Segmentation Of Data Research Articles

Vanderbilt Electronic Health Record Voice Assistant Supports Clinicians.

Electronic health records (EHRs) present navigation challenges due to time-consuming searches across segmented data. Voice assistants can improve clinical workflows by allowing natural language queries and contextually aware navigation of the EHR. To develop a voice-mediated EHR assistant and interview providers to inform its future refinement. The Vanderbilt EHR Voice Assistant (VEVA) was developed as a responsive web application and designed to accept voice inputs and execute the appropriate EHR commands. Fourteen providers from Vanderbilt Medical Center were recruited to participate in interactions with VEVA and to share their experience with the technology. The purpose was to evaluate VEVA's overall usability, gather qualitative feedback, and detail suggestions for enhancing its performance. VEVA's mean system usability scale score was 81 based on the 14 providers' evaluations, which was above the standard 50th percentile score of 68. For all five summaries evaluated (overview summary, A1C results, blood pressure, weight, and health maintenance), most providers offered a positive review of VEVA. Several providers suggested modifications to make the technology more useful in their practice, ranging from summarizing current medications to changing VEVA's speech rate. Eight of the providers (64%) reported they would be willing to use VEVA in its current form. Our EHR voice assistant technology was deemed usable by most providers. With further improvements, voice assistant tools such as VEVA have the potential to improve workflows and serve as a useful adjunct tool in health care.

A Deep Learning-Based Integrated Framework for Quality-Aware Undersampled Cine Cardiac MRI Reconstruction and Analysis.

Cine cardiac magnetic resonance (CMR) imaging is considered the gold standard for cardiac function evaluation. However, cine CMR acquisition is inherently slow and in recent decades considerable effort has been put into accelerating scan times without compromising image quality or the accuracy of derived results. In this article, we present a fully-automated, quality-controlled integrated framework for reconstruction, segmentation and downstream analysis of undersampled cine CMR data. The framework produces high quality reconstructions and segmentations, leading to undersampling factors that are optimised on a scan-by-scan basis. This results in reduced scan times and automated analysis, enabling robust and accurate estimation of functional biomarkers. To demonstrate the feasibility of the proposed approach, we perform simulations of radial k-space acquisitions using in-vivo cine CMR data from 270 subjects from the UK Biobank (with synthetic phase) and in-vivo cine CMR data from 16 healthy subjects (with real phase). The results demonstrate that the optimal undersampling factor varies for different subjects by approximately 1 to 2 seconds per slice. We show that our method can produce quality-controlled images in a mean scan time reduced from 12 to 4 seconds per slice, and that image quality is sufficient to allow clinically relevant parameters to be automatically estimated to lie within 5% mean absolute difference.

Semantic segmentation of raw multispectral laser scanning data from urban environments with deep neural networks

Real-time semantic segmentation of point clouds has increasing importance in applications related to 3D city modelling and mapping, automated inventory of forests, autonomous driving and mobile robotics. Current state-of-the-art point cloud semantic segmentation methods rely heavily on the availability of 3D laser scanning data. This is problematic in regards of low-latency, real-time applications that use data from high-precision mobile laser scanners, as those are typically 2D line scanning devices. In this study, we experiment with real-time semantic segmentation of high-density multispectral point clouds collected from 2D line scanners in urban environments using encoder - decoder convolutional neural network architectures. We introduce a rasterized multi-scan input format that can be constructed exclusively from the raw (non-georeferenced profiles) 2D laser scanner measurement stream without odometry information. In addition, we investigate the impact of multispectral data on the segmentation accuracy. The dataset used for training, validation and testing was collected with multispectral FGI AkhkaR4-DW backpack laser scanning system operating at the wavelengths of 905 nm and 1550 nm, and consists in total of 228 million points (39 583 scans). The data was divided into 13 classes that represent various targets in urban environments. The results show that the increased spatial context of the multi-scan format improves the segmentation performance on the single-wavelength lidar dataset from 45.4 mIoU (a single scan) to 62.1 mIoU (24 consecutive scans). In the multispectral point cloud experiments we achieved a 71 % and 28 % relative increase in the segmentation mIoU (43.5 mIoU) as compared to the purely single-wavelength reference experiments, in which we achieved 25.4 mIoU (905 nm) and 34.1 mIoU (1550 nm). Our findings show that it is possible to semantically segment 2D line scanner data with good results by combining consecutive scans without the need for odometry information. The results also serve as motivation for developing multispectral mobile laser scanning systems that can be used in challenging urban surveys.

