchmark datasets demonstrate that our methods, DDSR and EDDSR, achieved superior performance compared to state-of-the-art MR image SR methods with less computational load and memory usage.

Research progress of MR imaging for prediction of CT imaging

Medical images can provide clinicans with accurate and comprehensive patients’ information. Morphological or functional abnormalities caused by various diseases can be manifested in many aspects. Although MR images and CT images can highlight the medical image data of different tissue structures of patients, single MR images or CT images cannot fully reflect the complexity of diseases. Using MR image to predict CT image is one of the cross-modal prediction of medical images. In this paper, the methods of MR image prediction for CT image are classified into four categoriesincluding registration based on atlas, based on image segmentationmethod, based on learning method and based on deep learning method. In our research, we concluded that the method based on deep learning should bemore promoted in the future by compering the existing problems and future development of MR image predicting CT image method. 摘要： 医学图像可以为医生提供准确和全面的病患信息。由于人体因各种疾病引起的形态或功能异常可以表现在很 多方面, MR 图像和 CT 图像能重点呈现出患者不同组织结构的医学图像数据, 但单独的 MR 图像或者 CT 图像不能 全面反应出疾病的复杂性。MR 图像预测 CT 图像属于医学图像跨模态预测的一种, 将 MR 图像预测 CT 图像的方法 分为 4 类, 基于图集的方法、基于图像分割的方法、基于学习的方法和基于深度学习的方法。本文对 MR 图像预测 CT 图像的各类方法、存在问题和未来发展方向进行综述, 得出结论基于深度学习的方法应是未来跨模态预测的主要 方法。

Adaptive Bayesian filtering based restoration of MR images

MR images contain noise and for quantitative clinical diagnosis restoration is essential. For single coil MR images, noise follows Rician distribution when signal to noise ratio (SNR) is low and Gaussian distribution when SNR is high. Rician noise is signal dependent and introduces bias. The work proposes an adaptive Bayesian framework for restoration of 2D magnitude MR images. Restoration is achieved by Rician likelihood as the data attachment term with range and domain Gaussian filters, adaptive to noise as prior in Maximum A´ posterior framework. A good filtering behavior is achieved due to the domain component of the filter and crisp edges are preserved at the same time due to the noise adaptive range component. Rician likelihood aids the image restoration in terms of bias removal. Convergence of the proposed method further highlights the optimal filtering performance. Experiments conducted on publically available Brainweb phantom demonstrate enhanced performance in terms of signal to noise ratio, structural similarity index and overall performance.

Multimodal MRI synthesis using unified generative adversarial networks.

Complementary information obtained from multiple contrasts of tissue facilitates physicians assessing, diagnosing and planning treatment of a variety of diseases. However, acquiring multiple contrasts magnetic resonance images (MRI) for every patient using multiple pulse sequences is time-consuming and expensive, where, medical image synthesis has been demonstrated as an effective alternative. The purpose of this study is to develop a unified framework for multimodal MR image synthesis. A unified generative adversarial network consisting of only a single generator and a single discriminator was developed to learn the mappings among images of four different modalities. The generator took an image and its modality label as inputs and learned to synthesize the image in the target modality, while the discriminator was trained to distinguish between real and synthesized images and classify them to their corresponding modalities. The network was trained and tested using multimodal brain MRI consisting of four different contrasts which are T1-weighted (T1), T1-weighted and contrast-enhanced (T1c), T2-weighted (T2), and fluid-attenuated inversion recovery (Flair). Quantitative assessments of our proposed method were made through computing normalized mean absolute error (NMAE), peak signal-to-noise ratio (PSNR), structural similarity index measurement (SSIM), visual information fidelity (VIF), and naturalness image quality evaluator (NIQE). The proposed model was trained and tested on a cohort of 274 glioma patients with well-aligned multi-types of MRI scans. After the model was trained, tests were conducted by using each of T1, T1c, T2, Flair as a single input modality to generate its respective rest modalities. Our proposed method shows high accuracy and robustness for image synthesis with arbitrary MRI modality that is available in the database as input. For example, with T1 as input modality, the NMAEs for the generated T1c, T2, Flair respectively are 0.034±0.005, 0.041±0.006, and 0.041±0.006, the PSNRs respectively are 32.353±2.525dB, 30.016±2.577dB, and 29.091±2.795dB, the SSIMs are 0.974±0.059, 0.969±0.059, and 0.959±0.059, the VIF are 0.750±0.087, 0.706±0.097, and 0.654±0.062, and NIQE are 1.396±0.401, 1.511±0.460, and 1.259±0.358, respectively. We proposed a novel multimodal MR image synthesis method based on a unified generative adversarial network. The network takes an image and its modality label as inputs and synthesizes multimodal images in a single forward pass. The results demonstrate that the proposed method is able to accurately synthesize multimodal MR images from a single MR image.

