ABSTRACT Traffic data imputation is essential for intelligent transportation systems. Existing physics‐informed deep learning methods typically employ shared feature extraction networks and global traffic flow parameters, thereby overlooking spatiotemporal heterogeneity in speed‐flow coupling strength and fundamental diagram parameters. To address these limitations, we propose the Adaptive Physics‐Informed Spatiotemporal Heterogeneity Learning Network (API‐HNet). API‐HNet introduces a learnable bidirectional coupling mechanism based on cosine similarity to model heterogeneous speed‐flow interactions. Free‐flow speed and jam density are learned for each station‐day pair and incorporated into a physics‐consistency loss for end‐to‐end optimization. Experiments on the Shanghai and U.S. datasets showed that API‐HNet outperformed nine baseline methods under random missing and block missing scenarios, reducing RMSE by 3.29%–15.29% and MAE by 2.26%–13.01% while maintaining comparable inference efficiency. The model also captured variations in the coupling strength and fundamental diagram parameters that were consistent with traffic flow theory, thereby enhancing its physical interpretability.

Urban commuting flow prediction is crucial for optimizing public transportation and improving efficiency, yet traditional models often focus on geographic adjacency, overlooking the complex cross-regional interactions within transportation networks. To address this, we propose a Geography-Aware Graph Neural Network (GAGNN) model for commuting flow prediction. The model first jointly encodes the geographic adjacency matrix and semantic adjacency from public transportation networks, developing a comprehensive attention mechanism to fuse regional proximity with cross-regional semantic connectivity. Subsequently, a Graph Attention Network (GAT) is employed to embed the multiple adjacency relations and multi-source geographic knowledge. Finally, graph embeddings are combined with spatial factors into multidimensional feature vectors, fed into an MLP for commuting flow prediction. The model was validated with Fuzhou workday mobile phone data from January to February 2023, assessing the impact of semantic adjacency from different transportation networks on performance. The results show that: (1) We proposed the GAGNN outperforms both traditional models and advanced graph neural network models (e.g., GSGNN), reducing MAE by 14.9% and improving CPC by 2.1%; (2) The type of semantic adjacency significantly impacts model prediction accuracy. Road-based semantic connections perform best, especially for long-distance commuting flows, followed by metro and bus semantic connections, while the absence of semantic connections yields the worst performance. (3) Spatial scale significantly affects model prediction performance. Under road-based semantic adjacency, accuracy slightly declines with increasing scale, whereas metro, bus, and non-semantic connections, prediction accuracy improves. These findings offer effective support for accurate regional commuting flow modeling and public transportation networks optimization.

Fine-grained human mobility prediction at the urban scale is crucial for urban traffic management and sustainable development. However, existing models struggle to balance prediction accuracy and computational efficiency in urban-scale, fine-grained scenarios: models based on Euclidean structure suffer from limited accuracy, while models based on non-Euclidean structure face prohibitive computational costs. To address this challenge, we propose a Graph-Prior-Enhanced Euclidean Convolutional Network (GPECN) for efficient and accurate citywide mobility prediction. First, inspired by structural similarity, we present a virtual graph based on Wasserstein distance to capture the irregular and heterogeneous spatial relationships among OD pairs. Second, we embed graph priors into a Euclidean grid, forming a graph-prior-enhanced structure that preserves the accuracy advantages of graph models while retaining the computational efficiency of grid-based models. Finally, we explicitly incorporate the graph priors into the forward propagation of a classical convolutional neural network, enabling effective emulation of urban-scale graph feature aggregation under a low computational load. Using real world OD flow data from Xiamen and Fuzhou, China, we demonstrated the effectiveness of the proposed model at the urban scale. Experimental results show that, compared with five baseline models, the proposed GPECN model achieves prediction accuracy on a par with that of graph models while significantly improving computational efficiency.

