Independent Subsets Research Articles

Examining Optimal Weights for the Youth Assessment and Screening Instrument in North Dakota

The purpose of deriving weights for factors is to identify and discover what items have the strongest relation to the outcome of interest to create an assessment with the best predictive validity. The current study examined and compared weighted and nonweighted models for the pre-screen of the Youth Assessment and Screening Instrument (YASI). The total sample of 6,175 youth from the state of North Dakota, United States, was divided into the calibration sample (33%) to build the models and was cross-validated across five independent subsets. The results were that no method consistently nor substantially outperformed the original method. The one exception was the risk total score produced through the logistic regression model substantially improved the predictive accuracy for violent dispositions after the initial YASI administration in the validation subsets. Therefore, it may be worth considering alternative models and warrants further exploration for the users of the scale.

Proteomic and genomic evaluation of plasma protein biomarkers for atrial fibrillation

Abstract Background Atrial fibrillation (AF) is a common arrhythmia and an important cause of ischemic stroke. Several plasma proteins have been suggested as biomarkers for the risk of AF and subsequent stroke in previous studies, but validation in independent cohorts is needed. Large-scale integration of proteomics with genetic and clinical data holds potential for refinement of disease prediction and risk stratification. Purpose This study aimed to evaluate previously suggested plasma protein biomarkers of AF and ischemic stroke using plasma protein measurements in a large cohort. Methods Ten tentative protein biomarkers previously associated with AF or ischemic stroke in AF patients, were evaluated in a nested case-control study comprising 49,757 individuals from the UK biobank (UKB). Plasma proteins were measured by normalized protein expression on the Olink platform. Outcomes were incident AF and ischemic stroke. The proteins were evaluated in Cox regression models adjusted for sex, age, and clinical risk factors. To assess whether the proteins were independent predictors, they were evaluated in sensitivity analyses with additional adjustment for NT-proBNP levels. Polygenic risk scores were calculated for plasma protein concentrations using the PRS-CS-auto algorithm and tested for association with AF or ischemic stroke in an independent subset of the UKB cohort without plasma protein measurements (n=379,392). Results All ten plasma proteins were associated with incident AF in the primary model, and eight were associated with incident ischemic stroke (P-values < 0.0025 for all). Adjusted for NT-proBNP levels, the proteins interleukin-6, GDF15, angiopoietin 2, and Troponin-I remained independently associated with AF, and the proteins GDF15, Cystatin-C, interleukin-6, ADAMTS13, and FGF23 remained independently associated with ischemic stroke. Polygenic risk scores for NT-proBNP and angiopoietin 2 were associated with increased rates of AF. No polygenic scores were significantly associated with ischemic stroke. Conclusion Large-scale proteomic evaluation validated ten plasma proteins associated AF and found eight to be associated with ischemic stroke. Seven proteins were associated with AF or ischemic stroke independently of NT-proBNP levels. Additional genomic evaluation provided evidence of associations between incident AF and genetically predicted plasma levels of angiopoietin 2 and NT-proBNP. As proteomic analyses of the plasma proteome become cheaper and more readily available, these findings could contribute to future risk stratification and selection of individuals for AF screening programs.Proteins associated with AFProteins associated with ischemic stroke

A novel superposition coding scheme based on polar code for multi‐user detection in underwater acoustic OFDM communication system

SummarySuperposition coding (SC) is a non‐orthogonal multiple access (NOMA) scheme for the downlink communication system, which allows signals of different users to be stacked and forwarded simultaneously in the same frequency band. The codeword for each user is assigned a different weight at the transmitter to avoid signal conflicts and interference after stacking, and successive interference cancellation (SIC) is adopted to decode. A novel SC scheme based on polar code, suitable for the downlink multi‐user underwater acoustic (UWA) communication system using orthogonal frequency division multiplexing (OFDM), is proposed in this paper. Polar code is utilized to create a nested code structure that is divided into several subsets. We exploit this feature of the polar codes and assign each subset to the corresponding user in the channel coding process. The information bits are placed on the independent subset for each user, and the random bits are placed on the subset of other users while all the users share the same set of frozen bits. Then, the codewords are stacked with different weights. The selection range of power factors can be expanded through the double superposition of the coding and the power domains, and higher system throughput and fairness can be achieved. The simulation and tank experiment results significantly verify the effectiveness of the proposed scheme, making it most suitable for the downlink UWA multi‐user communication systems.

