U-Net Convolutional Neural Network Research Articles

Deep learning to overcome Zernike phase-contrast nanoCT artifacts for automated micro-nano porosity segmentation in bone.

Bone material contains a hierarchical network of micro- and nano-cavities and channels, known as the lacuna-canalicular network (LCN), that is thought to play an important role in mechanobiology and turnover. The LCN comprises micrometer-sized lacunae, voids that house osteocytes, and submicrometer-sized canaliculi that connect bone cells. Characterization of this network in three dimensions is crucial for many bone studies. To quantify X-ray Zernike phase-contrast nanotomography data, deep learning is used to isolate and assess porosity in artifact-laden tomographies of zebrafish bones. A technical solution is proposed to overcome the halo and shade-off domains in order to reliably obtain the distribution and morphology of the LCN in the tomographic data. Convolutional neural network (CNN) models are utilized with increasing numbers of images, repeatedly validated by `error loss' and `accuracy' metrics. U-Net and Sensor3D CNN models were trained on data obtained from two different synchrotron Zernike phase-contrast transmission X-ray microscopes, the ANATOMIX beamline at SOLEIL (Paris, France) and the P05 beamline at PETRA III (Hamburg, Germany). The Sensor3D CNN model with a smaller batch size of 32 and a training data size of 70 images showed the best performance (accuracy 0.983 and error loss 0.032). The analysis procedures, validated by comparison with human-identified ground-truth images, correctly identified the voids within the bone matrix. This proposed approach may have further application to classify structures in volumetric images that contain non-linear artifacts that degrade image quality and hinder feature identification.

Impact of training data composition on the generalizability of convolutional neural network aortic cross-section segmentation in four-dimensional magnetic resonance flow imaging

BackgroundTime-resolved, three-dimensional phase-contrast magnetic resonance imaging (4D flow MRI) plays an important role in assessing cardiovascular diseases. However, the manual or semi-automatic segmentation of aortic vessel boundaries in 4D flow data introduces variability and limits reproducibility of aortic hemodynamics visualization and quantitative flow-related parameter computation. This paper explores the potential of deep learning to improve 4D flow MRI segmentation by developing models for automatic segmentation and analyzes the impact of the training data on the generalization of the model across different sites, scanner vendors, sequences, and pathologies. MethodsThe study population consists of 260 4D flow MRI datasets, including subjects without known aortic pathology, healthy volunteers, and patients with bicuspid aortic valve (BAV) examined at different hospitals. The dataset was split to train segmentation models on subsets with different representations of characteristics such as pathology, gender, age, scanner model, vendor, and field strength. An enhanced 3D U-net convolutional neural network (CNN) architecture with residual units was trained for 2D+t aortic cross-sectional segmentation. The model performance was evaluated using Dice score, Hausdorff distance, and average symmetric surface distance on test data, datasets with characteristics not represented in the training set (model-specific), and an overall evaluation set. Standard diagnostic flow parameters were computed and compared with manual segmentation results using Bland-Altman analysis and interclass correlation. ResultsThe representation of technical factors such as scanner vendor and field strength in the training dataset had the strongest influence on the overall segmentation performance. Age had a greater impact than gender. Models solely trained on BAV patients’ datasets performed well on datasets of healthy subjects but not vice versa. ConclusionThis study highlights the importance of considering a heterogeneous dataset for the training of widely applicable automatic CNN segmentations in 4D flow MRI, with a particular focus on the inclusion of different pathologies and technical aspects of data acquisition.

An overview of image processing in biomedicine using U-Net convolutional neural network architecture

Image processing in biomedicine is a very broad field, which includes both medical and technical significance. The aim of this paper is to investigate the current trends in the domain of application of U-Net architecture in the period from 2018 to 2023. The PRISMA framework was used for the systematic literature review, and 4 research questions were asked. For the most part, U-Net architectures are used that can process complex high-resolution images in the fastest way in the context of semantic segmentation. Previous work in image processing has focused on overcoming problems such as the complexity of different architectures, image loss, image resolution, and quality, as well as the size of datasets and noise reduction. The most frequently used groups of datasets are BraTS, Data Science Bowl, and ISIC Challenge. The best general Dice score was obtained for the LUNA16, VESSEL12, and Kaggle Lung datasets with 0.98. It is concluded that the application of the U-Net network is growing, with a focus on solving specific challenges in the context of a certain modality and segment of biomedicine.

