Training Images Research Articles

Data-Driven Machine Fault Diagnosis of Multisensor Vibration Data Using Synchrosqueezed Transform and Time-Frequency Image Recognition with Convolutional Neural Network

Accurate vibration classification using inertial measurement unit (IMU) data is critical for various applications such as condition monitoring and fault diagnosis. This study proposes a novel convolutional neural network (CNN) based approach, the IMU6DoF-SST-CNN in six variants, for robust vibration classification. The method utilizes Fourier synchrosqueezed transform (FSST) and wavelet synchrosqueezed transform (WSST) for time-frequency analysis, effectively capturing the temporal and spectral characteristics of the vibration data. Additionally, was used the IMU6DoF-SST-CNN to explore three different fusion strategies for sensor data to combine information from the IMU’s multiple axes, allowing the CNN to learn from complementary information across various axes. The efficacy of the proposed method was validated using three datasets. The first dataset consisted of constant fan velocity data (three classes: idle, normal operation, and fault) at 200 Hz. The second dataset contained variable fan velocity data (also with three classes: normal operation, fault 1, and fault 2) at 2000 Hz. Finally, a third dataset of Case Western Reserve University (CWRU) comprised bearing fault data with thirteen classes, sampled at 12 kHz. The proposed method achieved a perfect validation accuracy for the investigated vibration classification task. While all variants of the method achieved high accuracy, a trade-off between training speed and image generation efficiency was observed. Furthermore, FSST demonstrated superior localization capabilities compared to traditional methods like continuous wavelet transform (CWT) and short-time Fourier transform (STFT), as confirmed by image representations and interpretability analysis. This improved localization allows the CNN to effectively capture transient features associated with faults, leading to more accurate vibration classification. Overall, this study presents a promising and efficient approach for vibration classification using IMU data with the proposed IMU6DoF-SST-CNN method. The best result was obtained for IMU6DoF-SST-CNN with FSST and sensor-type fusion.

Uncertainty inspired domain adaptation network for rail surface defect segmentation

Rail surface defect detection is an essential part of railroad maintenance to prevent safety accidents. Recently, deep learning-based defect detection algorithms have shown impressive detection performance. However, the performance improvement from deep neural networks is based on a strong assumption that the training images have the same data distribution as the test images. This requirement is unrealistic in practical railroad surface defect detection. The appearance of rails varies significantly in different railroad sections due to changes in working conditions and maintenance status. In this paper, we propose a novel uncertainty-inspired unsupervised domain adaptation framework (UIUDA) to improve the generalization of deep learning models on data with distribution differences. Specifically, domain adversarial learning is used to align the feature distributions of training and test data. To mitigate the negative transfer caused by global alignment, we design a locally aligned domain adaptation strategy. In addition, self-training is introduced to refine the decision boundaries of the model in the test data. In this process, we propose an uncertainty-inspired evaluation method to remove noisy pseudo labels. We conducted domain adaptation experiments with RSDD and NEU-RSDD as the source domain and FRSD dataset as the target domain. The results show that UIUDA can effectively improve the generalization ability of the model, and the mIoU metric of defect segmentation is 80.17%.

Comparative analysis of convolutional neural network and DenseNet121 transfer learning in agriculture focusing on crop leaf disease identification

PurposePlant cultivation holds a pivotal role in agriculture, necessitating precise disease identification for the overall health of plants. This research conducts a comprehensive comparative analysis between two prominent deep learning algorithms, convolutional neural network (CNN) and DenseNet121, with the goal of enhancing disease identification in tomato plant leaves.Design/methodology/approachThe dataset employed in this investigation is a fusion of primary data and publicly available data, covering 13 distinct disease labels and a total of 18,815 images for model training. The data pre-processing workflow prioritized activities such as normalizing pixel dimensions, implementing data augmentation and achieving dataset balance, which were subsequently followed by the modeling and testing phases.FindingsExperimental findings elucidated the superior performance of the DenseNet121 model over the CNN model in disease classification on tomato leaves. The DenseNet121 model attained a training accuracy of 98.27%, a validation accuracy of 87.47% and average recall, precision and F1-score metrics of 87, 88 and 87%, respectively. The ultimate aim was to implement the optimal classifier for a mobile application, namely Tanamin.id, and, therefore, DenseNet121 was the preferred choice.Originality/valueThe integration of private and public data significantly contributes to determining the optimal method. The CNN method achieves a training accuracy of 90.41% and a validation accuracy of 83.33%, whereas the DenseNet121 method excels with a training accuracy of 98.27% and a validation accuracy of 87.47%. The DenseNet121 architecture, comprising 121 layers, a global average pooling (GAP) layer and a dropout layer, showcases its effectiveness. Leveraging categorical_crossentropy as the loss function and utilizing the stochastic gradien descent (SGD) Optimizer with a learning rate of 0.001 guides the course of the training process. The experimental results unequivocally demonstrate the superior performance of DenseNet121 over CNN.