Spaceborne lidar measurement of global cloud properties through machine learning

With a large footprint size, multiple scattering measurements of clouds from spaceborne lidar provide useful information about cloud physical properties, such as cloud optical depths and cloud droplet size, both during daytime and nighttime. A neural network algorithm, with a subset of cloud backscatter profiles of dual-polarization and dual-wavelength Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) lidar measurements during daytime as input variables and cloud physical properties derived from collocated Moderate Resolution Imaging Spectroradiometer (MODIS) multi-spectral measurements as output, is developed and evaluated with an independent subset of the collocated CALIPSO and MODIS measurements. The study suggests that with a receiver footprint size of 110 m, CALIPSO lidar measurements are sensitive to liquid-phase cloud optical depth variations from 0 to 25. A larger footprint size, thus more multiple scattering, is required for lidar to have sensitivities to all liquid-phase clouds. The technique can be applied to all 17 years of CALIPSO daytime and nighttime measurements and, thus, provides useful information about global distributions of cloud physical properties both during day and night.

Decoding the impact of neighboring amino acids on ESI-MS intensity output through deep learning

Peptide-level quantification using mass spectrometry (MS) is no trivial task as the physicochemical properties affect both response and detectability. The specific amino acid (AA) sequence affects these properties, however the connection between sequence and intensity output remains poorly understood. In this work, we explore combinations of amino acid pairs (i.e., dimer motifs) to determine a potential relationship between the local amino acid environment and MS1 intensity. For this purpose, a deep learning (DL) model, consisting of an encoder-decoder with an attention mechanism, was built. The attention mechanism allowed to identify the most relevant motifs. Specific patterns were consistently observed where a bulky/aromatic and hydrophobic AA followed by a cationic AA as well as consecutive bulky/aromatic and hydrophobic AAs were found important for the prediction of the MS1 intensity. Correlating attention weights to mean MS1 intensities revealed that some important motifs, particularly containing Trp, His, and Cys, were linked with low responding peptides whereas motifs containing Lys and most bulky hydrophobic AAs were often associated with high responding peptides. Moreover, Asn-Gly was associated with low response. The model predicts MS1 response with a mean average percentage error of ∼11 % and a Pearson correlation coefficient of ∼0.64. While dimer representation of peptide sequences did not improve predictive capacity compared to single AA representation in earlier work, this work adds valuable insight for a better understanding of peptide response in MS analysis. SignificanceMass spectrometry is not inherently quantitative, and the response of a compound relies not only on its concentration but also on the molecular composition. For mass spectrometry-based analysis of peptides, such as in bottom-up proteomics, this directly implies that the response cannot be used directly to quantify individual peptides. Moreover, the dependency of the response on the amino acid sequence of individual peptides remains poorly understood. Using a deep learning model based on a recurrent neural network with an attention mechanism, we here investigate how the presence of dimer motifs within a peptide affects the MS1 response through the analysis of intended equimolar peptide pools comprising almost 200,000 unique peptides in total. Not only do we identify certain dimer classes and specific dimers that substantially affect the MS1 response, but the model is also able to predict peptide intensity with low error rates within the independent test subset. The findings not only improve our understanding of the link between sequence and response for peptides but also highlight the potential of utilizing deep learning for developing methods allowing for absolute, label-free peptide quantification.

Characterising the stellar differential rotation based on largest-spot statistics from ground-based photometry

