Undersampled Datasets Research Articles

Accelerating multi-coil MR image reconstruction using weak supervision.

Deep-learning-based MR image reconstruction in settings where large fully sampled dataset collection is infeasible requires methods that effectively use both under-sampled and fully sampled datasets. This paper evaluates a weakly supervised, multi-coil, physics-guided approach to MR image reconstruction, leveraging both dataset types, to improve both the quality and robustness of reconstruction. A physics-guided end-to-end variational network (VarNet) is pretrained in a self-supervised manner using a 4 under-sampled dataset following the self-supervised learning via data undersampling (SSDU) methodology. The pre-trained weights are transferred to another VarNet, which is fine-tuned using a smaller, fully sampled dataset by optimizing multi-scale structural similarity (MS-SSIM) loss in image space. The proposed methodology is compared with fully self-supervised and fully supervised training. Reconstruction quality improvements in SSIM, PSNR, and NRMSE when abundant training data is available (the high-data regime), and enhanced robustness when training data is scarce (the low-data regime) are demonstrated using weak supervision for knee and brain MR image reconstructions at 8 and 10 acceleration, respectively. Multi-coil physics-guided MR image reconstruction using both under-sampled and fully sampled datasets is achievable with transfer learning and fine-tuning. This methodology can provide improved reconstruction quality in the high-data regime and improved robustness in the low-data regime at high acceleration rates.

Rapid Hyperspectral Photothermal Mid-Infrared Spectroscopic Imaging from Sparse Data for Gynecologic Cancer Tissue Subtyping.

Ovarian cancer detection has traditionally relied on a multistep process that includes biopsy, tissue staining, and morphological analysis by experienced pathologists. While widely practiced, this conventional approach suffers from several drawbacks: it is qualitative, time-intensive, and heavily dependent on the quality of staining. Mid-infrared (MIR) hyperspectral photothermal imaging is a label-free, biochemically quantitative technology that, when combined with machine learning algorithms, can eliminate the need for staining and provide quantitative results comparable to traditional histology. However, this technology is slow. This work presents a novel approach to MIR photothermal imaging that enhances its speed by an order of magnitude. This method resolves the longstanding trade-off between imaging resolution and data collection speed, enabling the reconstruction of high-quality, high-resolution images from undersampled data sets and achieving a 10X improvement in data acquisition time. We assessed the performance of our sparse imaging methodology using a variety of quantitative metrics, including mean squared error (MSE), structural similarity index (SSIM), and tissue subtype classification accuracies, employing both random forest and convolutional neural network (CNN) models, accompanied by Receiver Operating Characteristic (ROC) curves. Our statistically robust analysis, based on data from 100 ovarian cancer patient samples and over 65 million data points, demonstrates the method's capability to produce superior image quality and accurately distinguish between different gynecological tissue types with segmentation accuracy exceeding 95%. Our work demonstrates the feasibility of integrating rapid MIR hyperspectral photothermal imaging with machine learning in enhancing ovarian cancer tissue characterization, paving the way for quantitative, label-free, automated histopathology.

Boosting efficiency in a clinical literature surveillance system with LightGBM.

Given the suboptimal performance of Boolean searching to identify methodologically sound and clinically relevant studies in large bibliographic databases, exploring machine learning (ML) to efficiently classify studies is warranted. To boost the efficiency of a literature surveillance program, we used a large internationally recognized dataset of articles tagged for methodological rigor and applied an automated ML approach to train and test binary classification models to predict the probability of clinical research articles being of high methodologic quality. We trained over 12,000 models on a dataset of titles and abstracts of 97,805 articles indexed in PubMed from 2012-2018 which were manually appraised for rigor by highly trained research associates and rated for clinical relevancy by practicing clinicians. As the dataset is unbalanced, with more articles that do not meet the criteria for rigor, we used the unbalanced dataset and over- and under-sampled datasets. Models that maintained sensitivity for high rigor at 99% and maximized specificity were selected and tested in a retrospective set of 30,424 articles from 2020 and validated prospectively in a blinded study of 5253 articles. The final selected algorithm, combining a LightGBM (gradient boosting machine) model trained in each dataset, maintained high sensitivity and achieved 57% specificity in the retrospective validation test and 53% in the prospective study. The number of articles needed to read to find one that met appraisal criteria was 3.68 (95% CI 3.52 to 3.85) in the prospective study, compared with 4.63 (95% CI 4.50 to 4.77) when relying only on Boolean searching. Gradient-boosting ML models reduced the work required to classify high quality clinical research studies by 45%, improving the efficiency of literature surveillance and subsequent dissemination to clinicians and other evidence users.