An emerging network for COVID-19 CT-scan classification using an ensemble deep transfer learning model

Over the past few years, the widespread outbreak of COVID-19 has caused the death of millions of people worldwide. Early diagnosis of the virus is essential to control its spread and provide timely treatment. Artificial intelligence methods are often used as powerful tools to reach a COVID-19 diagnosis via computed tomography (CT) samples. In this paper, artificial intelligence-based methods are introduced to diagnose COVID-19. At first, a network called CT6-CNN is designed, and then two ensemble deep transfer learning models are developed based on Xception, ResNet-101, DenseNet-169, and CT6-CNN to reach a COVID-19 diagnosis by CT samples. The publicly available SARS-CoV-2 CT dataset is utilized for our implementation, including 2481 CT scans. The dataset is separated into 2108, 248, and 125 images for training, validation, and testing, respectively. Based on experimental results, the CT6-CNN model achieved 94.66% accuracy, 94.67% precision, 94.67% sensitivity, and 94.65% F1-score rate. Moreover, the ensemble learning models reached 99.2% accuracy. Experimental results affirm the effectiveness of designed models, especially the ensemble deep learning models, to reach a diagnosis of COVID-19.

Perlin noise generation of physiologically realistic cardiac fibrosis

Fibrosis, a pathological increase in extracellular matrix proteins, is a significant health issue that hinders the function of many organs in the body, in some cases fatally. In the heart, fibrosis impacts on electrical propagation in a complex and poorly predictable fashion, potentially serving as a substrate for dangerous arrhythmias. Individual risk depends on the spatial manifestation of fibrotic tissue, and learning the spatial arrangement on the fine scale in order to predict these impacts still relies upon invasive ex vivo procedures. As a result, the effects of spatial variability on the symptomatic impact of cardiac fibrosis remain poorly understood. In this work, we address the issue of availability of such imaging data via a computational methodology for generating new realisations of cardiac fibrosis microstructure. Using the Perlin noise technique from computer graphics, together with an automated calibration process that requires only a single training image, we demonstrate successful capture of collagen texturing in four types of fibrosis microstructure observed in histological sections. We then use this generator to quantitatively analyse the conductive properties of these different types of cardiac fibrosis, as well as produce three-dimensional realisations of histologically-observed patterning. Owing to the generator’s flexibility and automated calibration process, we also anticipate that it might be useful in producing additional realisations of other physiological structures.

Pixel-MPS: Stochastic Embedding and Density-Based Clustering of Image Patterns for Pixel-Based Multiple-Point Geostatistical Simulation

Multiple-point geostatistics (MPS) is an established tool for the uncertainty quantification of Earth systems modeling, particularly when dealing with the complexity and heterogeneity of geological data. This study presents a novel pixel-based MPS method for modeling spatial data using advanced machine-learning algorithms. Pixel-based multiple-point simulation implies the sequential modeling of individual points on the simulation grid, one at a time, by borrowing spatial information from the training image and honoring the conditioning data points. The developed methodology is based on the mapping of the training image patterns database using the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm for dimensionality reduction, and the clustering of patterns by applying the Density-based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, as an efficient unsupervised classification technique. For the automation, optimization, and input parameter tuning, multiple stages are implemented, including entropy-based determination of the template size and a k-nearest neighbors search for clustering parameter selection, to ensure the proposed method does not require the user’s interference. The proposed model is validated using synthetic two- and three-dimensional datasets, both for conditional and unconditional simulations, and runtime information is provided. Finally, the method is applied to a case study gold mine for stochastic orebody modeling. To demonstrate the computational efficiency and accuracy of the proposed method, a two-dimensional training image with 101 by 101 pixels is simulated for 100 conditional realizations in 453 s (~4.5 s per realization) using only 361 hard data points (~3.5% of the simulation grid), and the resulting average simulation has a good visual match and only an 11.8% pixel-wise mismatch with the training image.