Data augmentation for 3D seismic fault interpretation using deep learning

Manual seismic interpretation of faults is a tedious and complicated process, which is prone to human error and bias. A semi-automatic approach for interpreting faults on seismic data is to use attributes that highlight discontinuities. These methods need to be optimized by the interpreter, thus constantly rely on the interpreter's knowledge. Recently, Machine Learning (ML) techniques in general and Convolutional Neural Networks (CNN) as part of Deep Neural Networks (DNN), in particular, have been used to detect and image faults on seismic data in order to make the process more automated and accurate. Supervised CNN learns and evolves from manually annotated or labeled fault interpretations.In this study, we have applied supervised 2D CNN to image faults on seismic data through binary segmentation. The study was performed on 3D marine seismic surveys in off-shore Norway collected from three separate locations along the Norwegian Continental Shelf, utilizing Efficient UNET CNN architecture. We applied data augmentation (geometric transformations) and hyperparameter tuning to improve the learning process and performance of the deep learning algorithm.We used noise content, acquisition type, imaging type, and fault scale as the main criteria to select seismic surveys. The application of data augmentation to the training and testing data generally led to improvement in the performance of 2D CNN on fault predictions, although the amount of improvement varied with respect to different surveys. The initial 2D CNN fault prediction improvement mainly relied on the quality, and size of faults present in the 3D seismic volumes. Further, improvement was achieved by the adjustment of certain hyperparameters affecting the training and testing process of the 2D CNN. However, we found little to no improvement on one seismic volume containing high levels of noise.

A novel deep learning method for large-scale analysis of bone marrow adiposity using UK Biobank Dixon MRI data

BackgroundBone marrow adipose tissue (BMAT) represents > 10% fat mass in healthy humans and can be measured by magnetic resonance imaging (MRI) as the bone marrow fat fraction (BMFF). Human MRI studies have identified several diseases associated with BMFF but have been relatively small scale. Population-scale studies therefore have huge potential to reveal BMAT’s true clinical relevance. The UK Biobank (UKBB) is undertaking MRI of 100,000 participants, providing the ideal opportunity for such advances. ObjectiveTo establish deep learning for high-throughput multi-site BMFF analysis from UKBB MRI data. Materials and methodsWe studied males and females aged 60–69. Bone marrow (BM) segmentation was automated using a new lightweight attention-based 3D U-Net convolutional neural network that improved segmentation of small structures from large volumetric data. Using manual segmentations from 61–64 subjects, the models were trained to segment four BM regions of interest: the spine (thoracic and lumbar vertebrae), femoral head, total hip and femoral diaphysis. Models were tested using a further 10–12 datasets per region and validated using datasets from 729 UKBB participants. BMFF was then quantified and pathophysiological characteristics assessed, including site- and sex-dependent differences and the relationships with age, BMI, bone mineral density, peripheral adiposity, and osteoporosis. ResultsModel accuracy matched or exceeded that for conventional U-Nets, yielding Dice scores of 91.2% (spine), 94.5% (femoral head), 91.2% (total hip) and 86.6% (femoral diaphysis). One case of severe scoliosis prevented segmentation of the spine, while one case of Non-Hodgkin Lymphoma prevented segmentation of the spine, femoral head and total hip because of T2 signal depletion; however, successful segmentation was not disrupted by any other pathophysiological variables. The resulting BMFF measurements confirmed expected relationships between BMFF and age, sex and bone density, and identified new site- and sex-specific characteristics. ConclusionsWe have established a new deep learning method for accurate segmentation of small structures from large volumetric data, allowing high-throughput multi-site BMFF measurement in the UKBB. Our findings reveal new pathophysiological insights, highlighting the potential of BMFF as a novel clinical biomarker. Applying our method across the full UKBB cohort will help to reveal the impact of BMAT on human health and disease.