Fully automatic classification of breast MRI background parenchymal enhancement using a transfer learning approach.

Marked enhancement of the fibroglandular tissue on contrast-enhanced breast magnetic resonance imaging (MRI) may affect lesion detection and classification and is suggested to be associated with higher risk of developing breast cancer. The background parenchymal enhancement (BPE) is qualitatively classified according to the BI-RADS atlas into the categories “minimal,” “mild,” “moderate,” and “marked.” The purpose of this study was to train a deep convolutional neural network (dCNN) for standardized and automatic classification of BPE categories.This IRB-approved retrospective study included 11,769 single MR images from 149 patients. The MR images were derived from the subtraction between the first post-contrast volume and the native T1-weighted images. A hierarchic approach was implemented relying on 2 dCNN models for detection of MR-slices imaging breast tissue and for BPE classification, respectively. Data annotation was performed by 2 board-certified radiologists. The consensus of the 2 radiologists was chosen as reference for BPE classification. The clinical performances of the single readers and of the dCNN were statistically compared using the quadratic Cohen's kappa.Slices depicting the breast were classified with training, validation, and real-world (test) accuracies of 98%, 96%, and 97%, respectively. Over the 4 classes, the BPE classification was reached with mean accuracies of 74% for training, 75% for the validation, and 75% for the real word dataset. As compared to the reference, the inter-reader reliabilities for the radiologists were 0.780 (reader 1) and 0.679 (reader 2). On the other hand, the reliability for the dCNN model was 0.815.Automatic classification of BPE can be performed with high accuracy and support the standardization of tissue classification in MRI.

A New Hybrid Brain MR Image Segmentation Algorithm With Super-Resolution, Spatial Constraint-Based Clustering and Fine Tuning

Tissue segmentation from a single brain MR image is of paramount importance for brain reconstruction and analysis. In this paper, we propose a new hybrid algorithm for brain MR image segmentation, combining super-resolution, spatial constraint based clustering and fine-tuning. To smooth noise and improve image clarity, we first amplify the brain MR image by using a super-resolution algorithm - cubic surface fitting with edges in the image as constraints. Then an improved fuzzy c-means clustering algorithm is performed on the amplified image for the global segmentation, in which a shape parameter and an anomaly detection parameter are introduced. With the introduction of these two parameters, the robustness of the clustering is enhanced, and the trade-off between noise smoothing and detail preservation can be controlled more accurately. Furthermore, the local regions around boundaries of different brain tissues (e.g., gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF)) are re-segmented in a fine-tuning process, and a soft voting strategy is adopted for adjusting the incorrect pixels, which makes full use of the boundary details of different tissues. Experimental results show the new algorithm can preserve major brain tissue structures and smooth out noise.

Gadoxetate acid disodium-enhanced MRI: Multiple arterial phases using differential sub-sampling with cartesian ordering (DISCO) may achieve more optimal late arterial phases than the single arterial phase imaging

BackgroundTo prospectively determine whether the use of a multiple arterial phase imaging (DISCO) improve the capturing rate of late arterial phase with less motion artifact than single arterial phase obtained with gadoxetate acid disodium. Materials and methodsFrom 06/2017 to 10/2018, prospectively acquired data of 132 patients who underwent either single (n = 67) or multiple arterial phase (n = 65) gadoxetate acid-enhanced MR imaging were analyzed. Two readers independently assessed arterial phase timing and the degree of motion artifact using a five-point scale. The kappa test was used to determine the agreement between the two readers, χ2 or fisher exact test were used for the categorical variables and Student t-test or Mann-Whitney U test were used for the comparison of the motion artifacts. ResultsGood to perfect inter-observer agreement was obtained for the arterial phase timing and degree of motion artifact (all kappa value >0.70). Optimal timing of arterial phase was observed in 95.4% (62/65) of multiple arterial phase compared with 73.1% (49/67) of single arterial phase (χ2 = 12.209, p < 0.001). Motion artifact score of the late arterial phase images measured using single arterial phase acquisition (3.22 ± 0.68) was significantly higher than the multiple arterial phase (2.42 ± 0.74) group (t = 5.921, p < 0.001). For the multiple arterial phase comparison, motion artifact score of the 2nd, 3rd and 4th phases were also significant reduced compared with 1st, 5th and 6th phases (all p < 0.05). ConclusionThe use of multiple arterial phase acquisition with gadoxetate acid disodium can improve the capturing rate of well-timed late arterial phase with less motion artifact.

Deep Learning Model to Assess Cancer Risk on the Basis of a Breast MR Image Alone.

