Network Prediction Research Articles

Causal relationship between complement C1QB and colorectal cancer: a drug target Mendelian randomization study.

Colorectal cancer is influenced by several factors such as unhealthy habits and genetic factors. C1QB has been linked to a number of malignancies. However, uncertainty surrounds the connection between C1QB and CRC. Therefore, this study aimed to explore a bidirectional causal relationship of C1QB as a drug target in CRC through Mendelian randomization (MR) analysis. The GWASs for C1QB and CRC were obtained from the Integrative Epidemiology Unit Open GWAS database. There were five strategies to investigate MR. Sensitivity analysis was carried out via tests for heterogeneity, horizontal pleiotropy and leave-one-out effects to evaluate the dependability of the MR analysis results. Furthermore, colocalization analysis of C1QB and CRC, protein-protein interaction network and drug prediction according to exposure factors as well as phenotype scanning were performed. The results of forward MR analysis demonstrated that C1QB was a risk factor for CRC (OR = 1.104, p = 0.033). However, we did not find a causal relationship between CRC and C1QB (reverse MR). Rs294180 and rs291985 corresponded to the same linkage interval and had the potential to influence C1QB and CRC, respectively. The PPI results demonstrated that C1QB interacted with 10 genes (C1QA, C1QC, C1R, C1S, C2, C4A, C4B, CALR, SERPING1, and VSIG4). Additionally, 21 medications were predicted to match C1QB. Molecular docking data, including for benzo(a)pyrene, 1-naphthylisothiocyanate, calcitriol and medroxyprogesterone acetate, revealed excellent binding for drugs and proteins. Moreover, we identified 29 diseases that were associated with C1QB and related medicines via disease prediction and intersection methods. As a therapeutic target for CRC, phenotypic scanning revealed that C1QB does not significantly affect weight loss, liver cirrhosis, or nonalcoholic fatty liver disease, but might have protective impacts on ovarian cancer and melanoma. The results highlight a causal relationship between C1QB and CRC and imply an oncogenic role for C1QB in CRC, as potential drug targets. Drugs designed to target C1QB have a greater chance of success in clinical trials and are expected to help prioritize CRC drug development and reduce drug development costs. That provided a theoretical foundation and reference for research on CRC and C1QB in MR.

Spatio-temporal ecological models via physics-informed neural networks for studying chronic wasting disease

To mitigate the negative effects of emerging wildlife diseases in biodiversity and public health it is critical to accurately forecast pathogen dissemination while incorporating relevant spatio-temporal covariates. Forecasting spatio-temporal processes can often be improved by incorporating scientific knowledge about the dynamics of the process using physical models. Ecological diffusion equations are often used to model epidemiological processes of wildlife diseases where environmental factors play a role in disease spread. Physics-informed neural networks (PINNs) are deep learning algorithms that constrain neural network predictions based on physical laws and therefore are powerful forecasting models useful even in cases of limited and imperfect training data. In this paper, we develop a novel ecological modeling tool using PINNs, which fits a feedforward neural network and simultaneously performs parameter identification in a partial differential equation (PDE) with varying coefficients. We demonstrate the applicability of our model by comparing it with the commonly used Bayesian stochastic partial differential equation method and traditional machine learning approaches, showing that our proposed model exhibits superior prediction and forecasting performance when modeling chronic wasting disease in deer in Wisconsin. Furthermore, our model provides the opportunity to obtain scientific insights into spatio-temporal covariates affecting spread and growth of diseases. This work contributes to future machine learning and statistical methodology development by studying spatio-temporal processes enhanced by prior physical knowledge.

Spatially explicit predictions of food web structure from regional-level data.

Knowledge about how ecological networks vary across global scales is currently limited given the complexity of acquiring repeated spatial data for species interactions. Yet, recent developments in metawebs highlight efficient ways to first document possible interactions within regional species pools. Downscaling metawebs towards local network predictions is a promising approach to using the current data to investigate the variation of networks across space. However, issues remain in how to represent the spatial variability and uncertainty of species interactions, especially for large-scale food webs. Here, we present a probabilistic framework to downscale a metaweb based on the Canadian mammal metaweb and species occurrences from global databases. We investigated how our approach can be used to represent the variability of networks and communities between ecoregions in Canada. Species richness and interactions followed a similar latitudinal gradient across ecoregions but simultaneously identified contrasting diversity hotspots. Network motifs revealed additional areas of variation in network structure compared with species richness and number of links. Our method offers the potential to bring global predictions down to a more actionable local scale, and increases the diversity of ecological networks that can be projected in space. This article is part of the theme issue 'Connected interactions: enriching food web research by spatial and social interactions'.