The aim of this study was to evaluate the effect of upfront autologous stem cell transplantation (ASCT) on the prognosis in patients with advanced-stage extra-nodal NK/T-cell lymphoma (NKTCL) achieving first complete remission (CR1) after frontline therapy. To this end, we retrospectively reviewed data from patients diagnosed with stage III or IV NKTCL who achieved CR1 after frontline treatment between 2006 and 2023 at 14 medical centers in China. A total of 107 patients were included in this study. The majority of patients (87%) received non-anthracycline-based first-line chemotherapy, and 38 (36%) received upfront ASCT consolidation at the time of CR1. With a median follow-up time of 39.8 months, progression-free survival (PFS) and overall survival (OS) rates were significantly better in the ASCT group than in the non-ASCT group (3-year PFS: 78.2% vs. 54.6%, p = 0.005; 3-year OS: 86.0% vs. 68.9%, p = 0.04). However, in the subgroup of patients receiving non-anthracycline-based chemotherapy (N = 93), ASCT was significantly associated with better PFS (3-year PFS: 77.6% vs. 59.3%, p = 0.025) but not with OS (3-year OS: 85.6% vs. 74.8%, p = 0.164). In multivariate analyzes, upfront ASCT was an independent predictor of better PFS (hazard ratio [HR] = 0.37, 95% confidence interval [CI]: 0.15-0.88, p = 0.025) but not OS after adjusting for baseline prognostic index of NK-cell lymphoma (PINK), types of first-line chemotherapy, and local radiotherapy. After propensity score matched analyzes, upfront ASCT remained significantly associated with better PFS but not with OS. These real-world data suggest upfront ASCT prolongs PFS in patients with advanced-stage NKTCL in CR1, yet its impact on OS remains unclear.

BACKGROUND:Primary central nervous system lymphoma (PCNSL) is a rare extranodal lymphoma characterized by a poor prognosis due to high relapse rates and a lack of standardized treatment. This study aimed to evaluate the impact of induction/consolidation therapy on long-term survival and to provide extended follow-up data. METHODS:In this retrospective analysis, 140 immunocompetent patients with diffuse large B-cell PCNSL (DLBCL-PCNSL) treated at two centers between 2014 and 2024 were enrolled. Treatment efficacy was assessed based on baseline characteristics, therapeutic regimens, and treatment response. Progression-free survival (PFS) and overall survival (OS) were estimated using the Kaplan-Meier method, and prognostic factors were identified using multivariate Cox proportional hazards regression models. RESULTS:With a median follow-up of 5.3 years (range: 0.1-11.0 years), the 2- and 5-year PFS rates were 50.4% (95% CI: 42.1-60.2) and 34.1% (95% CI: 25.5-45.0), respectively, while the corresponding OS rates were 85.3% (95% CI: 79.4-91.6) and 60.8% (95% CI: 52.0-71.1). No survival plateau was observed. Among patients, 94% received methotrexate-based induction therapy: 94 received rituximab-methotrexate-temozolomide (R-MT) and 17 received MT alone, with 2-year PFS rates of 57.7% and 39.7%, respectively. Overall, 75% of patients achieved remission (CR/CRu/PR) after induction, and among these, 55% underwent consolidation therapy, predominantly autologous stem cell transplantation (ASCT, 90%) or whole-brain radiotherapy (10%). Patients receiving ASCT exhibited superior survival outcomes compared to those who did not. CONCLUSIONS:R-MT induction combined with ASCT consolidation is associated with improved survival in PCNSL, although relapse risk remains substantial. Outcomes remain poor in refractory subgroups, highlighting the need for novel therapeutic strategies.

Although numerous models have been proposed to predict the intensity of human activities in urban areas, two major issues hamper the performance of existing models: (1) fail to incorporate appropriate prior knowledge instrumental for improving accuracy and interpretability; (2) fail to integrate probabilistic and deterministic predictions to achieve complementary strengths, namely uncertainty quantification and high predictive accuracy. To address these challenges, we proposed a prior-enhanced dual-mode spatiotemporal graph neural network (PED-STGNN) to support both probabilistic and deterministic predictions. Specifically, we introduced a hypergraph node-to-vector (hypernode2vec) method to capture the multivariate functional similarity prior derived from complex and multivariate relations between urban regions. This functional similarity characterizes urban systems more precisely than existing methods relying on first-order pairwise relations. It improves accuracy and interpretability while enabling spatial modeling of higher-order multivariate relations beyond first-order pairwise relations. We also designed a plug-and-play probabilistic prediction module that enables switches between probabilistic and deterministic modes. Experiments based on the human activity intensity in Fuzhou, China, demonstrated the advantages in accuracy, interpretability and multi-scenario applicability.