OBJECTIVE. The purpose of this study is to develop an image-based deep learning (DL) model to predict the 5-year risk of breast cancer on the basis of a single breast MR image from a screening examination. MATERIALS AND METHODS. We collected 1656 consecutive breast MR images from screening examinations performed for 1183 high-risk women from January 2011 to June 2013, to predict the risk of cancer developing within 5 years of the screening. Women who lacked a 5-year screening follow-up examination and women who had cancer other than primary breast cancer develop in their breast were excluded from the study. We developed a logistic regression model based on traditional risk factors (the risk factor logistic regression [RF-LR] model) and a DL model based on the MR image alone (the Image-DL model). Examinations occurring within 6 months of a cancer diagnosis were excluded from the testing sets in each fold of cross-validation. We compared our models against the Tyrer-Cuzick (TC) model. All models were evaluated using mean (± SD) AUC values and observed-to-expected (OE) ratios across 10-fold cross-validation. RESULTS. The RF-LR and Image-DL models achieved mean AUC values of 0.558 ± 0.108 and 0.638 ± 0.094, respectively. In contrast, the TC model achieved an AUC value of 0.493 ± 0.092. The Image-DL and RF-LR models achieved mean OE ratios of 0.993 ± 0.658 and 0.828 ± 0.181, compared with the mean OE ratio of 1.091 ± 0.255 obtained using the TC model. CONCLUSION. Our DL model can assess the 5-year cancer risk on the basis of a breast MR image alone, and it showed improved individual risk discrimination when compared with a state-of-the-art risk assessment model. These results offer promising preliminary data regarding the potential of image-based risk assessment models to support more personalized care.

MR-based treatment planning in radiation therapy using a deep learning approach.

PurposeTo develop and evaluate the feasibility of deep learning approaches for MR‐based treatment planning (deepMTP) in brain tumor radiation therapy.Methods and materialsA treatment planning pipeline was constructed using a deep learning approach to generate continuously valued pseudo CT images from MR images. A deep convolutional neural network was designed to identify tissue features in volumetric head MR images training with co‐registered kVCT images. A set of 40 retrospective 3D T1‐weighted head images was utilized to train the model, and evaluated in 10 clinical cases with brain metastases by comparing treatment plans using deep learning generated pseudo CT and using an acquired planning kVCT. Paired‐sample Wilcoxon signed rank sum tests were used for statistical analysis to compare dosimetric parameters of plans made with pseudo CT images generated from deepMTP to those made with kVCT‐based clinical treatment plan (CTTP).ResultsdeepMTP provides an accurate pseudo CT with Dice coefficients for air: 0.95 ± 0.01, soft tissue: 0.94 ± 0.02, and bone: 0.85 ± 0.02 and a mean absolute error of 75 ± 23 HU compared with acquired kVCTs. The absolute percentage differences of dosimetric parameters between deepMTP and CTTP was 0.24% ± 0.46% for planning target volume (PTV) volume, 1.39% ± 1.31% for maximum dose and 0.27% ± 0.79% for the PTV receiving 95% of the prescribed dose (V95). Furthermore, no significant difference was found for PTV volume (P = 0.50), the maximum dose (P = 0.83) and V95 (P = 0.19) between deepMTP and CTTP.ConclusionsWe have developed an automated approach (deepMTP) that allows generation of a continuously valued pseudo CT from a single high‐resolution 3D MR image and evaluated it in partial brain tumor treatment planning. The deepMTP provided dose distribution with no significant difference relative to a kVCT‐based standard volumetric modulated arc therapy plans.

"Ears of the Lynx" MRI Sign Is Associated with SPG11 and SPG15 Hereditary Spastic Paraplegia.

The "ears of the lynx" MR imaging sign has been described in case reports of hereditary spastic paraplegia with a thin corpus callosum, mostly associated with mutations in the spatacsin vesicle trafficking associated gene, causing Spastic Paraplegia type 11 (SPG11). This sign corresponds to long T1 and T2 values in the forceps minor of the corpus callosum, which appears hyperintense on FLAIR and hypointense on T1-weighted images. Our purpose was to determine the sensitivity and specificity of the ears of the lynx MR imaging sign for genetic cases compared with common potential mimics. Four independent raters, blinded to the diagnosis, determined whether the ears of the lynx sign was present in each of a set of 204 single anonymized FLAIR and T1-weighted MR images from 34 patients with causal mutations associated with SPG11 or Spastic Paraplegia type 15 (SPG15). 34 healthy controls, and 34 patients with multiple sclerosis. The interrater reliability for FLAIR images was substantial (Cohen κ, 0.66-0.77). For these images, the sensitivity of the ears of the lynx sign across raters ranged from 78.8 to 97.0 and the specificity ranged from 90.9 to 100. The accuracy of the sign, measured by area under the receiver operating characteristic curve, ranged from very good (87.1) to excellent (93.9). The ears of the lynx sign on FLAIR MR imaging is highly specific for the most common genetic subtypes of hereditary spastic paraplegia with a thin corpus callosum. When this sign is present, there is a high likelihood of a genetic mutation, particularly associated with SPG11 or SPG15, even in the absence of a family history.