Optimization Study of Steady-State Aerial-Towed Cable Circling Strategy Based on BP Neural Network Prediction

This paper presents models for UAV aerial-towed cables in free-end and fixed-end configurations, crucial for tasks like communication and aerial charging. By establishing a quasi steady-state model, computational results on cable shapes are obtained. To accelerate computations, a backpropagation (BP) neural network prediction model is trained, significantly reducing the computation time. An evaluation function has been developed that integrates both aircraft performance and cable shape considerations to evaluate circling parameters across various states. This function integrates techniques such as BP neural networks and particle swarm optimization (PSO) to refine parameters such as velocities and bank angles for both free-end and fixed-end cables. The results show that the BP neural network accurately predicts cable shapes, achieving a maximum error of 5% in towing force and verticality. Additionally, PSO efficiently optimizes circling parameters, thereby enhancing the effectiveness of the evaluation function in identifying optimal solutions. This approach significantly improves the efficiency of determining optimal circling parameters for UAV aerial-towed cables, thereby contributing to their operational efficacy.

Prediction and Dynamic Simulation Verification of Output Characteristics of Radial Piston Motors Based on Neural Networks

Radial piston motors are executive components in hydraulic systems, tasked with providing appropriate torque and speed according to load requirements in practical applications. The purpose of this study is to predict the output torque of radial piston hydraulic motors and confirm their suitable operating conditions. Efficiency determination experiments were conducted on physical models, yielding thirty sets of performance data. Torque (output torque) and mechanical efficiency from the experimental data were selected as prediction targets and fitted using two methods: multiple linear regression and neural networks. A dynamic simulation model was built using Adams2020 software to obtain theoretical torque values, enabling the verification of the alignment between the predicted values and simulation results. The results indicate that the error between the theoretical torque of the dynamic model and the physical experiments is 1.9%, with the error of the neural network predictions being within 2%. The dynamic simulation model can yield highly accurate theoretical torque values, providing a reference for the external load of hydraulic motors; additionally, neural networks offer accurate predictions of output torque, thus reducing experimental testing costs.

Bidirectional long-short term memory predictor for material removal rate in computer-controlled optical surfacing

In Computer-Controlled Optical Surfacing technology, the precision of the removal function directly affects the accuracy of computer-aided processing software predictions, which in turn influences subsequent polishing machine processing. A key parameter for constructing the removal function is the material removal rate, which is often challenging to obtain accurately. Currently, the Preston equation is widely used to describe the principles of material removal. However, as a linear equation that omits many factors, it struggles to accurately model the removal function in complex machining scenarios. Therefore, this paper proposes a hybrid neural network model combining Convolutional Neural Networks and Bidirectional Long Short-Term Memory to predict the material removal rate. The model's parameters are optimized using an improved Grey Wolf Optimization algorithm, ultimately establishing a removal function closely consistent with an actual removal function. We first tested our method on the PHM2016 Data Challenge dataset, achieving a mean squared error of 6.19 and an R2 of 0.9949, outperforming other mainstream neural network prediction models developed in recent years. Additionally, we further validated the performance of the neural network using a small grinding head polishing dataset, achieving MSE and R2 values of 1.9035 and 0.99902, respectively. Finally, we applied this method to construct the removal function on an actual small grinding head production line. Compared to the traditional Preston equation-based removal function, the predicted residual surface's PV and RMS errors were reduced from 28.24 % to 35.58 %–4.563 % and 4.86 %, respectively. These validation results demonstrate that the proposed method not only facilitates easier acquisition of the removal function model but also significantly enhances the accuracy of computer-aided processing software predictions, thereby better guiding ultra-precision machining processes.

Reducing Manual Annotation Costs for Cell Segmentation by Upgrading Low-Quality Annotations.

Deep-learning algorithms for cell segmentation typically require large data sets with high-quality annotations to be trained with. However, the annotation cost for obtaining such sets may prove to be prohibitively expensive. Our work aims to reduce the time necessary to create high-quality annotations of cell images by using a relatively small well-annotated data set for training a convolutional neural network to upgrade lower-quality annotations, produced at lower annotation costs. We investigate the performance of our solution when upgrading the annotation quality for labels affected by three types of annotation error: omission, inclusion, and bias. We observe that our method can upgrade annotations affected by high error levels from 0.3 to 0.9 Dice similarity with the ground-truth annotations. We also show that a relatively small well-annotated set enlarged with samples with upgraded annotations can be used to train better-performing cell segmentation networks compared to training only on the well-annotated set. Moreover, we present a use case where our solution can be successfully employed to increase the quality of the predictions of a segmentation network trained on just 10 annotated samples.