Accurately and quickly forecasting the future state of urban sensors is crucial for urban monitoring and management. Although many forecasting approaches have been proposed, existing models still face two major challenges. First, most approaches do not have the ability to automatically handle missing data. Second, most approaches have high complexity, neglecting the usability and lightweight of the approach. Therefore, we present a lightweight spatiotemporal dilation approach tolerating missing data (STDM) to address the aforementioned challenges. First, we integrate a missing data handling mechanism into the STDM approach to enhance its forecasting capability under missing scenarios. Second, we present a lightweight spatiotemporal dilation component to enhance the inference speed of the STDM approach. Finally, we design the STDM approach as a separable architecture and define a corresponding loss function, allowing the STDM approach to be compatible with both forecasting tasks under missing and non-missing scenarios. The approach underwent validation using traffic, PM2.5, and temperature datasets. It exhibited superior forecasting accuracy and inference speed across four missing scenarios, outperforming eight baselines. Codes and data are available at link on https ://doi.org/10.6084/m9.figshare.24289456.

Modeling population-level human mobility has been attracting multidisciplinary research attention due to its profound implications for sustainable urban development. However, previous studies have often neglected the explicit consideration of spatial heterogeneity of travel demand, which limits their abilities to accurately estimate mobility flows. In this study, we introduce a prior-guided, data-driven human mobility model that integrates the position of origins and destinations, spatial travel patterns, and physical models as priors to capture spatial heterogeneity of human mobility. Specifically, we introduce the concept of 'relative attractiveness' to emulate the underlying driving force for the formation of spatial heterogeneity in human mobility. To learn the embeddings of 'relative attractiveness', we propose a suite of methods that integrate prior knowledge and graph neural networks, mainly including a relative position encoding module to encode the position of different origin-destination (OD) pairs relative to the entire geographical space and a message-passing method inspired by the classical physical models to simulate the mechanisms of mobility flow generation. Finally, a gradient boosting regression tree is trained to generate the mobility flow based on the learned embeddings. Extensive experiments on two real-world datasets have showed our model outperforms state-of-the-art data-driven mobility models in terms of accuracy and generalization.

Instance segmentation of remote sensing imagery (RSI) is vital for applications like geographic information system (GIS) updates and urban planning. Due to RSI's diversity (e.g., scale variations and complex object shapes), instance segmentation models require extensive labeled data, which is costly and labor-intensive. To mitigate this dependence on extensive labeled data, we propose a reinforcement learning (RL)-driven iteratively-refined instance segmentation model for RSI by RL (RL-ISegNet) framework, which uses policy gradient optimization to enhance the learning of hard samples in limited labeled data, thereby improving data utilization efficiency under limited samples. Specifically, our RL-ISegNet defines a sequential optimization process for hard samples, using cumulative rewards from each subtask (region proposal, detection, and segmentation) to provide future feedback guidance for optimizing network parameters. Meanwhile, considering the conflicts between subtasks, we propose a metric to quantify the positive or negative impacts between subtasks as a loss constraint, thereby increasing the sampling probability of trajectories with positive impacts. Additionally, we developed a Fisher information-based method to freeze low uncertainty parameters, reducing the training time. The experimental results show that our RL-ISegNet improves data utilization efficiency by +12.2% and achieves improvements of +7.1% in instance segmentation metrics with 4% of the training dataset.

Objective:To investigate the efficacy and safety of zanubrutinib in the treatment of autoimmune cytopenia(AIC)secondary to indolent B-cell lymphoma(iBCL).Methods:A total of 23 patients with iBCL complicated with AIC who were admitted to our hospital from December 2019 to September 2023 were selected as the research subjects.All patients were administered zanubrutinib 160 mg,twice daily,and continued oral administration.The objective response rate(ORR)of AIC,the therapeutic effect on lymphoma,and the incidence of adverse reactions were observed.Results:After a median follow-up of 20(5 to 48)months,the median duration of response was 9(interquartile range[IQR]5-24)months.AICA efficacy assessment showed that there were 10 cases of complete remission(CR),9 cases of partial remission(PR),and 4 cases of no response(NR),and the ORR was 82.6％(19/23)(95％CI:61.2-95.0).Among them,for the 14 patients with autoimmune hemolytic anemia(AIHA),7 achieved CR,5 had PR,and 2 had NR.For the 4 patients with immune thrombocytopenia(ITP),1 reached CR,2 had PR,and 1 had NR.Regarding the 5 patients with Evans syndrome(ES),2 achieved CR,2 had PR,and 1 had NR.The assessment of lymphoma efficacy showed that there were 10 cases of CR,7 cases of PR,6 cases of stable disease(SD),and no progressive cases,with an ORR of 73.9％(17/23)(95％CI:51.6-89.8).The main adverse reactions during the treatment were infection,hemorrhage,neutropenia,elevated lymphocyte count,rash,and anemia.Most of these adverse reactions were grade 1-2 and tolerable.No arrhythmia and hypertension occurred,and no deaths due to adverse reactions.Conclusion:Zanubrutinib is effective and safe for AIC secondary to iBCL.