Aims. Stellar spot distribution has consequences on the observable periodic signals in long-time baseline ground-based photometry. We model the statistics of the dominating spots of two young and active Solar-type stars, V889 Her and LQ Hya, in order to obtain information on the underlying spot distribution, rotation of the star, as well as the orientation of the stellar axis of spin. Methods. By calculating estimates for spot-induced periodicities in independent subsets of photometric data, we obtain statistics based on the dominating spots in each subset, giving rise to largest-spot statistics accounting for stellar geometry and rotation, including differential rotation. Results. Our simple statistical models are able to reproduce the observed distribution of photometric signals rather well. This also enables us to estimate the dependence of angular velocity of the spots as a function of latitude. Our results indicate that V889 Her has a non-monotonic differential rotation curve with a maximum angular velocity between latitudes of 37–40 deg and lower angular velocity at the pole than the equator. Our results for LQ Hya indicate that the star rotates much like a rigid body. Furthermore, the results imply that the monotonic Solar differential rotation curve may not be a universal model for other solar-type stars. Conclusions. The non-monotonicity of the differential rotation of V889 Her is commonly produced in magnetohydrodynamic simulations, which indicates that our results are realistic from a theoretical perspective.

Detection and Segmentation of Mouth Region in Stereo Stream Using YOLOv6 and DeepLab v3+ Models for Computer-Aided Speech Diagnosis in Children

This paper describes a multistage framework for face image analysis in computer-aided speech diagnosis and therapy. Multimodal data processing frameworks have become a significant factor in supporting speech disorders’ treatment. Synchronous and asynchronous remote speech therapy approaches can use audio and video analysis of articulation to deliver robust indicators of disordered speech. Accurate segmentation of articulators in video frames is a vital step in this agenda. We use a dedicated data acquisition system to capture the stereovision stream during speech therapy examination in children. Our goal is to detect and accurately segment four objects in the mouth area (lips, teeth, tongue, and whole mouth) during relaxed speech and speech therapy exercises. Our database contains 17,913 frames from 76 preschool children. We apply a sequence of procedures employing artificial intelligence. For detection, we train the YOLOv6 (you only look once) model to catch each of the three objects under consideration. Then, we prepare the DeepLab v3+ segmentation model in a semi-supervised training mode. As preparation of reliable expert annotations is exhausting in video labeling, we first train the network using weak labels produced by initial segmentation based on the distance-regularized level set evolution over fuzzified images. Next, we fine-tune the model using a portion of manual ground-truth delineations. Each stage is thoroughly assessed using the independent test subset. The lips are detected almost perfectly (average precision and F1 score of 0.999), whereas the segmentation Dice index exceeds 0.83 in each articulator, with a top result of 0.95 in the whole mouth.

Deep transfer learning for detection of breast arterial calcifications on mammograms: a comparative study

IntroductionBreast arterial calcifications (BAC) are common incidental findings on routine mammograms, which have been suggested as a sex-specific biomarker of cardiovascular disease (CVD) risk. Previous work showed the efficacy of a pretrained convolutional network (CNN), VCG16, for automatic BAC detection. In this study, we further tested the method by a comparative analysis with other ten CNNs.Material and methodsFour-view standard mammography exams from 1,493 women were included in this retrospective study and labeled as BAC or non-BAC by experts. The comparative study was conducted using eleven pretrained convolutional networks (CNNs) with varying depths from five architectures including Xception, VGG, ResNetV2, MobileNet, and DenseNet, fine-tuned for the binary BAC classification task. Performance evaluation involved area under the receiver operating characteristics curve (AUC-ROC) analysis, F1-score (harmonic mean of precision and recall), and generalized gradient-weighted class activation mapping (Grad-CAM++) for visual explanations.ResultsThe dataset exhibited a BAC prevalence of 194/1,493 women (13.0%) and 581/5,972 images (9.7%). Among the retrained models, VGG, MobileNet, and DenseNet demonstrated the most promising results, achieving AUC-ROCs > 0.70 in both training and independent testing subsets. In terms of testing F1-score, VGG16 ranked first, higher than MobileNet (0.51) and VGG19 (0.46). Qualitative analysis showed that the Grad-CAM++ heatmaps generated by VGG16 consistently outperformed those produced by others, offering a finer-grained and discriminative localization of calcified regions within images.ConclusionDeep transfer learning showed promise in automated BAC detection on mammograms, where relatively shallow networks demonstrated superior performances requiring shorter training times and reduced resources.Relevance statementDeep transfer learning is a promising approach to enhance reporting BAC on mammograms and facilitate developing efficient tools for cardiovascular risk stratification in women, leveraging large-scale mammographic screening programs.Key points• We tested different pretrained convolutional networks (CNNs) for BAC detection on mammograms.• VGG and MobileNet demonstrated promising performances, outperforming their deeper, more complex counterparts.• Visual explanations using Grad-CAM++ highlighted VGG16’s superior performance in localizing BAC.Graphical