A COMPARATIVE MODELING AND OPTIMIZATION OF SURFACE ROUGHNESS IN THE END MILLING OF Al 3003 SUBJECTED TO NON-EQUAL CHANNEL ANGULAR PRESSING (NECAP)

This paper highlights the surface roughness optimization of a specific material, Al 3003, which has been subjected to the non-equal channel angular pressing (NECAP) process. Considering spindle speed, feed rate, and depth of cut as input variables and surface roughness as an output variable, experiments have been conducted based on the L27 orthogonal array of the Taguchi method. Four prediction models, namely exponential and response surface methodology (RSM) as mathematical models, and artificial neural networks (ANNs) prediction models with different training algorithms (Bayesian Regularization (BR) and Levenberg–Marquardt (LM)), are proposed. Applying effectiveness and performance criteria, the prediction accuracy of the exponential model (90.35%), RSM (93.07%), BR (97.83%), and LM (97.54%) shows that all proposed prediction models are efficient enough. The ANN model trained with BR is found to be the best fit for predicting surface roughness. In order to optimize surface roughness, a newly introduced optimization method called the Intelligible-in-time Logics Algorithm (ILA) is employed. High spindle speed (1000[Formula: see text]rev/min), low feed rate (100[Formula: see text]mm/min) and depth of cut (0.5[Formula: see text]mm) have been the optimum cutting parameter combinations to obtain minimum surface roughness (0.4956[Formula: see text][Formula: see text]m). The results have been verified by confirmation tests and Particle Swarm Optimization (PSO) method. ILA and PSO predict the same optimum parameter combinations and minimum surface roughness, while ILA performs optimization in less time (114.4[Formula: see text]s), about 3.5 times faster than PSO. The paper’s findings strongly advocate the application of ILA in machining data optimization.

Reinforcement Learning Empowered Grey System Theory: Advancing Intelligence in Traffic Prediction for Wireless Sensor Networks

Reinforcement Learning Empowered Grey System Theory: Advancing Intelligence in Traffic Prediction for Wireless Sensor Networks

Artificial Neural Network Prediction of Mortality in Cancer Patients Presenting for Radiation Therapy at a Multisite Institution.

For many decades, the management of cancer has utilized radiation therapy, which continues to evolve with technology to improve patient outcomes. However, despite the standardization of treatment plans and the establishment of best clinical practices based on prospective, randomized trials and adherence to National Comprehensive Cancer Network (NCCN) guidelines, the outcomes from radiation therapy are highly variable and dependent on a number of factors, including patient demographics, tumor characteristics/histology, and treatment parameters. In this study, we attempt to use available patient data and treatment parameters at the time of radiation therapy to predict future outcomes using artificial intelligence (AI). Six thousand five hundred ninety-five cases of patients who completed radiation treatment were selected retrospectively and used to train artificial neural networks (ANNs) and baseline models (i.e., logistic regression, random forest, support vector machines [SVMs], gradient boosting [XGBoost]) for binary classification of mortality at multiple time points ranging from six months to five years post-treatment. A hyperparameter grid search was used to identify the optimal network architecture for each time point, using sensitivity as the primary outcome metric. The median age was 75 years (range: 2-102 years). There were 63.8% females and 36.1% males. The results indicate that ANNs were able to successfully perform binary mortality prediction with an accuracy greater than random chance and greater sensitivity than baseline models used. The best-performing algorithm was the ANN, which achieved a sensitivity of 83.00% ± 4.89% for five-year mortality. The neural network was able to achieve higher sensitivity than Logistic Regression, SVM Random Forest, and XGBoost across all output target variables, demonstrating the utility of a neural network model for mortality prediction on the provided dataset.