CVT error prediction method based on improved attention mechanism transformer model

In order to enhance the performance of capacitive voltage transformer (CVT) error prediction, we propose a transformer model (AM-transformer) error prediction method based on an improved attention mechanism. The method starts from both input data and model improvement, adopting Pearson’s correlation coefficient to calculate the environmental factors that have a greater impact on the CVT measurement error, and enhancing the ability of local learning of the attention mechanism by adding causal convolution and data segmentation operations. The experimental results show that the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) of the method reach 0.016, 4.21% and 0.021, respectively, realizing a high prediction accuracy.

Application of distributed database in management information system

This article discusses the application methods of distributed databases in management information systems. First, the two main types of structured and isomorphic databases are analyzed. Then, it elaborates on the application technology of distributed databases from five aspects: structural design, data transmission, security strategy, function implementation, and data synchronization maintenance, and proposes the selection of network topology mode, data segmentation strategy, and optimized transmission algorithm according to business needs. , establish security protection mechanisms, and support OLTP and OLAP processing methods. Finally, distributed databases provide essential support for management information systems by horizontally expanding storage and computing capabilities but also bring technical challenges.

IMPLEMENTATION OF THE BIDIRECTIONAL GATED RECURRENT UNIT ALGORITHM ON CONSUMER PRICE INDEX DATA IN INDONESIA

The Consumer Price Index (CPI) is the main index in measuring the inflation rate. Changes in the CPI from time to time reflect inflation and deflation, namely the higher the CPI value, the higher the inflation rate. This study aims to apply Birectional Gated Recurrent Unit (BiGRU) model to the CPI data in Indonesia. BiGRU comprises two GRU layers so it captures sequences that are ignored by the GRU. The research data is in the form of CPI data in Indonesia from January 2006 to December 2022 sourced from the website of the Central Bureau of Statistics totaling 204 data. The data is divided into training data and testing data. Training data was taken from January 2006 to July 2019 as many as 163 data. Data testing was taken from August 2019 to December 2022 as many as 41 data. Before the data is processed, a sliding window process is carried out by dividing the data into segments to reduce the error value. The window size value used is 10. In the sliding window process, the number of segments is 194 data segments. Based on the experiment results, it was concluded that the application of BiGRU to the CPI data was carried out in an experiment with 20 BiGRU architectures. BiGRU architecture was obtained which produced the lowest MAPE value, namely an architecture with two BiGRU layers having 256 neurons and 400 units, and one dense layer. In addition, the epochs used are 200 epochs, the ReLU activation function, and Adam optimization. The experimental results of the BiGRU architecture obtained a MAPE value of 0.24% which indicates that the architectural performance is very good.

Quantifying the accuracy of inter-beat intervals acquired from consumer-grade photoplethysmography wristbands using an electrocardiogram-aided information-based similarity approach