Synthetic Image Generation Using Conditional GAN-Provided Single-Sample Face Image

The performance of facial recognition systems significantly decreases when faced with a lack of training images. This issue is exacerbated when there is only one image per subject available. Probe images may contain variations such as illumination, expression, and disguise, which are difficult to recognize accurately. In this work, we present a model that generates six facial variations from a single neutral face image. Our model is based on a CGAN, designed to produce six highly realistic facial expressions from one neutral face image. To evaluate the accuracy of our approach comprehensively, we employed several pre-trained models (VGG-Face, ResNet-50, FaceNet, and DeepFace) along with a custom CNN model. Initially, these models achieved only about 76% accuracy on single-sample neutral images, highlighting the SSPP challenge. However, after fine-tuning on the synthetic expressions generated by our CGAN from these single images, their accuracy increased significantly to around 99%. Our method has proven highly effective in addressing SSPP issues, as evidenced by the significant improvement achieved.

Reconstruction of High-Resolution 3D GPR Data from 2D Profiles: A Multiple-Point Statistical Approach

Ground-penetrating radar (GPR) is a popular geophysical tool for mapping the underground. High-resolution 3D GPR data carry a large amount of information and can greatly help to interpret complex subsurface geometries. However, such data require a dense collection along closely spaced parallel survey lines, which is time consuming and costly. In many cases, for the sake of efficiency, a choice is made during 3D acquisitions to use a larger spacing between the profile lines, resulting in a dense measurement spacing along the lines but a much coarser one in the across-line direction. Simple interpolation methods are then commonly used to increase the sampling before interpretation, which can work well when the subsurface structures are already well sampled in the across-line direction but can distort such structures when this is not the case. In this work, we address the latter problem using a novel multiple-point geostatistical (MPS) simulation methodology. For a considered 3D GPR dataset with reduced sampling in the across-line direction, we attempt to reconstruct a more densely spaced, high-resolution dataset using a series of 2D conditional stochastic simulations in both the along-line and across-line directions. For these simulations, the existing profile data serve as training images from which complex spatial patterns are quantified and reproduced. To reduce discontinuities in the generated 3D spatial structures caused by independent 2D simulations, the target profile being simulated is chosen randomly, and simulations in the along-line and across-line directions are performed alternately. We show the successful application of our approach to 100 MHz synthetic and 200 MHz field GPR data under multiple decimation scenarios where survey lines are regularly deleted from a dense 3D reference dataset, and the corresponding reconstructions are compared with the original data.

Advancements in diagnosing oral potentially malignant disorders: leveraging Vision transformers for multi-class detection

ObjectivesDiagnosing oral potentially malignant disorders (OPMD) is critical to prevent oral cancer. This study aims to automatically detect and classify the most common pre-malignant oral lesions, such as leukoplakia and oral lichen planus (OLP), and distinguish them from oral squamous cell carcinomas (OSCC) and healthy oral mucosa on clinical photographs using vision transformers.Methods4,161 photographs of healthy mucosa, leukoplakia, OLP, and OSCC were included. Findings were annotated pixel-wise and reviewed by three clinicians. The photographs were divided into 3,337 for training and validation and 824 for testing. The training and validation images were further divided into five folds with stratification. A Mask R-CNN with a Swin Transformer was trained five times with cross-validation, and the held-out test split was used to evaluate the model performance. The precision, F1-score, sensitivity, specificity, and accuracy were calculated. The area under the receiver operating characteristics curve (AUC) and the confusion matrix of the most effective model were presented.ResultsThe detection of OSCC with the employed model yielded an F1 of 0.852 and AUC of 0.974. The detection of OLP had an F1 of 0.825 and AUC of 0.948. For leukoplakia the F1 was 0.796 and the AUC was 0.938.ConclusionsOSCC were effectively detected with the employed model, whereas the detection of OLP and leukoplakia was moderately effective.Clinical relevanceOral cancer is often detected in advanced stages. The demonstrated technology may support the detection and observation of OPMD to lower the disease burden and identify malignant oral cavity lesions earlier.