The number of 1-nearly independent vertex subsets

Let G be a graph with vertex set V(G) and edge set E(G). A subset I of V(G) is an independent vertex subset if no two vertices in I are adjacent in G. We study the number, σ 1(G), of all subsets of V(G) that contain exactly one pair of adjacent vertices. We call those subsets 1-nearly independent vertex subsets. Recursive formulas of σ 1 are provided, as well as some cases of explicit formulas. We prove a tight lower (resp. upper) bound on σ 1 for graphs of order n. We deduce as a corollary that the star K 1,n−1 (the tree with degree sequence (n − 1, 1, . , 1)) is the n-vertex tree with smallest σ 1, while it is well known that K1,n−1 is the n-vertex tree with largest number of independent subsets.

BGS: Accelerate GNN training on multiple GPUs

Emerging Graph Neural Networks (GNNs) have made significant progress in processing graph-structured data, yet existing GNN frameworks face scalability issues when training large-scale graph data using multiple GPUs. Frequent feature data transfers between CPUs and GPUs are a major bottleneck, and current caching schemes have not fully considered the characteristics of multi-GPU environments, leading to inefficient feature extraction. To address these challenges, we propose BGS, an auxiliary framework designed to accelerate GNN training from a data perspective in multi-GPU environments. Firstly, we introduce a novel training set partition algorithm, assigning independent training subsets to each GPU to enhance the spatial locality of node access, thus optimizing the efficiency of the feature caching strategy. Secondly, considering that GPUs can communicate at high speeds via NVLink connections, we designed a feature caching placement strategy suitable for multi-GPU environments. This strategy aims to improve the overall hit rate by setting reasonable redundant caches on each GPU. Evaluations on two representative GNN models, GCN and GraphSAGE, show that BGS significantly improves the hit rate of feature caching strategies in multi-GPU environments and substantially reduces the time overhead of data loading, achieving a performance improvement of 1.5 to 6.2 times compared to the baseline.

Introducing a new hazard and exposure atlas for European winter storms

Winter storms are one of the riskiest natural hazards in Europe. They are often assessed either from a climatology perspective or the exposed land use type. However, for accurate storm protection management, analyzing compatible wind speed and land use data at the highest spatiotemporally accuracy is critical. Thus, the study goals were (1) modeling wind speed for the 27 most extreme winter storm events from 1993 to 2022 at high spatial resolution (250 m × 250 m) and (2) defining and developing new and representative storm hazard and storm exposure indices for forested area, built-up area, and population. The winter storm event-related wind speed was modeled using a least-squares boosting approach based on ERA5 reanalysis data and 1730 wind speed measurement sites from the integrated surface global hourly meteorological dataset. In combination with temporally variable population density, built-up area, and forested area grids, the storm exposure indices for the European NUTS3 regions were calculated. The coefficient of determination of the median of the modeled wind speed fields is 0.73 for an independent subset of measurement sites. The wind speed maps reveal the uniqueness of the winter storm events. Each storm only affects parts of Europe. The study also emphasizes the relevance of distinguishing between storm hazard and storm exposure. The exposure of built-up areas to winter storms has increased considerably in recent years. This example underlines that storm exposure is not temporally constant but depends on the development of wind speed and land use. Decision-makers can apply the results of this study to evaluate and improve the effectiveness of storm protection measures. The developed storm exposures indices enable a sector- and location-specific evaluation with high precision.