Objective. Although inter-beat intervals (IBI) and the derived heart rate variability (HRV) can be acquired through consumer-grade photoplethysmography (PPG) wristbands and have been applied in a variety of physiological and psychophysiological conditions, their accuracy is still unsatisfactory. Approach. In this study, 30 healthy participants concurrently wore two wristbands (E4 and Honor 5) and a gold-standard electrocardiogram (ECG) device under four conditions: resting, deep breathing with a frequency of 0.17 Hz and 0.1 Hz, and mental stress tasks. To quantitatively validate the accuracy of IBI acquired from PPG wristbands, this study proposed to apply an information-based similarity (IBS) approach to quantify the pattern similarity of the underlying dynamical temporal structures embedded in IBI time series simultaneously recorded using PPG wristbands and the ECG system. The occurrence frequency of basic patterns and their rankings were analyzed to calculate the IBS distance from gold-standard IBI, and to further calculate the signal-to-noise ratio (SNR) of the wristband IBI time series. Main results. The accuracies of both HRV and mental state classification were not satisfactory due to the low SNR in the wristband IBI. However, by rejecting data segments of SNR < 25, the Pearson correlation coefficients between the wristbands' HRV and the gold-standard HRV were increased from 0.542 ± 0.235 to 0.922 ± 0.120 for E4 and from 0.596 ± 0.227 to 0.859 ± 0.145 for Honor 5. The average accuracy of four-class mental state classification increased from 77.3% to 81.9% for E4 and from 79.3% to 83.3% for Honor 5. Significance. Consumer-grade PPG wristbands are acceptable for HR and HRV monitoring when removing low SNR segments. The proposed method can be applied for quantifying the accuracies of IBI and HRV indices acquired via any non-ECG system.

Integrating Navigation-Assisted Ablation in the Locoregional Treatment of Hepatocellular Carcinoma

With the increasing prevalence of hepatocellular carcinoma (HCC), ablative therapy is a critical treatment option to achieve a cancer-free state. The anatomic location of the tumor can be a challenge, and select hepatic locations of a tumor require laparoscopic-assisted strategies to safely reach and treat the lesion. To assess the association of real-time ultrasonography-augmented navigation for HCC ablation with patient survival, operative time, and rate of incomplete ablations. This retrospective case-control study was conducted among a prospectively collected database of more than 750 patients with HCC who were treated with ablation therapy with and without the use of navigation at a single quaternary medical center from June 2011 to January 2021. Data were analyzed from October 2022 through June 2023. Real-time ultrasonography-augmented navigation. The primary outcome was rate of incomplete ablations in patients undergoing HCC ablation with vs without navigation. Secondary outcomes included overall survival (OS), progression-free survival (PFS), and operative time. The analytic cohort included 467 patients (mean [SD] age, 62.4 [7.8] years; 355 male [76.0%]; 21 Hispanic [4.5%], 67 non-Hispanic Black [14.5%], and 347 Non-Hispanic White [75.0%] among 463 patients with race and ethnicity data). The most common etiology of liver disease was hepatitis C infection (187 patients with etiology data [40.0%]), and 348 of 458 patients with TMN staging data (76.0%) had TNM stage 1 disease. There were 187 individuals treated with navigation and 280 individuals treated without navigation. Patients who underwent navigation-assisted ablation were more likely to have stage 2 disease based on TNM staging (62 of 183 patients [33.9%] vs 47 of 275 patients [17.1%] with TMN data; P < .002) and had a higher mean (SD) number of lesions (1.3 [0.5] vs 1.2 [0.5] lesions; P = .002) and a longer mean (SD) operation time (113.2 [29.4] vs 109.6 [32.3] minutes; P = .04). Patients who underwent navigation were also more likely to have tumors in segment 8 (59 patients [32.1%] vs 53 of 275 patients with segment data [19.3%] with segment data; P = .005) and less likely to have tumors in segment 4 (20 patients [10.9%] vs 54 patients with segment data [19.6%]; P = .005). Overall mean (SD) time to recurrence after treatment was 10.0 (12.5) months, with similar rates for patients with navigation vs no navigation. There were no differences in incomplete ablation rate (10 patients [9.2%] vs 10 patients [10.5%]; P = .32), OS, or PFS between patients undergoing ablation with and without navigation. In this study, use of navigation was associated with comparable outcomes to undergoing ablation without navigation, although patients with navigation had more locally advanced disease. These findings suggest that use of real-time navigation in laparoscopic-assisted ablation of liver cancer should be considered as a useful tool for treating challenging tumors.