A CNN-based strategy to automate contour detection of the hip and proximal femur using DXA hip images from longitudinal databases (CLSA and CaMos)

ABSTRACT Hip fractures contribute significantly to mortality in older adults. New methods to identify those at risk use dual-energy X-ray absorptiometry (DXA) images and advanced image processing. However, DXA images have an overlapping femur and pelvis and may contain boundary lines, making automation challenging. Herein, a 5-layer U-net convolutional neural network (CNN) was developed to segment the femur from hip DXA images. Images were used from the Canadian Longitudinal Study on Aging (CLSA, N=104) and Canadian Multicentre Osteoporosis Study (CaMos, N=105) databases for training, with manual contour drawing defining the ‘true output’ of each image. An algorithm was then developed to mask each hip image, and trained to predict subsequent masks. The CNN was tested with 44 additional CLSA images and 42 CaMos images. This proposed approach had an accuracy and intersection over union (IoU) of 97% and 0.57, and 93% and 0.51, for CaMos and CLSA scans, respectively. Furthermore, a series of augmentation techniques was applied to increase the data size, with accuracies of 96% and 94%, and IoU of 0.53 and 0.50. Overall, our strategy automatically determined the contour of the proximal femur using various clinical DXA images, a key step to automate fracture risk assessment in clinical practice.

Variation Segmentation Layer in Deep Learning Network for SPECT Images Lesion Segmentation

Functional imaging, particularly SPECT Iodine-131 ablation imaging, has gained recognition as a useful clinical tool for diagnosing, treating, assessing as well as avoiding a variety of disorders, which includes metastasis. Nonetheless, SPECT imaging is conspicuously characterized by low resolution, high sensitivity, limited specificity, and a low signal-to-noise ratio. This is caused by the imaging data's visually similar characteristics of lesions amongst diseases. Concentrating on the automated diagnosis of diseases with SPECT Iodine-131 ablation imaging, in this work, three types of segmentation layers are used. This comprises a pixel classification layer, dice classification layer, and focal loss layer that will be tested to determine which segmentation layer is high for auto-segmentation lesions on SPECT Iodine-131 ablation imaging. The data preprocessing, which mostly entails data augmentation, is initially carried out to address the issue of small SPECT image sample sizes by using the geometric transformation operation. Deep Designer Network App was used to develop a 3D U-Net Convolutional Neural Network (CNN). The dice classification layer shows the highest accuracy for the thyroid uptake data set, which is 42.34, 0.7333, and 0.5789 for RMSD, DSC, and IoU, respectively. There is significance in using the dice classification layer in data sets that have various forms of ground truth labeling. On the other hand, the pixel classification layer is promising and workable for the multi-disease, multi-lesion classification task of SPECT Iodine-131 ablation imaging with a huge data set training.

Accelerating phase-field simulation of three-dimensional microstructure evolution in laser powder bed fusion with composable machine learning predictions

Phase-field (PF) modeling is a versatile physics-based computational method that has been used to simulate the evolution of microstructures. The PF method can produce accurate microstructures but suffers from a high computational cost, limiting its use in length scales relevant to additive manufacturing. Using small-scale PF simulations as training data, we trained a surrogate machine learning (ML) model as a computationally cheaper alternative. We use a three-dimensional (3D) U-Net convolutional neural network and learn microstructure evolution in a supervised fashion. With initial microstructure and thermal history as inputs, the ML model can predict the resulting grain orientations at a high accuracy compared to the PF model. Computationally, the ML model is orders of magnitudes faster than the direct PF simulation in a GPU implementation, and scales favorably with increasing number of cells. By spatio-temporally composing multiple ML model predictions at the small-scale, we demonstrate a large-scale 64-layer simulation of a 2 mm × 2 mm × 2 mm cube for a powder bed fusion additive manufacturing process. The ML results revealed a mixture of equiaxed, columnar, and curved grains, comparable to experimental observations. Microstructures resulting from different toolpath strategies, and lack of fusion defects are also demonstrated. This approach paves the way for microstructure-driven process design.

Denoising coherent Doppler lidar data based on a U-Net convolutional neural network.

The coherent Doppler wind lidar (CDWL) has long been thought to be the most suitable technique for wind remote sensing in the atmospheric boundary layer (ABL) due to its compact size, robust performance, and low-cost properties. However, as the coherent lidar exploits the Mie scattering from aerosol particles, the signal intensity received by the lidar is highly affected by the concentration of aerosols. Unlike air molecules, the concentration of aerosol varies greatly with time and weather, and decreases dramatically with altitude. As a result, the performance of the coherent lidar fluctuates greatly with time, and the detection range is mostly confined within the planetary boundary layer. The original data collected by the lidar are first transformed into a spectrogram and then processed into radial wind velocities utilizing algorithms such as a spectral centroid. When the signal-to-noise ratio (SNR) is low, these classic algorithms fail to retrieve the wind speed stably. In this work, a radial wind velocity retrieving algorithm based on a trained convolutional neural network (CNN) U-Net is proposed for denoising and an accurate estimate of the Doppler shift in a low-SNR regime. The advantage of the CNN is first discussed qualitatively and then proved by means of a numerical simulation. Simulated spectrum data are used for U-Net training and testing, which show that the U-Net is not only more accurate than the spectral centroid but also achieves a further detection range. Finally, joint observation data from the lidar and radiosonde show excellent agreement, demonstrating that the U-Net-based retrieving algorithm has superior performance over the traditional spectral centroid method both in accuracy and detection range.