Development of methods for generating examples that train a neural network to detect and classify interference in the structure of a useful signal

The main purpose of the information transmission system is to transfer useful information between its users reliably and accurately. At the same time, it would be advisable to supplement the functionality of the information transmission system with the ability to analyze interference in the communication channel during the transmission of a message. In the case of analysis of damage to the useful signal, which may occur due to interference of various types in the propagation medium, the information transmission system can use the results of such analysis to improve noise immunity, to collect statistics, to control electromagnetic compatibility. It is known that optimal processing of noise-like signals is carried out using a correlator or a matched filter, but these classical devices might be complemented with a neural network, and then in addition to determining the correlation function, it is the neural network that is able to analyze a signal damaged by interference, as shown by the authors in their previous works. In this case, for the correct training of a neural network, samples of a useful signal damaged by various types of jamming are needed. This article discusses methods for the formation of training images for a neural network designed to determine the period of pulse interference acting on an M-sequence as a known signal. Two techniques are suggested. They are digital modelling and experimental (laboratory) method. The first one is based on the inverting some elements (chips) of the original M-sequence located in certain segments whose lengths are random. The inversion is performed by the program written in C++ language. The second approach involves a special installation that consists of a personal computer, an Arduino microcontroller, a useful signal transmitter, a receiver and a periodic interference generator. The objective of each method is to obtain a matrix of damaged chips in a chain of M-sequences. Such matrix was called an “error profile” by the authors. The characteristics of the error profile obtained for certain interference parameters (jamming pulse period and duration) according to suggested methods can be used as training data for the neural network designed for jamming classification.

Patient-specific placental vessel segmentation with limited data

A major obstacle in applying machine learning for medical fields is the disparity between the data distribution of the training images and the data encountered in clinics. This phenomenon can be explained by inconsistent acquisition techniques and large variations across the patient spectrum. The result is poor translation of the trained models to the clinic, which limits their implementation in medical practice. Patient-specific trained networks could provide a potential solution. Although patient-specific approaches are usually infeasible because of the expenses associated with on-the-fly labeling, the use of generative adversarial networks enables this approach. This study proposes a patient-specific approach based on generative adversarial networks. In the presented training pipeline, the user trains a patient-specific segmentation network with extremely limited data which is supplemented with artificial samples generated by generative adversarial models. This approach is demonstrated in endoscopic video data captured during fetoscopic laser coagulation, a procedure used for treating twin-to-twin transfusion syndrome by ablating the placental blood vessels. Compared to a standard deep learning segmentation approach, the pipeline was able to achieve an intersection over union score of 0.60 using only 20 annotated images compared to 100 images using a standard approach. Furthermore, training with 20 annotated images without the use of the pipeline achieves an intersection over union score of 0.30, which, therefore, corresponds to a 100% increase in performance when incorporating the pipeline. A pipeline using GANs was used to generate artificial data which supplements the real data, this allows patient-specific training of a segmentation network. We show that artificial images generated using GANs significantly improve performance in vessel segmentation and that training patient-specific models can be a viable solution to bring automated vessel segmentation to the clinic.

Weed Detection and Classification with Computer Vision Using a Limited Image Dataset

In agriculture, as precision farming increasingly employs robots to monitor crops, the use of weeding and harvesting robots is expanding the need for computer vision. Currently, most researchers and companies address these computer vision tasks with CNN-based deep learning. This technology requires large datasets of plant and weed images labeled by experts, as well as substantial computational resources. However, traditional feature-based approaches to computer vision can extract meaningful parameters and achieve comparably good classification results with only a tenth of the dataset size. This study presents these methods and seeks to determine the minimum number of training images required to achieve reliable classification. We tested the classification results with 5, 10, 20, 40, 80, and 160 images per weed type in a four-class classification system. We extracted shape features, distance transformation features, color histograms, and texture features. Each type of feature was tested individually and in various combinations to determine the best results. Using six types of classifiers, we achieved a 94.56% recall rate with 160 images per weed. Better results were obtained with more training images and a greater variety of features.