Price forecasting through neural networks for crude oil, heating oil, and natural gas

Building price projections of various energy commodities has long been an important endeavor for a wide range of participants in the energy market. We study the forecast problem in this paper by concentrating on four significant energy commodities. Using nonlinear autoregressive neural network models, we investigate the daily prices of WTI and Brent crude oil as well as the monthly prices of Henry Hub natural gas and New York Harbor No. 2 heating oil. We investigate prediction performance resulting from various model configurations, including training techniques, hidden neurons, delays, and data segmentation. Based on the investigation, relatively straightforward models are built that yield quite accurate and reliable performance. Specifically, performance in terms of relative root mean square errors is 1.96%/1.81%/9.75%/21.76%, 1.96%/1.80%/8.76%/14.41%, and 1.87%/1.78%/9.10%/16.97% for model training, validation, and testing, respectively, and the overall relative root mean square error is 1.95%/1.80%/9.51%/20.35% for the whole sample for WTI crude oil/Brent crude oil/New York Harbor No. 2 heating oil/Henry Hub natural gas. The outcomes of this projection might be used in technical analysis or integrated with other fundamental forecasts for policy analysis.

DeepD3, an open framework for automated quantification of dendritic spines.

Dendritic spines are the seat of most excitatory synapses in the brain, and a cellular structure considered central to learning, memory, and activity-dependent plasticity. The quantification of dendritic spines from light microscopy data is usually performed by humans in a painstaking and error-prone process. We found that human-to-human variability is substantial (inter-rater reliability 82.2±6.4%), raising concerns about the reproducibility of experiments and the validity of using human-annotated 'ground truth' as an evaluation method for computational approaches of spine identification. To address this, we present DeepD3, an open deep learning-based framework to robustly quantify dendritic spines in microscopy data in a fully automated fashion. DeepD3's neural networks have been trained on data from different sources and experimental conditions, annotated and segmented by multiple experts and they offer precise quantification of dendrites and dendritic spines. Importantly, these networks were validated in a number of datasets on varying acquisition modalities, species, anatomical locations and fluorescent indicators. The entire DeepD3 open framework, including the fully segmented training data, a benchmark that multiple experts have annotated, and the DeepD3 model zoo is fully available, addressing the lack of openly available datasets of dendritic spines while offering a ready-to-use, flexible, transparent, and reproducible spine quantification method.

Automated forest inventory: Analysis of high-density airborne LiDAR point clouds with 3D deep learning

Detailed forest inventories are critical for sustainable and flexible management of forest resources, to conserve various ecosystem services. Modern airborne laser scanners deliver high-density point clouds with great potential for fine-scale forest inventory and analysis, but automatically partitioning those point clouds into meaningful entities like individual trees or tree components remains a challenge. The present study aims to fill this gap and introduces a deep learning framework, termed ForAINet, that is able to perform such a segmentation across diverse forest types and geographic regions. From the segmented data, we then derive relevant biophysical parameters of individual trees as well as stands. The system has been tested on FOR-Instance, a dataset of point clouds that have been acquired in five different countries using surveying drones. The segmentation back-end achieves over 85% F-score for individual trees, respectively over 73% mean IoU across five semantic categories: ground, low vegetation, stems, live branches and dead branches. Building on the segmentation results our pipeline then densely calculates biophysical features of each individual tree (height, crown diameter, crown volume, DBH, and location) and properties per stand (digital terrain model and stand density). Especially crown-related features are in most cases retrieved with high accuracy, whereas the estimates for DBH and location are less reliable, due to the airborne scanning setup.

Reconstructing microvascular network skeletons from 3D images: What is the ground truth?