Fully automatic system to detect and segment the proximal femur in pelvic radiographic images for Legg-Calvé-Perthes disease.

This study aimed to develop a method using computer vision techniques to accurately detect and delineate the proximal femur in radiographs of Legg-Calvé-Perthes disease (LCPD) patients. Currently, evaluating femoral head deformity, a crucial predictor of LCPD outcomes, relies on unreliable categorical and qualitative classifications. To address this limitation, we employed the pretrained object detection model YOLOv5 to detect the proximal femur on over 2000 radiographs, including images of shoulders and chests, to enhance robustness and generalizability. Subsequently, we utilized the U-Net convolutional neural network architecture for image segmentation of the proximal femur in more than 800 manually annotated images of stage IV LCPD. The results demonstrate outstanding performance, with the object detection model achieving high accuracy (mean average precision of 0.99) and the segmentation model attaining an accuracy score of 91%, dice coefficient of 0.75, and binary IoU score of 0.85 on the held-out test set. The proposed fully automatic proximal femur detection and segmentation system offers a promising approach to accurately detect and delineate the proximal femoral bone contour in radiographic images, which is essential for further image analysis in LCPD patients.Clinical significance: This study highlights the potential of computer vision techniques for enhancing the reliability ofLegg-Calvé-Perthes diseasestaging and outcome prediction.

Vegetation coverage precisely extracting and driving factors analysis in drylands

Fractional Vegetation Coverage (FVC) is an essential indicator that captures variations in vegetation and documents the impacts of climate change and human activity for environmental assessment. However, conventional methods encounter challenges in accurately extracting fine-scale FVC in drylands due to the vegetation distribution being very heterogeneous in space with patches and inter-patches. Using the lower Tarim River Basin as a typical study case, we investigated three deep convolutional neural networks—Unet, Pspnet, and Deeplabv3 + —to generate high-precision FVC in drylands with high-resolution (0.8 m) remote sensing images. Among these models, the Unet model performed better, with an accuracy of 93.38%, while the accuracy of Pspnet and Deeplabv3+ was 88.14% and 88.91%, respectively. Comparison with the FVC derived from normalized difference vegetation index (NDVI), and land use/land cover data from ESRI and ESA indicated that the FVC map produced by Unet was more consistent with on-site field observations. Delving into drivers influencing dryland FVC, we found that groundwater depth plays a pivotal role compared to topographical and climatic variables. Specifically, when the groundwater depth exceeds −3 m, the probability of occurring high FVC is reduced to 50%. This study innovatively extracted the FVC of drylands with high vegetation spatial heterogeneity, which better solves the insufficient accuracy of the existing dataset, serves as a valuable reference for monitoring vegetation change, and facilitates more precise quantification of carbon storage.

Domain independent post-processing with graph U-nets: applications to electrical impedance tomographic imaging⋆ ⋆SH and BH were supported by the National Institute Of Biomedical Imaging And Bioengineering of the National Institutes of Health under Award Number R21EB028064. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. AH

Objective. To extend the highly successful U-Net Convolutional Neural Network architecture, which is limited to rectangular pixel/voxel domains, to a graph-based equivalent that works flexibly on irregular meshes; and demonstrate the effectiveness on electrical impedance tomography (EIT). Approach. By interpreting the irregular mesh as a graph, we develop a graph U-Net with new cluster pooling and unpooling layers that mimic the classic neighborhood based max-pooling important for imaging applications. Main results. The proposed graph U-Net is shown to be flexible and effective for improving early iterate total variation (TV) reconstructions from EIT measurements, using as little as the first iteration. The performance is evaluated for simulated data, and on experimental data from three measurement devices with different measurement geometries and instrumentations. We successfully show that such networks can be trained with a simple two-dimensional simulated training set, and generalize to very different domains, including measurements from a three-dimensional device and subsequent 3D reconstructions. Significance. As many inverse problems are solved on irregular (e.g. finite element) meshes, the proposed graph U-Net and pooling layers provide the added flexibility to process directly on the computational mesh. Post-processing an early iterate reconstruction greatly reduces the computational cost which can become prohibitive in higher dimensions with dense meshes. As the graph structure is independent of ‘dimension’, the flexibility to extend networks trained on 2D domains to 3D domains offers a possibility to further reduce computational cost in training.