The Fine-Tuned Large Language Model for Extracting the Progressive Bone Metastasis from Unstructured Radiology Reports.

Early detection of patients with impending bone metastasis is crucial for prognosis improvement. This study aimed to investigate the feasibility of a fine-tuned, locally run large language model (LLM) in extracting patients with bone metastasis in unstructured Japanese radiology report and to compare its performance with manual annotation. This retrospective study included patients with "metastasis" in radiological reports (April 2018-January 2019, August-May 2022, and April-December 2023 for training, validation, and test datasets of 9559, 1498, and 7399 patients, respectively). Radiologists reviewed the clinical indication and diagnosis sections of the radiological report (used as input data) and classified them into groups 0 (no bone metastasis), 1 (progressive bone metastasis), and 2 (stable or decreased bone metastasis). The data for group 0 was under-sampled in training and test datasets due to group imbalance. The best-performing model from the validation set was subsequently tested using the testing dataset. Two additional radiologists (readers 1 and 2) were involved in classifying radiological reports within the test dataset for testing purposes. The fine-tuned LLM, reader 1, and reader 2 demonstrated an accuracy of 0.979, 0.996, and 0.993, sensitivity for groups 0/1/2 of 0.988/0.947/0.943, 1.000/1.000/0.966, and 1.000/0.982/0.954, and time required for classification (s) of 105, 2312, and 3094 in under-sampled test dataset (n = 711), respectively. Fine-tuned LLM extracted patients with bone metastasis, demonstrating satisfactory performance that was comparable to or slightly lower than manual annotation by radiologists in a noticeably shorter time.

Characteristic value decomposition: A unifying paradigm for data-driven modal analysis

Data-driven modal analysis is an indispensable means to understand the dynamic behavior of engineered structures and natural systems. It effectively captures active dynamic behavior and embeds them in a set of identified linear modal subspaces, unveiling complex dynamic behavior via the superposition of simpler hierarchical spatiotemporal dynamics. Despite the existence of many methods, there is no unified perspective connecting these modal analysis techniques. This paper proposes a new modal decomposition method, the characteristic value decomposition (CVD), that provides a unifying paradigm for data-driven modal analysis. CVD identifies identical spatial modes and temporal coordinates from two measurement data matrices. The modal parameters are identified through an eigendecomposition to a quotient of auto-covariance and cross-covariance matrices of the two data matrices. This paper shows that several data-driven modal analysis methods, namely, the state-variable decomposition, the dynamic mode decomposition, the Ibrahim time domain method, and the eigensystem realization algorithm, are special cases of CVD when the two data matrices are appropriately selected. Moreover, the CVD identifies the modal parameters with higher accuracy and is robust to added noise compared to established methods, as demonstrated using numerical experiments. It is also shown that the CVD flexibly utilizes data from multiple experiments and undersampled data sets to unveil system dynamics.

Accelerated 3D multi-echo spin-echo sequence with a subspace constrained reconstruction for whole mouse brain mapping.

To accelerate whole-brain quantitative mapping in preclinical imaging setting. A three-dimensional (3D) multi-echo spin echo sequence was highly undersampled with a variable density Poisson distribution to reduce the acquisition time. Advanced iterative reconstruction based on linear subspace constraints was employed to recover high-quality raw images. Different subspaces, generated using exponential or extended-phase graph (EPG) simulations or from low-resolution calibration images, were compared. The subspace dimension was investigated in terms of precision. The method was validated on a phantom containing a wide range of and was then applied to monitor metastasis growth in the mouse brain at 4.7T. Image quality and estimation were assessed for 3 acceleration factors (6/8/10). The EPG-based dictionary gave robust estimations of a large range of . A subspace dimension of 6 was the best compromise between precision and image quality. Combining the subspace constrained reconstruction with a highly undersampled dataset enabled the acquisition of whole-brain maps, the detection and the monitoring of metastasis growth of less than 500 . Subspace-based reconstruction is suitable for 3D mapping. This method can be used to reach an acceleration factor up to 8, corresponding to an acquisition time of 25 min for an isotropic 3D acquisition of 156 m on the mouse brain, used here for monitoring metastases growth.