Evaluation method for the intactness of tunnel face surrounding rock based on tunnel face images

The intactness of rock masses is a fundamental parameter in classifying surrounding rocks. Due to limitations imposed by the extent of tunnel face outcrops, the assessment of rock mass intactness necessitates the manual extraction of the positions, orientations and spacing of joints/fissures. To mitigate the labour-intensive nature of this process, in this paper deep learning is employed to develop an integrated method for the automated extraction of joints/fissures and the quantitative analysis of rock mass intactness. We introduce an image preprocessing method based on multiscale histogram equalization to obtain high-contrast, low-noise images. The DeepIntactness model, which incorporates the strategy of curriculum learning to utilize a large number of unlabelled tunnel rock images for model training, is introduced for the extraction of joints/fissures. Following the extraction of joints/fissures, a multiline centre statistic method based on the rock mass block index method is employed to evaluate the intactness of the most vulnerable part of the tunnel face. By applying this approach to an engineering structure, its capacity to automatically extract and quantitatively evaluate the engineering properties of the surrounding rock mass intactness is demonstrated. Hence, this method provides a novel approach to evaluating the tunnel surrounding rock intactness using two-dimensional images. Supplementary material: The I-RBI calculation process of the cases in this paper is available at https://doi.org/10.6084/m9.figshare.c.7154677

Object detection of the bornean orangutan nests using drone and YOLOv5

<p class="tm6">Object detection methods when applied to ecology and conservation can help to identify and monitor endangered species and their habitats. Using drones for this purpose has become increasingly popular due to their ability to cover large areas quickly and efficiently. In this study, we aim to implement object detection using YOLOv5 to detect orangutan nests in forests. To conduct our experiment, we collect drone imagery under different conditions. We propose to use the original YOLOv5 to implement our model. The detection and monitoring of orangutan nests can help conservationists to identify critical habitats, monitor population, and design effective conservation strategies. Additionally, the use of drones can reduce the need for on-the-ground surveys, which can be time-consuming, expensive, and logistically challenging. In our study proposes a model for detecting orangutan nests in forests using a drone and the YOLOv5. Our model predicted 1,970 training images and 414 labeled orangutan nests, with a precision of 0.973, recall 0.949, accuracy mean average precision (mAP)_0.5 is 0.969, and mAP_0.5:0.95 is 0.630. The model finished 217 epochs in 58 hours and had a high object detection accuracy. The model has a 99.9% accuracy in detecting the number of orangutan nests.</p>

Correlation of Noninvasive Cardiac MRI Measures of Left Ventricular Myocardial Function and Invasive Pressure-Volume Parameters in a Porcine Ischemia-Reperfusion Model.

Purpose To assess the correlation between noninvasive cardiac MRI-derived parameters with pressure-volume (PV) loop data and evaluate changes in left ventricular function after myocardial infarction (MI). Materials and Methods Sixteen adult female swine were induced with MI, with six swine used as controls and 10 receiving platelet-derived growth factor-AB (PDGF-AB). Load-independent measures of cardiac function, including slopes of end-systolic pressure-volume relationship (ESPVR) and preload recruitable stroke work (PRSW), were obtained on day 28 after MI. Cardiac MRI was performed on day 2 and day 28 after infarct. Global longitudinal strain (GLS) and global circumferential strain (GCS) were measured. Ventriculo-arterial coupling (VAC) was derived from PV loop and cardiac MRI data. Pearson correlation analysis was performed. Results GCS (r = 0.60, P = .01), left ventricular ejection fraction (LVEF) (r = 0.60, P = .01), and cardiac MRI-derived VAC (r = 0.61, P = .01) had a significant linear relationship with ESPVR. GCS (r = 0.75, P < .001) had the strongest significant linear relationship with PRSW, followed by LVEF (r = 0.67, P = .005) and cardiac MRI-derived VAC (r = 0.60, P = .01). GLS was not significantly correlated with ESPVR or PRSW. There was a linear correlation (r = 0.82, P < .001) between VAC derived from cardiac MRI and from PV loop data. GCS (-3.5% ± 2.3 vs 0.5% ± 1.4, P = .007) and cardiac MRI-derived VAC (-0.6 ± 0.6 vs 0.3 ± 0.3, P = .001) significantly improved in the animals treated with PDGF-AB 28 days after MI compared with controls. Conclusion Cardiac MRI-derived parameters of MI correlated with invasive PV measures, with GCS showing the strongest correlation. Cardiac MRI-derived measures also demonstrated utility in assessing therapeutic benefit using PDGF-AB. Keywords: Cardiac MRI, Myocardial Infarction, Pressure Volume Loop, Strain Imaging, Ventriculo-arterial Coupling Supplemental material is available for this article. © RSNA, 2024.