Structural changes to microvascular networks are increasingly highlighted as markers of pathogenesis in a wide range of disease, e.g. Alzheimer’s disease, vascular dementia and tumour growth. This has motivated the development of dedicated 3D imaging techniques, alongside the creation of computational modelling frameworks capable of using 3D reconstructed networks to simulate functional behaviours such as blood flow or transport processes. Extraction of 3D networks from imaging data broadly consists of two image processing steps: segmentation followed by skeletonisation. Much research effort has been devoted to segmentation field, and there are standard and widely-applied methodologies for creating and assessing gold standards or ground truths produced by manual annotation or automated algorithms.The Skeletonisation field, however, lacks widely applied, simple to compute metrics for the validation or optimisation of the numerous algorithms that exist to extract skeletons from binary images. This is particularly problematic as 3D imaging datasets increase in size and visual inspection becomes an insufficient validation approach. In this work, we first demonstrate the extent of the problem by applying 4 widely-used skeletonisation algorithms to 3 different imaging datasets. In doing so we show significant variability between reconstructed skeletons of the same segmented imaging dataset. Moreover, we show that such a structural variability propagates to simulated metrics such as blood flow. To mitigate this variability we introduce a new, fast and easy to compute super metric that compares the volume, connectivity, medialness, bifurcation point identification and homology of the reconstructed skeletons to the original segmented data. We then show that such a metric can be used to select the best performing skeletonisation algorithm for a given dataset, as well as to optimise its parameters. Finally, we demonstrate that the super metric can also be used to quickly identify how a particular skeletonisation algorithm could be improved, becoming a powerful tool in understanding the complex implication of small structural changes in a network.

A method for denoising active source seismic data via Fourier transform and spectrum reconstruction

Denoising is a critical step in signal processing. We develop a method for random noise reduction in active source seismic data using spectrum reconstruction. Two methods are developed for modifying the observed data’s amplitude spectrum: one substitutes it with the source wavelet’s amplitude spectrum, whereas the other involves multiplying the source wavelet’s amplitude spectrum with the observed data’s amplitude spectrum. By reconstructing the modified amplitude spectrum while preserving the observed data’s phase spectrum, noise suppression is achieved. Extensive testing with theoretical models, synthetic shot gathers, and field data indicate a notable improvement in the signal-to-noise ratio (S/N) compared with the traditional band-pass filtering method. This method proves particularly effective for enhancing the S/N in the context of active source wide-angle seismic data used in offshore structural studies, eliminating the need for data segmentation based on offset, and thereby improving processing efficiency. Our method relies solely on a single complete cycle of the source wavelet, making it a purely data-driven solution. It has broad applications in processing active source or controlled source data with consistent source wavelets, including but not limited to seismic exploration, acoustic detection, and signal denoising in various ground-penetrating radars used on Mars, the moon, and earth.

The Spatiotemporal Evolution Mechanism of Urban Rail Transit Fault Propagation in Networked Operation Modes

The cascading propagation and evolution of metro operation failures can significantly impact the safety of metro operation. To overcome this challenge, this study pre-processes a massive amount of metro operation log data through noise reduction. Moreover, a professional terminology dictionary is constructed along with a custom stop-word dictionary to segment the preprocessed data. Subsequently, the AFP-tree algorithm is employed to mine the segmented log data and identify key hazards. A weighted urban rail transit network is established, considering the effective path time cost, and the shortest travel OD path. To simulate the dynamic evolution of the failure chain propagation, a model based on disaster propagation theory is constructed. Taking the Shanghai Metro line as a case, multiple simulation scenarios are established with 25 key hazards as triggering points, and the number of cascade failure stations affected under different scenarios is outputted. The results indicate that the fault stations caused by the large passenger flow are the largest. Meanwhile, the number of stations affected by the door clamp is the smallest. The scale of fault stations reaches a maximum value in 16–20 min. Through case analysis, a positive correlation is found when the self-recovery factor is between 14 and 18, and the number of fault stations shows a significant increasing trend. The research results can provide decision-making support and theoretical guidance for rail transit operation safety management enterprises.

Comparing eye-hand coordination between controller-mediated virtual reality, and a real-world object interaction task.