Heat transfer evaluation of combined nanofluids in vertical annulus: An experimental and artificial neural network approach

Enhancing the thermal properties of the working fluid is a practical way to improve heat transfer performance. One of the suggested fluids for this purpose is nanofluids, which are suspensions of nanoparticles within a base fluid. Nanofluids have the potential to operate as a primary coolant or as an emergency coolant in a nuclear reactor. This paper investigates the thermal performance of a homogeneous hybrid nanofluid of Al2O3 and TiO2 in deionized water. The thermo-hydraulic performance of the nanofluid is analyzed in a high-pressure test loop capable of 25 bar pressure. The test section has an annular geometry, generating a cosine heat flux pattern. This flux pattern is representative of nuclear reactor cores. The experimental study is conducted to investigate the effects of nanoparticle concentration and Reynolds number on the heat transfer performance of nanofluids. The results show that increasing these two parameters significantly improved heat transfer. Specifically, raising the nanoparticle concentration from 0% to 1% leads to a reduction in surface temperature of 19% and 8% at Reynolds numbers of 825 and 4128, respectively. The calculated Nusselt numbers are compared with correlations from previous studies. Furthermore, an artificial neural network model is developed to predict heat transfer data for the studied system. The model consists of a single-layer feed-forward neural network trained with the stochastic gradient descent algorithm. To determine the optimal network, the number of neurons is optimized and evaluated based on the network's predictability. The neural network model demonstrates substantial improvements over previous experimental correlations (MAPE <2.5% and MSE <0.001).

Predicting bridge condition index using an improved back-propagation neural network

The predicted Bridge Condition Index (BCI) is a guiding indicator for formulating preventive maintenance strategies for bridges and is hardly available even if using the most advanced neural network prediction model, owing to the few measured BCIs and neural network model defects. In this paper, an improved Back-Propagation Neural Network (BPNN) model named CPSO-BP-MC is proposed to predict BCI, which has a BPNN as its core, combined with the Coordinated Particle Swarm Optimization (CPSO) algorithm aims to faster convergence and avoid local optimal solution even with less training sample, and the Markov Chain (MC) algorithm aims to further revise the fluctuation of CPSO-BP prediction output due to maintenance and reinforcement. A case study is presented to demonstrate the efficiency of CPSO-BP-MC predicting BCI with fewer training samples. Comparing the predicted BCI of CPSO-BP-MC with the measured BCI and other models, the results show that CPSO-BP-MC converges faster than other models, and can predict the BCI more stably and accurately, and the error between prediction and measured BCI is less than 3 %.

Integrating Spatio-Temporal Graph Convolutional Networks with Convolutional Neural Networks for Predicting Short-Term Traffic Speed in Urban Road Networks

The rapid expansion of large urban areas underscores the critical importance of road infrastructure. An accurate understanding of traffic flow on road networks is essential for enhancing civil services and reducing fuel consumption. However, traffic flow is influenced by a complex array of factors and perpetually changing conditions, making comprehensive prediction of road network behavior challenging. Recent research has leveraged deep learning techniques to identify and forecast traffic flow and road network conditions, enhancing prediction accuracy by extracting key features from diverse factors. In this study, we performed short-term traffic speed predictions for road networks using data from Mobileye sensors mounted on taxis in Daegu City, Republic of Korea. These sensors capture the road network flow environment and the driver’s intentions. Utilizing these data, we integrated convolutional neural networks (CNNs) with spatio-temporal graph convolutional networks (STGCNs). Our experimental results demonstrated that the combined STGCN and CNN model outperformed the standalone STGCN and CNN models. The findings of this study contribute to the advancement of short-term traffic speed prediction models, thereby improving road network flow management.

Utilizing connectome fingerprinting functional MRI models for motor activity prediction in presurgical planning: A feasibility study.

Presurgical planning prior to brain tumor resection is critical for the preservation of neurologic function post-operatively. Neurosurgeons increasingly use advanced brain mapping techniques pre- and intra-operatively to delineate brain regions which are "eloquent" and should be spared during resection. Functional MRI (fMRI) has emerged as a commonly used non-invasive modality for individual patient mapping of critical cortical regions such as motor, language, and visual cortices. To map motor function, patients are scanned using fMRI while they perform various motor tasks to identify brain networks critical for motor performance, but it may be difficult for some patients to perform tasks in the scanner due to pre-existing deficits. Connectome fingerprinting (CF) is a machine-learning approach that learns associations between resting-state functional networks of a brain region and the activations in the region for specific tasks; once a CF model is constructed, individualized predictions of task activation can be generated from resting-state data. Here we utilized CF to train models on high-quality data from 208 subjects in the Human Connectome Project (HCP) and used this to predict task activations in our cohort of healthy control subjects (n = 15) and presurgical patients (n = 16) using resting-state fMRI (rs-fMRI) data. The prediction quality was validated with task fMRI data in the healthy controls and patients. We found that the task predictions for motor areas are on par with actual task activations in most healthy subjects (model accuracy around 90%-100% of task stability) and some patients suggesting the CF models can be reliably substituted where task data is either not possible to collect or hard for subjects to perform. We were also able to make robust predictions in cases in which there were no task-related activations elicited. The findings demonstrate the utility of the CF approach for predicting activations in out-of-sample subjects, across sites and scanners, and in patient populations. This work supports the feasibility of the application of CF models to presurgical planning, while also revealing challenges to be addressed in future developments. PRACTITIONER POINTS: Precision motor network prediction using connectome fingerprinting. Carefully trained models' performance limited by stability of task-fMRI data. Successful cross-scanner predictions and motor network mapping in patients with tumor.