A Generalized Structured Low-Rank Matrix Completion Algorithm for MR Image Recovery.

Recent theory of mapping an image into a structured low-rank Toeplitz or Hankel matrix has become an effective method to restore images. In this paper, we introduce a generalized structured low-rank algorithm to recover images from their undersampled Fourier coefficients using infimal convolution regularizations. The image is modeled as the superposition of a piecewise constant component and a piecewise linear component. The Fourier coefficients of each component satisfy an annihilation relation, which results in a structured Toeplitz matrix. We exploit the low-rank property of the matrices to formulate a combined regularized optimization problem. In order to solve the problem efficiently and to avoid the high-memory demand resulting from the large-scale Toeplitz matrices, we introduce a fast and a memory-efficient algorithm based on the half-circulant approximation of the Toeplitz matrix. We demonstrate our algorithm in the context of single and multi-channel MR images recovery. Numerical experiments indicate that the proposed algorithm provides improved recovery performance over the state-of-the-art approaches.

Generating synthetic CTs from magnetic resonance images using generative adversarial networks.

While MR-only treatment planning using synthetic CTs (synCTs) offers potential for streamlining clinical workflow, a need exists for an efficient and automated synCT generation in the brain to facilitate near real-time MR-only planning. This work describes a novel method for generating brain synCTs based on generative adversarial networks (GANs), a deep learning model that trains two competing networks simultaneously, and compares it to a deep convolutional neural network (CNN). Post-Gadolinium T1-Weighted and CT-SIM images from fifteen brain cancer patients were retrospectively analyzed. The GAN model was developed to generate synCTs using T1-weighted MRI images as the input using a residual network (ResNet) as the generator. The discriminator is a CNN with five convolutional layers that classified the input image as real or synthetic. Fivefold cross-validation was performed to validate our model. GAN performance was compared to CNN based on mean absolute error (MAE), structural similarity index (SSIM), and peak signal-to-noise ratio (PSNR) metrics between the synCT and CT images. GAN training took ~11h with a new case testing time of 5.7±0.6s. For GAN, MAEs between synCT and CT-SIM were 89.3±10.3 Hounsfield units (HU) and 41.9±8.6HU across the entire FOV and tissues, respectively. However, MAE in the bone and air was, on average, ~240-255HU. By comparison, the CNN model had an average full FOV MAE of 102.4±11.1HU. For GAN, the mean PSNR was 26.6±1.2 and SSIM was 0.83±0.03. GAN synCTs preserved details better than CNN, and regions of abnormal anatomy were well represented on GAN synCTs. We developed and validated a GAN model using a single T1-weighted MR image as the input that generates robust, high quality synCTs in seconds. Our method offers strong potential for supporting near real-time MR-only treatment planning in the brain.

Middle patellar tendon to posterior cruciate ligament (PT–PCL) and normalized PT–PCL: New magnetic resonance indices for tibial tubercle position in patients with patellar instability

BackgroundTo demonstrate whether the distance between the middle point of the patellar tendon and posterior cruciate ligament (PT–PCL) calculated on a single axial MR image could be an alternative measure to tibial tubercle–PCL (TT–PCL) distance for TT lateralization without the need of imaging processing. To show that normalization of PT–PCL (nPT-PCL) against the maximum diameter of the tibial plateau may help to identify patients with patellar instability (PI). MethodsMR scans of 30 patients (13 females, age 32 ± 13 years) with known PI and 60 patients (31 females, age 39 ± 19 years) with no history of PI were reviewed. Two operators calculated TT–PCL, and PT–PCL nPT-PCL. Intraclass correlation coefficient, Student's t-test, Receiver Operator Characteristic curves, Spearman's Rho and McNemar's test were used. ResultsInterobserver reproducibility was 0.894 for PT-PCL for TT-PCL (95% CI = 0.839–0.930) and 0.866 for TT-PCL (95% CI = 0.796–0.912). The PT–PCL was 23.5 ± 3.8 mm in patients and 20.0 ± 2.7 mm in controls (P < 0.001). The TT–PCL was 22.9 ± 3.9 mm in patients and 20.5 ± 2.7 mm in controls (P = 0.002). Correlation between the PT–PCL and TT–PCL was R = 0.838, P < 0.001. The PT–PCL had 66.6% (95% CI = 0.542–0.790) diagnostic yield. The nPT–PCL was significantly higher in patients (0.302 ± 0.03) than controls (0.271 ± 0.03; P < 0.001) with 73.9% (95% CI = 0.628–0.851) diagnostic yield. ConclusionThe PT–PCL correlated with TT–PCL, with 66.6% diagnostic yield. The nPT–PCL may represent an additional index, with 73.9% diagnostic yield.