ВИЯВЛЕННЯ ШАХРАЙСТВА З КРЕДИТНИМИ КАРТКАМИ МЕТОДАМИ МАШИННОГО НАВЧАННЯ

In the work, a study of large volumes of transactions was conducted. The problem of fraud detection is unique because it is necessary to take into account: data imbalance (ie, fraudulent transactions are usually less than 1% compared to normal ones); fraud scenarios change over time and need to be detected quickly; transactions usually contain numerous categorical characteristics; as a result of the confidentiality of transactions, there are no publicly available datasets. All this creates problems with the development of classification methods and with the selection of performance evaluation metrics. Threshold indicators for evaluating the effectiveness of classifiers were studied and the expediency of using thresholdless metrics was substantiated. There is currently no consensus on which set of performance indicators should be used. The primary analysis of research data was carried out and two techniques for eliminating class imbalance were applied: random undersampling, SMOTE technique. Removal of outliers was shown to improve the accuracy of the classification methods by more than 3%. A number of classifiers were built to determine fraudulent operations, statistical processing of the obtained results was carried out, which allowed to assess the adequacy of the built classifiers, to determine their optimal parameters, at which the classifiers work with maximum efficiency. It should be noted that all classifiers were tested on real data. Determined: The logistic regression classifier performs best on both the training and cross-validation sets. The Precision-Recall indicator was used to assess the effectiveness of the logistic regression model. For undersampling and oversampling, fully connected neural networks with one hidden layer were constructed and their accuracy was compared. It should be noted that the neural network on the oversampled data set predicts fewer correct fraud transactions than the model using the undersampled data set.

Comparison of Cluster-Based Sampling Approaches for Imbalanced Data of Crashes Involving Large Trucks

Severe and fatal crashes involving large trucks result in significant social and economic losses for human society. Unfortunately, the notably low proportion of severe and fatal injury crashes involving large trucks creates an imbalance in crash data. Models trained on imbalanced crash data are likely to produce erroneous results. Therefore, there is a need to explore novel sampling approaches for imbalanced crash data, and it is crucial to determine the appropriate combination of a machine learning model, sampling approach, and ratio. This study introduces a novel cluster-based under-sampling technique, utilizing the k-prototypes clustering algorithm. After initial cluster-based under-sampling, the consolidated cluster-based under-sampled data set was further resampled using three different sampling approaches (i.e., adaptive synthetic sampling (ADASYN), NearMiss-2, and the synthetic minority oversampling technique + Tomek links (SMOTETomek)). Later, four machine learning models (logistic regression (LR), random forest (RF), gradient-boosted decision trees (GBDT), and the multi-layer perceptron (MLP) neural network) were trained and evaluated using the geometric mean (G-Mean) and area under the receiver operating characteristic curve (AUC) scores. The findings suggest that cluster-based under-sampling coupled with the investigated sampling approaches improve the performance of the machine learning models developed on crash data significantly. In addition, the GBDT model combined with ADASYN or SMOTETomek is likely to yield better predictions than any model combined with NearMiss-2. Regarding changes in sampling ratios, increasing the sampling ratio with ADASYN and SMOTETomek is likely to improve the performance of models up to a certain level, whereas with NearMiss-2, performance is likely to drop significantly beyond a specific point. These findings provide valuable insights for selecting optimal strategies for treating the class imbalance issue in crash data.