Multi-layer Bundling as a New Approach for Determining Multi-scale Correlations Within a High-Dimensional Dataset

The growing complexity of biological data has spurred the development of innovative computational techniques to extract meaningful information and uncover hidden patterns within vast datasets. Biological networks, such as gene regulatory networks and protein-protein interaction networks, hold critical insights into biological features’ connections and functions. Integrating and analyzing high-dimensional data, particularly in gene expression studies, stands prominent among the challenges in deciphering these networks. Clustering methods play a crucial role in addressing these challenges, with spectral clustering emerging as a potent unsupervised technique considering intrinsic geometric structures. However, spectral clustering’s user-defined cluster number can lead to inconsistent and sometimes orthogonal clustering regimes. We propose the Multi-layer Bundling (MLB) method to address this limitation, combining multiple prominent clustering regimes to offer a comprehensive data view. We call the outcome clusters “bundles”. This approach refines clustering outcomes, unravels hierarchical organization, and identifies bridge elements mediating communication between network components. By layering clustering results, MLB provides a global-to-local view of biological feature clusters enabling insights into intricate biological systems. Furthermore, the method enhances bundle network predictions by integrating the bundle co-cluster matrix with the affinity matrix. The versatility of MLB extends beyond biological networks, making it applicable to various domains where understanding complex relationships and patterns is needed.

Heat and mass transfer dynamics in curved Ш-Chip: ISPH simulations and ANN analysis

This research investigates heat and mass transfer behavior in nano-enhanced phase change materials (NEPCM) within reactive systems using incompressible smoothed particle hydrodynamics (ISPH) coupled with artificial neural network (ANN) predictions. It introduces a novel model comprising six vertical rods and one curved rod, forming a unique Ш-Chip configuration. The complexity of the NEPCM model demonstrates its relevance in fluid dynamics and heat transfer analyses applicable across diverse fields including food processing, electronics cooling, chip manufacturing, heat exchangers, and solar energy systems. The study explores the influence of various dimensionless parameters such as the Frank-Kamenetskii number (Fk), Hartmann number (Ha), Soret and Dufour numbers (Sr&Du), Lewis number (Le), Rayleigh number (Ra), and fractional order parameter (α) on heat and mass transfer phenomena. Heat/mass sources from embedded vertical and curved rods filled with NEPCM are considered within the model. An artificial neural network (ANN) model employing a multilayer perceptron (MLP) structure is utilized to accurately predict the average Nusselt (Nu‾) and Sherwood (Sh‾) numbers, demonstrating its applicability in fluid dynamics and heat transfer analyses. Key insights highlight the role of Fk and Ra in enhancing convection flow and nanofluid velocity within the curved Ш-Chip. Furthermore, variations in Ha lead to observable velocity reductions due to intensified Lorentz forces, while increased Le values result in decreased velocities as thermal diffusion becomes dominant. The fractional order parameter (α) aids in the transition from unsteady to steady state. The unique configuration of the curved Ш-Chip, combined with the incorporation of heat/mass sources from embedded particles, offers promising applications across diverse fields such as food processing, electronics cooling, chip manufacturing, heat exchangers, and solar energy systems. This research underscores the crucial role of understanding the pertinent parameters to ensure the effective development of systems tailored to specific needs. Additionally, it emphasizes the practical value of employing ANN models in predictive modeling to tackle the complexities encountered in engineering applications.