The accurate prediction of origin-destination (OD) flows is essential for advancing sustainable urban mobility and supporting resilient urban planning. However, the inherent heterogeneity of mobility patterns results in complex geographic unit relations, diverse spatial organizational structures, and the long-tailed effect on OD flow distribution. This study proposes a novel OD flow prediction method based on graph-based deep learning (named as HMCG-LGBM). Specifically, 1) a modularity-based graph reconstruction strategy is presented for geographic unit relation augmentation by eliminating weak connections; 2) the heterogeneous spatial organization of OD flows is captured by combining the community detection approach and graph attention mechanism with the introduction of socio-economic and spatial features; and 3) a weighted loss function with distribution smoothing paradigm is developed to enhance the prediction for low-probability mobility events, addressing the challenges posed by long-tailed distributions. Extensive experiments conducted on real-world datasets show that the predictive performance of the proposed method is significantly improved, with the RMSE and MAE reduced from the baselines by 11.1%-33.3% and 14.1%-22.2%, respectively. The results also demonstrate the robustness of the proposed method for dealing with imbalanced OD flow distributions, providing valuable insights for spatial interaction predictive modeling in the context of sustainable urban systems.

Spatiotemporal prediction is one attractive research topic in urban computing, which is of great significance to urban planning and management. At present, there are many attempts to predict the spatiotemporal state of systems using various deep learning models. However, most existing models tend to improve prediction accuracy with larger parameter scale and time consumption, but ignoring ease of use in practice. To overcome this question, we propose a lightweight spatiotemporal graph dilated convolutional network called STGDN with satisfactory prediction accuracy and lower model complexity. More specifically, we propose a novel dilated convolution operator and integrate it into traditional causal convolutional networks and graph convolutional networks to greatly improve the efficiency of prediction. The proposed dilated convolution operator can significantly reduce the depth of the model, thereby reducing the parameter scale and improving the computational efficiency of the model. We conducted on multi experiments on three real-world spatiotemporal datasets (traffic dataset, PM2.5 dataset, and temperature dataset) to prove the effectiveness and advantage of our proposed STGDN. The experimental results show that the proposed STGDN model outperforms or achieves comparable prediction accuracy of the existing nine baselines with higher operational efficiency and fewer model parameters. Codes are available at anonymous private link on https://doi.org/10.6084/m9.figshare.23935683.

Forecasting citywide traffic congestion on large road networks has long been a nontrivial research problem due to the challenge of modeling complex evolution patterns of congestion in highly stochastic traffic environments. Arguing that purely data‐driven methods may not perform well for congestion forecasting, we propose a deep marked graph process model for predicting the congestion indices and the occurrence time of traffic congestion events for complex signalized road networks. Traffic congestion is considered as a nonrigorous spatiotemporal extreme event. We extend the traditional point process model by integrating a specially designed spatiotemporal graph convolutional network. This hybrid strategy takes advantage of the simple form of the point process model as well as the ability of graph neural networks to emulate the evolution of congestion. Experiments on real‐world congestion data sets show that the proposed method outperforms state‐of‐the‐art baseline methods, yielding satisfactory prediction results on a large signalized road network with superior computational efficiency.