Predicting Highway Risk Event with Trajectory Data: A Joint Approach of Traffic Flow and Vehicle Kinematics

Real-time collision risk prediction is essential for improving highway safety and reducing traffic accidents. However, previous studies have mainly used crash data and associated spatially discrete and temporally continuous traffic data, overlooking the potential of vehicle trajectory data, which provides comprehensive spatio-temporal information to characterize traffic near a specific location. Moreover, researchers have typically focused on either traffic flow characteristics or inter-vehicle microscopic kinematic characteristics for real-time risk prediction, with a dearth of studies integrating these two aspects. Given that risk events transpire more frequently than accidents and exhibit a strong correlation with them, it is imperative to concentrate more on risk events to proactively diminish crash probabilities. This study introduces a novel approach that extracts traffic flow and inter-vehicle kinematic features from risk events. It also provides a comparative analysis of the effectiveness of five machine-learning methods (Logistic Regression, K-Nearest Neighbors, eXtreme Gradient Boosting, Random Forests, and Multilayer Perceptron) and two data-processing strategies (oversampling and undersampling) in addressing risk identification and prediction issues. The results showed that (1) the synergistic use of traffic flow and inter-vehicle kinematic features surpasses the use of a single feature in identifying and predicting risks; (2) The eXtreme Gradient Boosting model, trained on the undersampled dataset, emerges as the optimal model for risk identification, boasting an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.976 and an F1 score of 0.604; (3) The RF model exhibits commendable performance under both risk prediction conditions (5 s ahead prediction and 10 s prediction), demonstrating the highest performance with F1 scores of 0.377 and 0.374, respectively. Additionally, it was discovered that the resampling strategy does not always prove effective in developing risk analysis models and should be chosen based on the model’s characteristics and target metrics. This offers valuable insights into the selection of data-processing strategies when handling unbalanced data. Finally, the study’s limitations and potential enhancements are discussed.

LAFEM: A Scoring Model to Evaluate Functional Landscape of Lysine Acetylome

Protein lysine acetylation is a critical post-translational modification involved in a wide range of biological processes. To date, about 20,000 acetylation sites of Homo sapiens were identified through mass spectrometry–based proteomic technology, but more than 95% of them have unclear functional annotations because of the lack of existing prioritization strategy to assess the functional importance of the acetylation sites on large scale. Hence, we established a lysine acetylation functional evaluating model (LAFEM) by considering eight critical features surrounding lysine acetylation site to high-throughput estimate the functional importance of given acetylation sites. This was achieved by selecting one of the random forest models with the best performance in 10-fold cross-validation on undersampled training dataset. The global analysis demonstrated that the molecular environment of acetylation sites with high acetylation functional scores (AFSs) mainly had the features of larger solvent-accessible surface area, stronger hydrogen bonding–donating abilities, near motif and domain, higher homology, and disordered degree. Importantly, LAFEM performed well in validation dataset and acetylome, showing good accuracy to screen out fitness directly relevant acetylation sites and assisting to explain the core reason for the difference between biological models from the perspective of acetylome. We further used cellular experiments to confirm that, in nuclear casein kinase and cyclin-dependent kinase substrate 1, acetyl-K35 with higher AFS was more important than acetyl-K9 with lower AFS in the proliferation of A549 cells. LAFEM provides a prioritization strategy to large scale discover the fitness directly relevant acetylation sites, which constitutes an unprecedented resource for better understanding of functional acetylome.

Anomaly Detection Based on a 3D Convolutional Neural Network Combining Convolutional Block Attention Module Using Merged Frames.

With the recent rise in violent crime, the real-time situation analysis capabilities of the prevalent closed-circuit television have been employed for the deterrence and resolution of criminal activities. Anomaly detection can identify abnormal instances such as violence within the patterns of a specified dataset; however, it faces challenges in that the dataset for abnormal situations is smaller than that for normal situations. Herein, using datasets such as UBI-Fights, RWF-2000, and UCSD Ped1 and Ped2, anomaly detection was approached as a binary classification problem. Frames extracted from each video with annotation were reconstructed into a limited number of images of 3×3, 4×3, 4×4, 5×3 sizes using the method proposed in this paper, forming an input data structure similar to a light field and patch of vision transformer. The model was constructed by applying a convolutional block attention module that included channel and spatial attention modules to a residual neural network with depths of 10, 18, 34, and 50 in the form of a three-dimensional convolution. The proposed model performed better than existing models in detecting abnormal behavior such as violent acts in videos. For instance, with the undersampled UBI-Fights dataset, our network achieved an accuracy of 0.9933, a loss value of 0.0010, an area under the curve of 0.9973, and an equal error rate of 0.0027. These results may contribute significantly to solve real-world issues such as the detection of violent behavior in artificial intelligence systems using computer vision and real-time video monitoring.