Artifact Augmentation for Enhanced Tissue Detection in Microscope Scanner Systems.

As the field of routine pathology transitions into the digital realm, there is a surging demand for the full automation of microscope scanners, aiming to expedite the process of digitizing tissue samples, and consequently, enhancing the efficiency of case diagnoses. The key to achieving seamless automatic imaging lies in the precise detection and segmentation of tissue sample regions on the glass slides. State-of-the-art approaches for this task lean heavily on deep learning techniques, particularly U-Net convolutional neural networks. However, since samples can be highly diverse and prepared in various ways, it is almost impossible to be fully prepared for and cover every scenario with training data. We propose a data augmentation step that allows artificially modifying the training data by extending some artifact features of the available data to the rest of the dataset. This procedure can be used to generate images that can be considered synthetic. These artifacts could include felt pen markings, speckles of dirt, residual bubbles in covering glue, or stains. The proposed approach achieved a 1-6% improvement for these samples according to the F1 Score metric.

U-Net Performance for Beach Wrack Segmentation: Effects of UAV Camera Bands, Height Measurements, and Spectral Indices

This study delves into the application of the U-Net convolutional neural network (CNN) model for beach wrack (BW) segmentation and monitoring in coastal environments using multispectral imagery. Through the utilization of different input configurations, namely, “RGB”, “RGB and height”, “5 bands”, “5 bands and height”, and “Band ratio indices”, this research provides insights into the optimal dataset combination for the U-Net model. The results indicate promising performance with the “RGB” combination, achieving a moderate Intersection over Union (IoU) of 0.42 for BW and an overall accuracy of IoU = 0.59. However, challenges arise in the segmentation of potential BW, primarily attributed to the dynamics of light in aquatic environments. Factors such as sun glint, wave patterns, and turbidity also influenced model accuracy. Contrary to the hypothesis, integrating all spectral bands did not enhance the model’s efficacy, and adding height data acquired from UAVs decreased model precision in both RGB and multispectral scenarios. This study reaffirms the potential of U-Net CNNs for BW detection, emphasizing the suitability of the suggested method for deployment in diverse beach geomorphology, requiring no high-end computing resources, and thereby facilitating more accessible applications in coastal monitoring and management.

Multi-class classification of thyroid nodules from automatic segmented ultrasound images: Hybrid ResNet based UNet convolutional neural network approach

Background and objectivesEarly detection and diagnosis of thyroid nodule types are important because they can be treated more effectively in their early stages. The types of thyroid nodules are generally stated as atypia of undetermined significance/follicular lesion of undetermined significance (AUS/FLUS), benign follicular, and papillary follicular. The risk of malignancy for AUS/FLUS is typically stated to be between 5 and 15 %, while some studies indicate a risk as high as 25 %. Without complete histology, it is difficult to classify nodules and these diagnostic operations are pricey and risky. To minimize laborious workload and misdiagnosis, recently various AI-based decision support systems have been developed. MethodsIn this study, a novel AI-based decision support system has been developed for the automated segmentation and classification of the types of thyroid nodules. This system is based on a hybrid deep-learning procedure that makes both an automatic thyroid nodule segmentation and classification tasks, respectively. In this framework, the segmentation is executed with some U-Net architectures such as ResUNet and ResUNet++ integrating with the feature extraction and upsampling with dropout operations to prevent overfitting. The nodule classification task is achieved by various deep nets architecture such as VGG-16, DenseNet121, ResNet-50, and Inception ResNet-v2 considering some accurate classification criteria such as Intersection over Union (IOU), Dice coefficient, accuracy, precision, and recall. ResultsIn analysis, a total of 880 patients with ages ranging from 10 to 90 years were included by taking the ultrasound images and demographics. The experimental evaluations showed that ResUNet++ demonstrated excellent segmentation outcomes, attaining remarkable evaluation scores including a dice coefficient of 92.4 % and a mean IOU of 89.7 %. ResNet-50 and Inception ResNet-v2 trained over the images segmented with UNets have shown better performance in terms of achieving high evaluation scores for the classification accuracy such as 96.6 % and 95.0 %, respectively. In addition, ResNet-50 and Inception ResNet-v2 classified AUS/FLUS from the images segmented with UNets with AUC=97.0 % and 96.0 %, respectively. ConclusionsThe proposed AI-based decision support system improves the automatic segmentation performance of AUS/FLUS and it has shown better performance than available approaches in the literature with respect to ACC, Jaccard and DICE losses. This system has great potential for clinical use by both radiologists and surgeons as well.