The puzzle of loneliness: A sociostructural and transnational analysis of International Chinese Students' networks in Germany

AbstractThe experience of loneliness is common among international students who study in a foreign country, away from their home. Loneliness refers to the difference between the actual and desired social relationships highlighting not only the quantity but also the quality of relationships. We argue for a perspective on loneliness that emphasizes its social context rather than relying mainly on individualistic or psychological interpretations prevalent thus far. In this article, we explore network predictors of loneliness among Chinese international students in Germany. Through our analysis of a unique data set, we show that when compared to the local students, international students tend to feel lonelier. Surprisingly, when we discover that when international students have the same network size, those with more emotionally supportive ties tend to experience more loneliness compared to those with fewer emotionally supportive ties. Not surprisingly, having a romantic partner is associated with lower levels of loneliness, especially when they also live in the country of education. Additionally, we find that international Chinese students who struggle to form cross‐national friendships report higher levels of loneliness. By delving into these findings, our research offers a sociological account of loneliness among international Chinese and local students in Germany.

Сетевая платформа автоматизации прогнозирования отказов карьерных самосвалов

The article considers the possibilities of developing an automated system for monitoring and predicting the technical condition of in-pit vehicles at the operation stage based on failure statistics and network analysis of data received from health sensors of the mining machines. This study seeks to reduce emergency downtime in the mining industry by introducing modern information and communication technologies. The applicability of existing methods to analyze digital signals received from the sensors installed on the mining equipment was assessed. A promising approach is considered, using the progress achieved in network engineering and conversion of the time series signals into the integrated networks. A sequence of operations is proposed as an innovation, including collection and analysis of data, development of network prediction models and practical implementation of the results. It is expected that using such a sequence of steps will be able to promptly notify of the need to repair equipment, thereby reducing downtime, which in turn will increase productivity and reduce the operating costs. The main stages of the study are formulated and presented, the implementation of which is aimed at predicting the health of the equipment, identifying the need for unscheduled repairs, which will lead to a decrease in the number of emergency failures or their prevention in real operating conditions of mining enterprises.

Impact of surface temperature changes on subgrade settlement of permafrost regions railway: a case study of the Qinghai-Tibet railway, China

ABSTRACT Parts of sections in permafrost regions of the Qinghai-Tibet Railway (QTR) have been deformed due to permafrost degradation, the global warming and human activities. Therefore, monitoring the subgrade settlement of the QTR in permafrost regions has been a major working in QTR maintenance. For this reason, this paper investigates the subsidence characteristics and derivative mechanisms of railway subgrades in the permafrost regions of the QTR. Additionally, it analyses the impact mechanisms of surface temperature changes on the subsidence of permafrost subgrades. Building on this investigation, a C-I-GWO-BP neural network prediction model was developed to forecast subgrade settlement in permafrost regions. The findings reveal a substantial correlation (R2 = 0.834) between the deformation of railway subgrades in permafrost regions and changes in surface temperature. Meanwhile，with the intensification of global warming and human activities, the subsidence of permafrost subgrade has shown an increasing trend year by year. The primary areas of deformation are predominantly concentrated in ice-rich permafrost regions, including rivers, lakes, and snowmelt runoff areas. Additionally, when compared to traditional BP neural network models, the C-I-GWO-BP neural network prediction model proposed in this article showcases diminished prediction errors and heightened accuracy in identifying optimal weights and thresholds. This model offers technical support for the efficient, accurate, and automated prediction of railway subgrade subsidence in permafrost regions.

ScribSD+: Scribble-supervised medical image segmentation based on simultaneous multi-scale knowledge distillation and class-wise contrastive regularization

Despite that deep learning has achieved state-of-the-art performance for automatic medical image segmentation, it often requires a large amount of pixel-level manual annotations for training. Obtaining these high-quality annotations is time-consuming and requires specialized knowledge, which hinders the widespread application that relies on such annotations to train a model with good segmentation performance. Using scribble annotations can substantially reduce the annotation cost, but often leads to poor segmentation performance due to insufficient supervision. In this work, we propose a novel framework named as ScribSD+ that is based on multi-scale knowledge distillation and class-wise contrastive regularization for learning from scribble annotations. For a student network supervised by scribbles and the teacher based on Exponential Moving Average (EMA), we first introduce multi-scale prediction-level Knowledge Distillation (KD) that leverages soft predictions of the teacher network to supervise the student at multiple scales, and then propose class-wise contrastive regularization which encourages feature similarity within the same class and dissimilarity across different classes, thereby effectively improving the segmentation performance of the student network. Experimental results on the ACDC dataset for heart structure segmentation and a fetal MRI dataset for placenta and fetal brain segmentation demonstrate that our method significantly improves the student’s performance and outperforms five state-of-the-art scribble-supervised learning methods. Consequently, the method has a potential for reducing the annotation cost in developing deep learning models for clinical diagnosis.