En-WBF: A Novel Ensemble Learning Approach to Wastewater Quality Prediction Based on Weighted BoostForest

With the development of urbanization, the accurate prediction of effluent quality has become increasingly critical for the real-time control of wastewater treatment processes. The conventional method for measuring effluent biochemical oxygen demand (BOD) suffers from significant time delays and high equipment costs, making it less feasible for timely effluent quality assessment. To tackle this problem, we propose a novel approach called En-WBF (ensemble learning based on weighted BoostForest) to predict effluent BOD in a soft-sensing manner. Specifically, we sampled several independent subsets from the original training set by weighted bootstrap aggregation to train a series of gradient BoostTrees as the base models. Then, the predicted effluent BOD was derived by weighting the base models to produce the final prediction. Experiments on real datasets demonstrated that on the UCI dataset, the proposed En-WBF approach achieved a series of improvements, including by 28.4% in the MAE, 40.9% in the MAPE, 29.8% in the MSE, 18.2% in the RMSE, and 2.3% in the R2. On the Fangzhuang dataset, the proposed En-WBF approach achieved a series of improvements, including by 8.8% in the MAE, 9.0% in the MAPE, 12.8% in the MSE, 6.6% in the RMSE, and 1.5% in the R2. This paper contributes a cost-effective and timely solution for wastewater treatment management in real practice with a more accurate effluent BOD prediction, validating the research in the application of ensemble learning methods for environmental monitoring and management.

Enhancing the soluble expression of α-1,2-fucosyltransferase in E. coli using high-throughput flow cytometry screening coupled with a split-GFP

2’-Fucosyllactose (2’-FL), one of the major human milk oligosaccharides, was produced in several engineered microorganisms. However, the low solubility of α-1,2-fucosyltransferase (α1,2-FucT) often becomes a bottleneck to produce maximum amount of 2’-FL in the microorganisms. To overcome this solubility issue, the following studies were conducted to improve the soluble expression of α1,2-FucT. Initially, hydrophobic amino acids in the hydrophilic region of the 6 α-helices were mutated, adhering to the α-helix rule. Subsequently, gfp11 was fused to the C-terminal of futC gene encoding α1,2-FucT (FutC), enabling selection of high-fluorescence mutants through split-GFP. Each mutant library was screened via fluorescence activated cell sorting (FACS) to separate soluble mutants for high-throughput screening. As a result, L80C single mutant and A121D/P124A/L125R triple mutant were found, and a combined quadruple mutant was created. Furthermore, we combined mutations of conserved sequences (Q150H/C151R/Q239S) of FutC, which showed positive effects in the previous studies from our lab, with the above quadruple mutants (L80C/A121D/P124A/L125R). The resulting strain produced approximately 3.4-fold higher 2′-FL titer than that of the wild-type, suggesting that the conserved sequence mutations are an independent subset of the mutations that further improve the solubility of the target protein acquired by random mutagenesis using split-GFP.

Claudin 18.2 as a novel therapeutic target.

Claudin 18.2, a tight-junction molecule predominantly found in the nonmalignant gastric epithelium, becomes accessible on the tumour cell surface during malignant transformation, thereby providing an appealing target for cancer therapy. Data from two phase III trials testing the anti-claudin 18.2 antibody zolbetuximab have established claudin 18.2-positive advanced-stage gastric cancers as an independent therapeutic subset that derives benefit from the addition of this agent to chemotherapy. This development has substantially increased the percentage of patients eligible for targeted therapy. Furthermore, newer treatments, such as high-affinity monoclonal antibodies, bispecific antibodies, chimeric antigen receptor T cells and antibody-drug conjugates capable of bystander killing effects, have shown considerable promise in patients with claudin 18.2-expressing gastric cancers. This new development has resulted from drug developers moving beyond traditional targets, such as driver gene alterations or growth factors. In this Review, we highlight the biological rationale and explore the clinical activity of therapies that target claudin 18.2 in patients with advanced-stage gastric cancer and explore the potential for expansion of claudin 18.2-targeted therapies to patients with other claudin 18.2-positive solid tumours.

PECAN Predicts Patterns of Cancer Cell Cytostatic Activity of Natural Products Using Deep Learning.

Many machine learning techniques are used as drug discovery tools with the intent to speed characterization by determining relationships between compound structure and biological function. However, particularly in anticancer drug discovery, these models often make only binary decisions about the biological activity for a narrow scope of drug targets. We present a feed-forward neural network, PECAN (Prediction Engine for the Cytostatic Activity of Natural product-like compounds), that simultaneously classifies the potential antiproliferative activity of compounds against 59 cancer cell lines. It predicts the activity to be one of six categories, indicating not only if activity is present but the degree of activity. Using an independent subset of NCI data as a test set, we show that PECAN can reach 60.1% accuracy in a six-way classification and present further evidence that it classifies based on useful structural features of compounds using a "within-one" measure that reaches 93.0% accuracy.