Circular economy oriented future building information processing: PSO for CNN approach

With the acceleration of urbanization, resource depletion is becoming increasingly serious. The innovative strategy of resource utilization is of great significance to the development of new green building. Therefore, this paper combines artificial intelligence technology and deeply excavates the basic information of the old building images collected by the survey. It determines the specific future building construction information from the direction of sustainable development. Based on the traditional convolutional neural network U-Net model, this paper improves its shortcomings, proposes an improved IU-Net model and optimizes the full convolutional neural network structure based on Particle Swarm Optimization (PSO). This paper analyzes and designs the future building information system platform oriented to circular economy, and it compares the improved IU-Net model with other models through experiments to obtain the accuracy and other indicators. The experimental results indicate that the system using the IU-Net model of full convolutional neural network based on PSO has higher prediction accuracy. It has better functionality for the analysis and decision-making of future building.

U-net convolutional neural network applied to progressive fibrotic interstitial lung disease: Is progression at CT scan associated with a clinical outcome?

BackgroundComputational advances in artificial intelligence have led to the recent emergence of U-Net convolutional neural networks (CNNs) applied to medical imaging. Our objectives were to assess the progression of fibrotic interstitial lung disease (ILD) using routine CT scans processed by a U-Net CNN developed by our research team, and to identify a progression threshold indicative of poor prognosis. MethodsCT scans and clinical history of 32 patients with idiopathic fibrotic ILDs were retrospectively reviewed. Successive CT scans were processed by the U-Net CNN and ILD quantification was obtained. Correlation between ILD and FVC changes was assessed. ROC curve was used to define a threshold of ILD progression rate (PR) to predict poor prognostic (mortality or lung transplantation). The PR threshold was used to compare the cohort survival with Kaplan Mayer curves and log-rank test. ResultsThe follow-up was 3.8 ± 1.5 years encompassing 105 CT scans, with 3.3 ± 1.1 CT scans per patient. A significant correlation between ILD and FVC changes was obtained (p = 0.004, ρ = -0.30 [95% CI: -0.16 to -0.45]). Sixteen patients (50%) experienced unfavorable outcome including 13 deaths and 3 lung transplantations. ROC curve analysis showed an aera under curve of 0.83 (p < 0.001), with an optimal cut-off PR value of 4%/year. Patients exhibiting a PR ≥ 4%/year during the first two years had a poorer prognosis (p = 0.001). ConclusionsApplying a U-Net CNN to routine CT scan allowed identifying patients with a rapid progression and unfavorable outcome.

U2-Net: A Very-Deep Convolutional Neural Network for Detecting Distracted Drivers

In recent years, the number of deaths and injuries resulting from traffic accidents has been increasing dramatically all over the world due to distracted drivers. Thus, a key element in developing intelligent vehicles and safe roads is monitoring driver behaviors. In this paper, we modify and extend the U-net convolutional neural network so that it provides deep layers to represent image features and yields more precise classification results. It is the basis of a very deep convolution neural network, called U2-net, to detect distracted drivers. The U2-net model has two paths (contracting and expanding) in addition to a fully-connected dense layer. The contracting path is used to extract the context around the objects to provide better object representation while the symmetric expanding path enables precise localization. The motivation behind this model is that it provides precise object features to provide a better object representation and classification. We used two public datasets: MI-AUC and State Farm, to evaluate the U2 model in detecting distracted driving. The accuracy of U2-net on MI-AUC and State Farm is 98.34 % and 99.64%, respectively. These evaluation results show higher accuracy than achieved by many other state-of-the-art methods.

Deep learning-based ultrasound auto-segmentation of the prostate with brachytherapy implanted needles.