Structured low-rank reconstruction for navigator-free water/fat separated multi-shot diffusion-weighted EPI.

Multi-shot diffusion-weighted EPI allows an increase in image resolution and reduced geometric distortions and can be combined with chemical-shift encoding (Dixon) to separate water/fat signals. However, such approaches suffer from physiological motion-induced shot-to-shot phase variations. In this work, a structured low-rank-based navigator-free algorithm is proposed to address the challenge of simultaneously separating water/fat signals and correcting for physiological motion-induced shot-to-shot phase variations in multi-shot EPI-based diffusion-weighted MRI. We propose an iterative, model-based reconstruction pipeline that applies structured low-rank regularization to estimate and eliminate the shot-to-shot phase variations in a data-driven way, while separating water/fat images. The algorithm is tested in different anatomies, including head-neck, knee, brain, and prostate. The performance is validated in simulations and in-vivo experiments in comparison to existing approaches. In-vivo experiments and simulations demonstrated the effectiveness of the proposed algorithm compared to extra-navigated and an alternative self-navigation approach. The proposed algorithm demonstrates the capability to reconstruct in the multi-shot/Dixon hybrid space domain under-sampled datasets, using the same number of acquired EPI shots compared to conventional fat-suppression techniques but eliminating fat signals through chemical-shift encoding. In addition, partial Fourier reconstruction can also be achieved by using the concept of virtual conjugate coils in conjunction with the proposed algorithm. The proposed algorithm effectively eliminates the shot-to-shot phase variations and separates water/fat images, making it a promising solution for future DWI on different anatomies.

Deep learning reconstruction of pressure fluctuations in supersonic shock–boundary layer interaction

The long short-term memory deep-learning model is applied to supersonic shock–boundary layer interaction flow. The study aims to show how near-wall pressure fluctuations can be reconstructed from reduced (under-sampled) datasets of pressure signals. Predicting pressure fluctuations from reduced datasets could allow predictions using less expensive simulations and experiments. The training of the deep learning model is based on direct numerical simulations of supersonic ramp flows, focusing on the regions upstream of and around the shock–boundary layer interaction region. During the pre-processing stage, cubic spline functions increase the fidelity of the sparse signals and feed them to the long-short memory model for an accurate reconstruction. Comparisons are also carried out for different sparsity factors and assess the model's accuracy both qualitatively through the pressure signals and quantitatively using the root mean square error and the power spectra. The deep learning predictions are promising and can be extended to include other aerodynamic or aeroelastic parameters of interest.

Malware Image Classification Using Deep Learning InceptionResNet-V2 and VGG-16 Method

Malware is intentionally designed to damage computers, servers, clients or computer networks. Malware is a general term used to describe any program designed to harm a computer or server. The goal is to commit a crime, such as gaining unauthorized access to a particular system, so as to compromise user security. Most malware still uses the same code to produce another different form of malware variants. Therefore, the ability to classify similar malware variant characteristics into malware families is a good strategy to stop malware. The research is useful for classifying malware on malware samples presented as bytemap grayscale images. The malware classification research focused on 25 malware classes with a total of 9,029 images from the Malimg dataset. This research implements the VGG-16 and InceptionResNet-V2 architectures by running 2 different scenarios, scenario 1 uses the original dataset and the other scenario uses the undersampled dataset. After building the model, each scenario will get an evaluation form such as accuracy, precision, recall, and f1-score. The highest score was obtained in scenario 2 on the VGG-16 method with a score of 94.8% and the lowest in scenario 2 on the InceptionResNet-V2 method with a score of 85.1%.