Freeway traffic volume is strongly correlated with the intensity of regional socioeconomic spatial interactions and the road network structure. Although existing studies have proposed indicators of betweenness centrality (BC) integrated into regional spatial interactions, the socio-economic drivers of freeway traffic volume formation have been neglected. More importantly, existing studies have not established a non-linear response relationship among BC, city socio-economic spatial interactions, and road traffic volume, which severely limits the comprehensive quantification of the role of freeway traffic flow drivers. Therefore, this study proposes a freeway traffic volume inference method that integrates spatial interaction to enhance BC. First, the socioeconomic factors of the origin and destination cities are incorporated into the BC indicator to create an enhanced betweenness centrality indicator (ODBC), which quantifies the strength of spatial interactions between cities. Second, a machine learning approach is used to develop the non-linear response relationship between ODBC and freeway traffic flow to accurately infer traffic volume. Finally, utilizing the SHapley additive explanation approach, the role vectors of intercity freeway traffic volume drivers are quantified. Experiments conducted on data from freeway toll stations demonstrate that the proposed method surpasses the baseline method based on BC and weighted by BC considering only the potential destination or origin city attractiveness, with an improvement in R2 of 14%, 4.2%, and 4%, and a maximum reduction in RMSE of 40%, 24.5%, and 26%. The proposed method yields higher accuracy for unknown road segments and is easily interpretable.

Explainable spatio-temporal prediction gains attraction in the development of geospatial artificial intelligence. The neural ordinal differential equation (NODE) emerges as a new solution for explainable spatio-temporal prediction. However, challenges still need to be solved in most existing NODE-based prediction models, such as difficulty modeling spatial data and mining long-term temporal dependencies in data. In this study, we propose a spatio-temporal attentional NODE (STA-ODE) to address the two challenges above. First, we define a spatio-temporal ordinary differential equation to predict a value at each time iteratively by a novel spatio-temporal derivative network. Second, we develop an attention mechanism to fuse multiple prediction values for capturing long-term temporal dependencies in data. To train the STA-ODE model, we design a loss function that aligns the prediction results in spatial dimension with prediction results in temporal dimension to calibrate the parameters of the model. The proposed model was validated with three real-world spatio-temporal datasets (traffic flow dataset, PM2.5 monitoring dataset, and temperature monitoring dataset). Experimental results showed that STA-ODE outperformed seven existing baselines regarding prediction accuracy. In addition, we used visualization to demonstrate the sound interpretability and prediction accuracy of the STA-ODE model.

Accurate rainstorm forecasting is crucial for the sustainable development of human society. Recently, machine learning-based rainstorm prediction methods have shown promising results. However, these methods often fail to adequately consider the prior knowledge of rainstorms and do not explicitly account for the dynamic spatio-temporal patterns of rainstorm events. This study introduces a novel end-to-end prior-informed rainstorm forecasting model that incorporates both fundamental physical priors and the spatio-temporal development patterns of rainstorms. The model utilizes a gated convolutional encoder-decoder network to effectively represent the spatio-temporal patterns of rainstorm events. A key component of the representation network is the Substantial Derivative-GuIded gated convolutional Unit (SDGiU), which updates latent states under the constraints of physical priors. Additionally, an integrated loss function is designed to minimize reconstruction errors on multiple scales and facilitate the generation of forecasts that reproduce the actual spatio-temporal patterns of rainstorm formation, development and dissipation. Experimental results on two reanalysis datasets show that the proposed forecasting model outperforms competing state-of-the-art baselines by at least 19.7% (15.0%) in overall Critical Success Index (Heidke Skill Score). Qualitative analysis indicates that the proposed model can generate predictions that are both physically consistent and spatially-temporally coherent.

Spatiotemporal prediction is a research topic in urban planning and management. Most existing spatiotemporal prediction models currently face challenges. More specifically, most prediction models are sensitive to missing data, meaning most prediction models are only tested on spatiotemporal data assuming no missing data. Although missing data can be imputed, spatiotemporal prediction models with the capability of handling missing data are needed. In this study, we propose a novel missing-data-tolerant causal graph attention model called CGATM to address the above challenges. To enable the CGATM model to be tested on spatiotemporal data with missing data, we propose a novel missing data handling mechanism that automatically handles missing data according to the probability of data missing patterns. To improve the nonlinear fitting ability of the CGATM model, we propose a novel causal graph attention method that represents geospatial heterogeneity by adjacent nodes with different weights. In addition, we design the CGTAM model as an Imputer-Predictor architecture and define a novel loss function to optimize model parameters. The proposed model was validated on three real-world spatiotemporal datasets (traffic dataset, PM2.5 dataset, and temperature dataset). Experimental results showed that the proposed model has better prediction performance under four missing scenarios, and outperforms eight existing baselines regarding prediction accuracy.