Accurate segmentation of the clinical target volume (CTV) corresponding to the prostate with or without proximal seminal vesicles is required on transrectal ultrasound (TRUS) images during prostate brachytherapy procedures. Implanted needles cause artifacts that may make this task difficult and time-consuming. Thus, previous studies have focused on the simpler problem of segmentation in the absence of needles at the cost of reduced clinical utility. To use a convolutional neural network (CNN) algorithm for segmentation of the prostatic CTV in TRUS images post-needle insertion obtained from prostate brachytherapy procedures to better meet the demands of the clinical procedure. A dataset consisting of 144 3-dimensional (3D) TRUS images with implanted metal brachytherapy needles and associated manual CTV segmentations was used for training a 2-dimensional (2D) U-Net CNN using a Dice Similarity Coefficient (DSC) loss function. These were split by patient, with 119 used for training and 25 reserved for testing. The 3D TRUS training images were resliced at radial (around the axis normal to the coronal plane) and oblique angles through the center of the 3D image, as well as axial, coronal, and sagittal planes to obtain 3689 2D TRUS images and masks for training. The network generated boundary predictions on 300 2D TRUS images obtained from reslicing each of the 25 3D TRUS images used for testing into 12 radial slices (15° apart), which were then reconstructed into 3D surfaces. Performance metrics included DSC, recall, precision, unsigned and signed volume percentage differences (VPD/sVPD), mean surface distance (MSD), and Hausdorff distance (HD). In addition, we studied whether providing algorithm-predicted boundaries to the physicians and allowing modifications increased the agreement between physicians. This was performed by providing a subset of 3D TRUS images of five patients to five physicians who segmented the CTV using clinical software and repeated this at least 1week apart. The five physicians were given the algorithm boundary predictions and allowed to modify them, and the resulting inter- and intra-physician variability was evaluated. Median DSC, recall, precision, VPD, sVPD, MSD, and HD of the 3D-reconstructed algorithm segmentations were 87.2 [84.1, 88.8]%, 89.0 [86.3, 92.4]%, 86.6 [78.5, 90.8]%, 10.3 [4.5, 18.4]%, 2.0 [-4.5, 18.4]%, 1.6 [1.2, 2.0]mm, and 6.0 [5.3, 8.0]mm, respectively. Segmentation time for a set of 12 2D radial images was 2.46 [2.44, 2.48]s. With and without U-Net starting points, the intra-physician median DSCs were 97.0 [96.3, 97.8]%, and 94.4 [92.5, 95.4]% (p<0.0001), respectively, while the inter-physician median DSCs were 94.8 [93.3, 96.8]% and 90.2 [88.7, 92.1]%, respectively (p<0.0001). The median segmentation time for physicians, with and without U-Net-generated CTV boundaries, were 257.5 [211.8, 300.0]s and 288.0 [232.0, 333.5]s, respectively (p=0.1034). Our algorithm performed at a level similar to physicians in a fraction of the time. The use of algorithm-generated boundaries as a starting point and allowing modifications reduced physician variability, although it did not significantly reduce the time compared to manual segmentations.

Femurs segmentation by machine learning from CT scans combined with autonomous finite elements in orthopedic and endocrinology applications

Efficient, accurate and reliable segmentation of femurs from CT-scans is of major importance for patient-specific autonomous finite element analysis (AFE) to determine bone's stiffness and strength. We present a fully automated segmentation algorithm for whole and partial femurs with or without tumors, and clinical applications of the AFE [1] in clinical practice.The segmentation is based on an U-Net convolutional neural network, resulting a 3D mask representing the desired femur in a CT scan. It is robust, independent of the scanning parameters such as slice spacing, pixel size, scanner manufacturer or the femoral length available in the scan. The U-Net was trained on 178 manually segmented femurs (23,721 images) and tested on 43. The performance evaluation resulted in a Dice similarity score (DSC) of 0.9924, intersection over union (IoU) of 0.9849, Hausdorff distance of 4.3315 mm and symmetric average surface distance (ASD) of 0.1326 mm. The algorithm is competitive with the best state-of-the-art femoral segmentation methodologies available.Based on the segmentation an automatic p-FE mesh is generated and physiological boundary conditions representing sidewise falls or stance are being applied automatically to improve the performance of the AFE described in [1]. New examples of the usage of the AFE in endocrinology and orthopedic oncology demonstrate this disruptive technology in actual clinical practice. We present the use of AFE for predicting hip fracture risk in the elderly population due to a sidewise fall and the identification of patients who require a prophylactic surgery due to metastatic tumors in their femurs.