Virtual reality (VR) technology has advanced significantly in recent years, with many potential applications. However, it is unclear how well VR simulations mimic real-world experiences, particularly in terms of eye-hand coordination. This study compares eye-hand coordination from a previously validated real-world object interaction task to the same task re-created in controller-mediated VR. We recorded eye and body movements and segmented participants' gaze data using the movement data. In the real-world condition, participants wore a head-mounted eye tracker and motion capture markers and moved a pasta box into and out of a set of shelves. In the VR condition, participants wore a VR headset and moved a virtual box using handheld controllers. Unsurprisingly, VR participants took longer to complete the task. Before picking up or dropping off the box, participants in the real world visually fixated the box about half a second before their hand arrived at the area of action. This 500-ms minimum fixation time before the hand arrived was preserved in VR. Real-world participants disengaged their eyes from the box almost immediately after their hand initiated or terminated the interaction, but VR participants stayed fixated on the box for much longer after it was picked up or dropped off. We speculate that the limited haptic feedback during object interactions in VR forces users to maintain visual fixation on objects longer than in the real world, altering eye-hand coordination. These findings suggest that current VR technology does not replicate real-world experience in terms of eye-hand coordination.

A CNN-SAM-LSTM hybrid neural network for multi-state estimation of lithium-ion batteries under dynamical operating conditions

Accurately estimating the state of charge (SOC), state of energy (SOE), and state of health (SOH) online is a critical and urgent concern in the management of lithium-ion batteries for electric vehicle applications, particularly in terms of safety and reliability. This paper develops a hybrid neural network, abbreviated as CNN-SAM-LSTM model, which combines a convolutional neural network, self-attention mechanism, and long-short term memory neural network to jointly estimate the state parameters of lithium-ion batteries, including SOC, SOE, and SOH. Additionally, a joint loss function considering homoscedastic uncertainty is developed to optimize weight adjustments for the training losses associated with the three state parameters. Experimental data collected under UDDS, BBDST and CC discharge conditions are employed to showcase the effectiveness of the proposed CNN-SAM-LSTM model. The results demonstrate that the proposed model is capable of simultaneously and accurately estimating SOC, SOE, and SOH for lithium-ion batteries under different dynamical operating conditions. Moreover, when applied to randomly segmented data, the proposed model exhibits robustness, effectively handling deviations from random discharge segments.

Storing Images in DNA via base128 Encoding.

Current DNA storage schemes lack flexibility and consistency in processing highly redundant and correlated image data, resulting in low sequence stability and image reconstruction rates. Therefore, according to the characteristics of image storage, this paper proposes storing images in DNA via base128 encoding (DNA-base128). In the data writing stage, data segmentation and probability statistics are carried out, and then, the data block frequency and constraint encoding set are associated with achieving encoding. When the image needs to be recovered, DNA-base128 completes internal error correction by threshold setting and drift comparison. Compared with representative work, the DNA-base128 encoding results show that the undesired motifs were reduced by 71.2-90.7% and that the local guanine-cytosine content variance was reduced by 3 times, indicating that DNA-base128 can store images more stably. In addition, the structural similarity index (SSIM) and multiscale structural similarity (MS-SSIM) of image reconstruction using DNA-base128 were improved by 19-102 and 6.6-20.3%, respectively. In summary, DNA-base128 provides image encoding with internal error correction and provides a potential solution for DNA image storage. The data and code are available at the GitHub repository: https://github.com/123456wk/DNA_base128.

Semantic Segmentation of Remote Sensing Data Based on Channel Attention and Feature Information Entropy.