Artificial intelligence to automatically measure glenoid inclination, humeral alignment, and the lateralization and distalization shoulder angles on postoperative radiographs after reverse shoulder arthroplasty

BackgroundRadiographic evaluation of the implant configuration after reverse shoulder arthroplasty (RSA) is time-consuming and subject to interobserver disagreement. The final configuration is a combination of implant features and surgical execution. Artificial intelligence (AI) algorithms have been shown to perform accurate and efficient analysis of images. The purpose of this study was to develop an AI algorithm to automatically measure glenosphere inclination, humeral component inclination, and the lateralization and distalization shoulder angles (DSAs) on postoperative anteroposterior radiographs after RSA. MethodsThe Digital Imaging and Communications in Medicine files corresponding to postoperative anteroposterior radiographs obtained after implantation of 143 RSAs were retrieved and used in this study. Four angles were analyzed: (1) glenoid inclination angle (GIA, between the central fixation feature of the glenoid and the floor of the supraspinatus fossa), (2) humeral alignment angle (HAA, between the long axis of the humeral shaft and a perpendicular to the metallic bearing of the prosthesis), (3) DSA, and (4) lateralization shoulder angle (LSA). A UNet segmentation model was trained to segment bony and implant elements using manually segmented training (n = 89) and validation (n = 22) images. Then, an image-processing–based pipeline was developed to measure all 4 angles using AI-segmented images. Measures performed by 3 physician observers and the AI algorithm were then completed in 32 additional images. The agreements among human observers and between observers and the AI algorithm were evaluated using intraclass correlation coefficients (ICCs) and absolute differences in degree. ResultsThe ICCs (95% confidence interval) for manual measurements of LSA, DSA, GIA, and HAA were 0.79 (0.55, 0.90), 0.90 (0.80, 0.95), 0.96 (0.93, 0.98), and 0.99 (0.97, 0.99), respectively. The AI algorithm measured the 32 images in the test set in less than 2 minutes. The agreement between observers and the AI algorithm was lowest when measuring the LSA for observer 2, with an ICC of 0.77 (0.52, 0.89), and an absolute difference in degrees (median [interquartile range]) of 5 (4). Better agreements were found between the AI measurements and the average manual measurements: absolute differences in degree for LSA, DSA, GIA, and HAA were 3 (5), 2 (3), 2 (2), and 2 (1), respectively; ICCs for LSA, DSA, GIA, and HAA were 0.89 (0.79, 0.95), 0.96 (0.93, 0.98), 0.85 (0.68, 0.93), and 0.98 (0.95, 0.99), respectively. ConclusionThe AI algorithm developed in this study can automatically measure the GIA, HAA, LSA, and DSA on postoperative anteroposterior radiographs obtained after implantation on RSA.

Left Atrial Peak Systolic Strain as an Indicator Pathway of Diastolic Dysfunction of the Left Ventricle.