Three remarks on W2 graphs

Let k≥1. A graph G is Wk if for any k pairwise disjoint independent vertex subsets A1,…,Ak in G, there exist k pairwise disjoint maximum independent sets S1,…,Sk in G such that Ai⊆Si for i∈[k]. Recognizing W1 graphs is coNP-hard, as shown by Chvátal and Slater (1993) and, independently, by Sankaranarayana and Stewart (1992). Extending this result and answering a recent question of Levit and Tankus, we show that recognizing Wk graphs is coNP-hard for k≥2. On the positive side, we show that recognizing Wk graphs is, for each k≥2, FPT parameterized by clique-width and by tree-width. Finally, we construct graphs G that are not W2 such that, for every vertex v in G and every maximal independent set S in G−N[v], the largest independent set in N(v)∖S consists of a single vertex, thereby refuting a conjecture of Levit and Tankus.

Don’t be late! Timely identification of cognitive impairment in people with multiple sclerosis: a study protocol

BackgroundCognitive impairment occurs in up to 65% of people with multiple sclerosis (PwMS), negatively affecting daily functioning and health-related quality of life. In general, neuropsychological testing is not part of standard MS-care due to insufficient time and trained personnel. Consequently, a baseline assessment of cognitive functioning is often lacking, hampering early identification of cognitive decline and change within a person over time. To assess cognitive functioning in PwMS in a time-efficient manner, a BICAMS-based self-explanatory digital screening tool called the Multiple Screener©, has recently been developed. The aim of the current study is to validate the Multiple Screener© in a representative sample of PwMS in the Netherlands. Additionally, we aim to investigate how cognitive functioning is related to psychological factors, and both work and societal participation.MethodsIn this cross-sectional multicentre study, 750 PwMS (aged 18–67 years) are included. To obtain a representative sample, PwMS are recruited via 12 hospitals across the Netherlands. They undergo assessment with the Minimal Assessment of Cognitive Functioning in MS (MACFIMS; reference-standard) and the Multiple Screener©. Sensitivity, specificity, and predictive values for identifying (mild) cognitive impairment are determined in a subset of 300 participants. In a second step, the identified cut-off values are tested in an independent subset of at least 150 PwMS. Moreover, test–retest reliability for the Multiple Screener© is determined in 30 PwMS. Information on psychological and work-related factors is assessed with questionnaires.DiscussionValidating the Multiple Screener© in PwMS and investigating cognition and its determinants will further facilitate early identification and adequate monitoring of cognitive decline in PwMS.

Effect of feature optimization on performance of machine learning models for predicting traffic incident duration

Developing a high-performing traffic incident-duration prediction model is considered a key component for evaluating the impact of these incidents on the roadway network. Various research studies have developed robust incident-duration prediction models. Still, they have faced many issues in providing an accurate prediction result due to the countless data modeling issues, such as complex correlations, highly skewed data distributions, heteroscedasticity, and outliers. This paper investigates the impact of feature optimization (FO) - a relatively new term encompassing two already-known topics: feature engineering (FE) and feature selection (FS) techniques - on the performance of several machine learning models developed for predicting incident durations. The models developed included multivariate linear regression, decision trees, support vector regressors, K-Nearest Neighbors regression, ensembles, and artificial neural networks. Various FO techniques have been used for each model to derive the massive traffic incidents dataset and repeat the prediction process. Our results show that the proposed filtering, wrapper, and embedded FS techniques can successfully reduce the number of features without sacrificing the prediction performance. Using log-normal transformation to deal with continuous data skewness, min-max normalization to deal with data variability, and principal component analysis (PCA) to reform the dataset into a smaller independent feature subset, FE techniques can enhance the accuracy of incident duration estimation over the assessed ML models. The best-performing FE technique was the PCA since performance improvements were observed across all developed ML models. The best-performing FS technique was the Recursive Feature Elimination, outperforming other tested techniques in reducing model complexity while maintaining model accuracy.