The common channel attention mechanism maps feature statistics to feature weights. However, the effectiveness of this mechanism may not be assured in remotely sensing images due to statistical differences across multiple bands. This paper proposes a novel channel attention mechanism based on feature information called the feature information entropy attention mechanism (FEM). The FEM constructs a relationship between features based on feature information entropy and then maps this relationship to their importance. The Vaihingen dataset and OpenEarthMap dataset are selected for experiments. The proposed method was compared with the squeeze-and-excitation mechanism (SEM), the convolutional block attention mechanism (CBAM), and the frequency channel attention mechanism (FCA). Compared with these three channel attention mechanisms, the mIoU of the FEM in the Vaihingen dataset is improved by 0.90%, 1.10%, and 0.40%, and in the OpenEarthMap dataset, it is improved by 2.30%, 2.20%, and 2.10%, respectively. The proposed channel attention mechanism in this paper shows better performance in remote sensing land use classification.

Dosiomics and radiomics-based prediction of pneumonitis after radiotherapy and immune checkpoint inhibition: The relevance of fractionation

ObjectivesPost-therapy pneumonitis (PTP) is a relevant side effect of thoracic radiotherapy and immunotherapy with checkpoint inhibitors (ICI). The influence of the combination of both, including dose fractionation schemes on PTP development is still unclear. This study aims to improve the PTP risk estimation after radio(chemo)therapy (R(C)T) for lung cancer with and without ICI by investigation of the impact of dose fractionation on machine learning (ML)-based prediction. Materials and MethodsData from 100 patients who received fractionated R(C)T were collected. 39 patients received additional ICI therapy. Computed Tomography (CT), RT segmentation and dose data were extracted and physical doses were converted to 2-Gy equivalent doses (EQD2) to account for different fractionation schemes. Features were reduced using Pearson intercorrelation and the Boruta algorithm within 1000-fold bootstrapping. Six single (clinics, Dose Volume Histogram (DVH), ICI, chemotherapy, radiomics, dosiomics) and four combined models (radiomics + dosiomics, radiomics + DVH + Clinics, dosiomics + DVH + Clinics, radiomics + dosiomics + DVH + Clinics) were trained to predict PTP. Dose-based models were tested using physical dose and EQD2. Four ML-algorithms (random forest (rf), logistic elastic net regression, support vector machine, logitBoost) were trained and tested using 5-fold nested cross validation and Synthetic Minority Oversampling Technique (SMOTE) for resampling in R. Prediction was evaluated using the area under the receiver operating characteristic curve (AUC) on the test sets of the outer folds. ResultsThe combined model of all features using EQD2 surpassed all other models (AUC = 0.77, Confidence Interval CI 0.76–0.78). DVH, clinical data and ICI therapy had minor impact on PTP prediction with AUC values between 0.42 and 0.57. All EQD2-based models outperformed models based on physical dose. ConclusionsRadiomics + dosiomics based ML models combined with clinical and dosimetric models were found to be suited best for PTP prediction after R(C)T and could improve pre-treatment decision making. Different RT dose fractionation schemes should be considered for dose-based ML approaches.

A bearing remaining life prediction method under variable operating conditions based on cross-transformer fusioning segmented data cleaning

Bearing remaining useful life (RUL) prediction research based on deep learning mostly emphasizes model performance and effective feature vectors, overlooking different densities of outlier distributions in vibration signals at varying degradation stages. Moreover, forecasting models focus on capturing cross-time dependencies, ignoring the dependencies between different variables. To solve these problems, this paper proposes an unsupervised segmented data cleaning algorithm and a RUL prediction framework adaptable to variable operating conditions. The method consists of four steps: (1) Multi-domain feature extraction and selection establish a feature vector space reflecting degradation trends. (2) Segmented data cleaning divides degradation stages, using different penalty factors for outlier cleaning. (3) Cleaned vibration signals undergo a second round of multidomain feature engineering and degradation-stage division. (4) A two-stage Cross-Transformer model is used for RUL prediction. The method proposed has been validated on the prognostics and health management (PHM) bearing degradation dataset. In the constant condition prediction task, the root mean square error (RMSE) and mean absolute error (MAE) were improved to 1.88 and 5.78, respectively. In the variable condition prediction task, the proposed method outperformed existing methods, with an improvement of 59.10 % in RMSE, demonstrating strong generalization performance and practical application value.