Reduction of SPECT acquisition time using deep learning: A phantom study

Single photon emission computed tomography (SPECT) procedures are characterized by long acquisition time to acquire diagnostically acceptable image data. The goal of this investigation was to assess the feasibility of using a deep convolutional neural network (DCNN) to reduce the acquisition time. The DCNN was implemented using the PyTorch and trained using image data from standard SPECT quality phantoms. The under-sampled image dataset is provided to neural network as input, while missing projections were provided as targets. The network is to produce for the output the missing projections. The baseline method of calculating the missing projections as arithmetic means of adjacent ones was introduced. The obtained synthesized projections and reconstructed images were compared to original data and baseline data across several parameters using PyTorch and PyTorch Image Quality code libraries. Results obtained from comparisons of projection and reconstructed image data show the DCNN clearly outperforming the baseline method. However, subsequent analysis revealed the synthesized image data being more comparable to under-sampled than to fully-sampled image data. The results of this investigation imply that neural network can replicate coarser objects better. However, densely sampled clinical image datasets, coarse reconstruction matrices and patient data featuring coarse structures combined with a lack of baseline data generation methods will hamper the ability to analyse the neural network outputs correctly. This study calls for use of phantom image data and introduction of a baseline method in the evaluation of neural network outputs.

Fast Low Rank Column-Wise Compressive Sensing for Accelerated Dynamic MRI

This work develops a novel set of algorithms, alternating Gradient Descent (GD) and minimization for MRI (altGDmin-MRI1 and altGDmin-MRI2), for accelerated dynamic MRI by assuming an approximate low-rank (LR) model on the matrix formed by the vectorized images of the sequence. The LR model itself is well-known in the MRI literature; our contribution is the novel GD-based algorithms which are much faster, memory-efficient, and ‘general’ compared with existing work; and careful use of a 3-level hierarchical LR model. By ‘general,’ we mean that, with a single choice of parameters, our method provides accurate reconstructions for multiple accelerated dynamic MRI applications, multiple sampling rates and sampling schemes. We show that our methods outperform many of the popular existing approaches while also being faster than all of them, on average. This claim is based on comparisons on 8 different retrospectively undersampled multi-coil dynamic MRI applications, sampled using either 1D Cartesian or 2D pseudo-radial undersampling, at multiple sampling rates. Evaluations on some prospectively undersampled datasets are also provided. Our second contribution is a mini-batch subspace tracking extension that can process new measurements and return reconstructions within a short delay after they arrive. The recovery algorithm itself is also faster than its batch counterpart.

Semi-Supervised Learning of MRI Synthesis Without Fully-Sampled Ground Truths.

Learning-based translation between MRI contrasts involves supervised deep models trained using high-quality source- and target-contrast images derived from fully-sampled acquisitions, which might be difficult to collect under limitations on scan costs or time. To facilitate curation of training sets, here we introduce the first semi-supervised model for MRI contrast translation (ssGAN) that can be trained directly using undersampled k-space data. To enable semi-supervised learning on undersampled data, ssGAN introduces novel multi-coil losses in image, k-space, and adversarial domains. The multi-coil losses are selectively enforced on acquired k-space samples unlike traditional losses in single-coil synthesis models. Comprehensive experiments on retrospectively undersampled multi-contrast brain MRI datasets are provided. Our results demonstrate that ssGAN yields on par performance to a supervised model, while outperforming single-coil models trained on coil-combined magnitude images. It also outperforms cascaded reconstruction-synthesis models where a supervised synthesis model is trained following self-supervised reconstruction of undersampled data. Thus, ssGAN holds great promise to improve the feasibility of learning-based multi-contrast MRI synthesis.

Calibrated Regression Models Based on the Risk of Clinical Nodal Metastasis Should be Used as Decision Aids for Prostate Cancer Staging to Reduce Unnecessary Imaging.

Radionuclide imaging will change the role of computed tomography and magnetic resonance imaging (CT/MRI) for prostate cancer (CaP) staging. Current guidelines recommend abdominopelvic imaging for new cases of CaP categorized as unfavorable intermediate risk (UIR) or higher. We assessed the performance characteristics of CT/MRI based on the National Comprehensive Cancer Network (NCCN) guidelines and developed a model that predicts cN1 disease using conventional imaging. We selected patients in the National Cancer Database diagnosed with CaP from 2010 to 2016 with available age, prostate specific antigen, clinical locoregional staging, biopsy Gleason grading, and core information. Multivariate logistic regression (MLR) was used on a undersampled training dataset using cN1 as the outcome. Performance characteristics were compared to those of the three most recent versions of the NCCN guidelines. A total of 443,640 men were included, and 2.5% had cN1 disease. Using CT/MRI only, the current NCCN guidelines have a sensitivity of 99%, and the number needed to image (NNI) is 24. At the same sensitivity, the cN1 risk was 1.6% using the MLR. The NNI for UIR alone is 341. Using the MLR model and a threshold of 10%, the PPV is 10.3% and 64% of CTs/MRIs could be saved at a cost of missing 6% of cN1 patients (or 0.15% of all patients). The NCCN guidelines are sensitive for detecting cN1 with CT/MRI, however, the number needed to image is 24. Obtaining CT/MRI for nodal staging when patients have a cN1 risk of 10% would reduce total imaging while still remaining sensitive. As novel PET tracers becomes increasingly used for initial CaP staging, well calibrated prediction models trained on the outcome of interest should be developed as decision aids for obtaining imaging.