Left atrial peak systolic strain (LA-PSS) imaging is an emerging index of left atrial function, and it was shown to be decreased in heart failure with preserved ejection fraction (EF). We aimed to determine whether LA-PSS could be used as an additional diagnostic parameter to current existing guidelines for the presence of left ventricle diastolic dysfunction (LVDD). A total of 190 consecutive adult patients with cardiovascular risk factors and normal left ventricle EF with no prior history of heart failure were included in the study. Speckle tracking software was used to study ventricular parietal deformity, left ventricle global longitudinal systolic strain, and LA-PSS. The median left ventricle global longitudinal systolic strain was -19%, with a significant difference ( P < 0.001) between patients with normal diastolic function versus those with LVDD. The median LA-PSS was 33% (30% to 38%) ( P < 0.001). Most patients (61%) had grade 1 atrial dysfunction based on PSS (range 24%-35%). The analysis of the area under the receiver operating characteristic curve of the LA-PSS as a potential indicator pathway of LVDD was 67% [95% confidence interval (CI), 62-72], and 75% (95% CI, 70-80), when the indeterminate pattern was included. The decreased LA-PSS made it possible to reclassify patients with an indeterminate pattern of diastolic function in 96% of cases. These results support the potential role of LA-PSS as an additional parameter for the diagnosis of LVDD in patients with normal EF, and may be integrated into the guidelines for routine evaluation of patients.

Crack segmentation for high-speed imaging: detection of fractures in thermally toughened glass

Fracture morphology characterization in broken glass panes is crucial for designing laminated safety glass (LSG) in civil engineering. Verifying completely broken LSG systems requires destructive sampling, increasing costs and hindering development. Therefore, to determine the residual load-bearing capacity, the link between the pre-fracture characteristics and the fracture morphology must be known. However, when the crack propagation needs to be directly captured with high-speed imaging, conventional methods are no longer sufficient for detecting cracks. To enable such investigations, we propose a novel machine learning framework for crack segmentation in high-speed imaging that addresses the complexity of glass fracture and minimises the required labour costs. In this study, the crack propagation of a sample was recorded and analysed at 2,000,000 images per second. The results showcase accuracies surpassing 97% while requiring only two labeled images for training, thus streamlining practical implementation. Furthermore, we show the method's robustness to the extent that hyperparameter tuning becomes unnecessary. Instead, we offer guidelines for selecting the most crucial hyperparameters depending on the problem. Our method offers a promising approach for non-linear temporal interpolation of noisy images, with implications for various applications extending beyond glass fracture analysis.

Incorporating Season and Solar Specificity Into Renderings Made by a NeRF Architecture Using Satellite Images.

As a result of Shadow NeRF and Sat-NeRF, it is possible to take the solar angle into account in a NeRF-based framework for rendering a scene from a novel viewpoint using satellite images for training. Our work extends those contributions and shows how one can make the renderings season-specific. Our main challenge was creating a Neural Radiance Field (NeRF) that could render seasonal features independently of viewing angle and solar angle while still being able to render shadows. We teach our network to render seasonal features by introducing one more input variable - time of the year. However, the small training datasets typical of satellite imagery can introduce ambiguities in cases where shadows are present in the same location for every image of a particular season. We add additional terms to the loss function to discourage the network from using seasonal features for accounting for shadows. We show the performance of our network on eight Areas of Interest containing images captured by the Maxar WorldView-3 satellite. This evaluation includes tests measuring the ability of our framework to accurately render novel views, generate height maps, predict shadows, and specify seasonal features independently from shadows. Our ablation studies justify the choices made for network design parameters.

How many specimens make a sufficient training set for automated three-dimensional feature extraction?

Deep learning has emerged as a robust tool for automating feature extraction from three-dimensional images, offering an efficient alternative to labour-intensive and potentially biased manual image segmentation methods. However, there has been limited exploration into the optimal training set sizes, including assessing whether artficial expansion by data augmentation can achieve consistent results in less time and how consistent these benefits are across different types of traits. In this study, we manually segmented 50 planktonic foraminifera specimens from the genus Menardella to determine the minimum number of training images required to produce accurate volumetric and shape data from internal and external structures. The results reveal unsurprisingly that deep learning models improve with a larger number of training images with eight specimens being required to achieve 95% accuracy. Furthermore, data augmentation can enhance network accuracy by up to 8.0%. Notably, predicting both volumetric and shape measurements for the internal structure poses a greater challenge compared with the external structure, owing to low contrast differences between different materials and increased geometric complexity. These results provide novel insight into optimal training set sizes for precise image segmentation of diverse traits and highlight the potential of data augmentation for enhancing multivariate feature extraction from three-dimensional images.