End-to-end volumetric segmentation of white matter hyperintensities using deep learning

Background and objectivesReliable detection of white matter hyperintensities (WMH) is crucial for studying the impact of diffuse white-matter pathology on brain health and monitoring changes in WMH load over time. However, manual annotation of 3D high-dimensional neuroimages is laborious and can be prone to biases and errors in the annotation procedure. In this study, we evaluate the performance of deep learning (DL) segmentation tools and propose a novel volumetric segmentation model incorporating self-attention via a transformer-based architecture. Ultimately, we aim to evaluate diverse factors that influence WMH segmentation, aiming for a comprehensive analysis of the state-of-the-art algorithms in a broader context. MethodsWe trained state-of-the-art DL algorithms, and incorporated advanced attention mechanisms, using structural fluid-attenuated inversion recovery (FLAIR) image acquisitions. The anatomical MRI data utilized for model training was obtained from healthy individuals aged 62–70 years in the Live active Successful Aging (LISA) project. Given the potential sparsity of lesion volume among healthy aging individuals, we explored the impact of incorporating a weighted loss function and ensemble models. To assess the generalizability of the studied DL models, we applied the trained algorithm to an independent subset of data sourced from the MICCAI WMH challenge (MWSC). Notably, this subset had vastly different acquisition parameters compared to the LISA dataset used for training. ResultsConsistently, DL approaches exhibited commendable segmentation performance, achieving the level of inter-rater agreement comparable to expert performance, ensuring superior quality segmentation outcomes. On the out of sample dataset, the ensemble models exhibited the most outstanding performance. ConclusionsDL methods generally surpassed conventional approaches in our study. While all DL methods performed comparably, incorporating attention mechanisms could prove advantageous in future applications with a wider availability of training data. As expected, our experiments indicate that the use of ensemble-based models enables the superior generalization in out-of-distribution settings. We believe that introducing DL methods in the WHM annotation workflow in heathy aging cohorts is promising, not only for reducing the annotation time required, but also for eventually improving accuracy and robustness via incorporating the automatic segmentations in the evaluation procedure.

Snowfall deposition in mountainous terrain: a statistical downscaling scheme from high-resolution model data on simulated topographies

One of the primary causes of non-uniform snowfall deposition on the ground in mountainous regions is the preferential deposition of snow, which results from the interaction of near-surface winds with topography and snow particles. However, producing high-resolution snowfall deposition patterns can be computationally expensive due to the need to run full atmospheric models. To address this, we developed two statistical downscaling schemes that can efficiently downscale near-surface, low-resolution snowfall data to fine-scale snow deposition accounting for the effect of preferential deposition in mountainous regions. Our approach relies on a comprehensive, model database generated using 3D wind fields from an atmospheric model and a preferential deposition model on several thousand simulated topographies covering a broad range in terrain characteristics. Both snowfall downscaling schemes rely on fine-scale topographic scaling parameters and low-resolution wind speed as input. While one scheme, referred to as the “wind scheme”, further necessitates fine-scale vertical wind components, a second scheme, referred to as the “aspect scheme”, does not require fine-scale temporal input. We achieve this by additionally downscaling near-surface vertical wind speed solely using topographic scaling parameters and low-resolution wind direction. We assess the performance of our downscaling schemes using an independent subset of the model database on simulated topographies, model data on actual terrain, and spatially measured new snow depth obtained through a photogrammetric drone survey following a snowfall event on previously snow-free ground. While the assessments show that our downscaling schemes perform well (relative errors ≤ ±3% with modeled and ≤ ±6% with measured snowfall deposition), they also demonstrate comparable results to benchmark downscaling models. However, our schemes notably outperform the benchmark models in representing fine-scale patterns. Our downscaling schemes possess several key features, including high computational efficiency, versatility enabled by the comprehensive model database, and independence from fine-scale temporal input data (aspect scheme), indicating their potential for widespread applicability. Therefore, our downscaling schemes for near-surface snowfall and vertical wind speed can be beneficial for various applications at fine grid resolutions such as in atmospheric and climate sciences, snow hydrology, glaciology, remote sensing, and avalanche sciences.