Prospectively Accelerated T2-Weighted Imaging of the Prostate by Combining Compressed SENSE and Deep Learning in Patients with Histologically Proven Prostate Cancer

To assess the performance of prospectively accelerated and deep learning (DL) reconstructed T2-weighted (T2w) imaging in volunteers and patients with histologically proven prostate cancer (PCa). Prospectively undersampled T2w datasets were acquired with acceleration factors of 1.7 (reference), 3.4 and 4.8 in 10 healthy volunteers and 23 patients with histologically proven PCa. Image reconstructions using compressed SENSE (C-SENSE) and a combination of C-SENSE and DL-based artificial intelligence (C-SENSE AI) were analyzed. Qualitative image comparison was performed using a 6-point Likert scale (overall image quality, noise, motion artifacts, lesion detection, diagnostic certainty); the T2 and PI-RADS scores were compared between the two reconstructions. Additionally, quantitative image parameters were assessed (apparent SNR, apparent CNR, lesion size, line profiles). All C-SENSE AI-reconstructed images received a significantly higher qualitative rating compared to the C-SENSE standard images. Analysis of the quantitative parameters supported this finding, with significantly higher aSNR and aCNR. The line profiles demonstrated a significantly steeper signal change at the border of the prostatic lesion and the adjacent normal tissue in the C-SENSE AI-reconstructed images, whereas the T2 and PI-RADS scores as well as the lesion size did not differ. In this prospective study, we demonstrated the clinical feasibility of a novel C-SENSE AI reconstruction enabling a 58% acceleration in T2w imaging of the prostate while obtaining significantly better image quality.

Accelerated two-dimensional phase-contrast for cardiovascular MRI using deep learning-based reconstruction with complex difference estimation.

To develop and validate a deep learning-based reconstruction framework for highly accelerated two-dimensional (2D) phase contrast (PC-MRI) data with accurate and precise quantitative measurements. We propose a modified DL-ESPIRiT reconstruction framework for 2D PC-MRI, comprised of an unrolled neural network architecture with a Complex Difference estimation (CD-DL). CD-DL was trained on 155 fully sampled 2D PC-MRI pediatric clinical datasets. The fully sampled data ( ) was retrospectively undersampled (6-11 ) and reconstructed using CD-DL and a parallel imaging and compressed sensing method (PICS). Measurements of peak velocity and total flow were compared to determine the highest acceleration rate that provided accuracy and precision within . Feasibility of CD-DL was demonstrated on prospectively undersampled datasets acquired in pediatric clinical patients ( ) and compared to traditional parallel imaging (PI) and PICS. The retrospective evaluation showed that 9 accelerated 2D PC-MRI images reconstructed with CD-DL provided accuracy and precision (bias, [95 confidence intervals]) within . CD-DL showed higher accuracy and precision compared to PICS for measurements of peak velocity (2.8 [ , 4.5] vs. 3.9 [ , 4.9]) and total flow (1.8 [ , 3.4] vs. 2.9 [ , 6.9]). The prospective feasibility study showed that CD-DL provided higher accuracy and precision than PICS for measurements of peak velocity and total flow. In a retrospective evaluation, CD-DL produced quantitative measurements of 2D PC-MRI peak velocity and total flow with error in both accuracy and precision for up to 9 acceleration. Clinical feasibility was demonstrated using a prospective clinical deployment of our 8 undersampled acquisition and CD-DL reconstruction in a cohort of pediatric patients.