BACKGROUND:Major depressive disorder (MDD) ranks among the foremost contributors to disability worldwide, yet its neurophysiological mechanisms remain poorly understood. Excitation-inhibition (E/I) imbalance has been implicated in MDD pathophysiology, but cortex-wide E/I ratio and its molecular substrates in MDD remain unknown. METHODS:Resting-state functional magnetic resonance imaging data from 254 MDD patients and 451 healthy controls (HCs) across six sites were analyzed. The Hurst exponent, a biophysically confirmed proxy of E/I balance, was estimated using a fractionally integrated process framework. Neurobiological decoding analyses were performed to map the transcriptomic and neurochemical signatures of cortical E/I imbalance in MDD. An independent ketamine clinical trial dataset (32 treatment-resistant depression patients and 21 HCs) was used to examine ketamine-induced changes in cortical E/I balance. RESULTS:Patients with MDD demonstrated significantly reduced Hurst exponent values, predominantly encompassing the parietal and prefrontal-cingulate cortices. Transcriptomic analysis identified enrichment for neuronal structural organization, nucleic acid metabolism, and mitochondrial function, with preferential overlap with excitatory and inhibitory neuron-specific gene sets. Neurochemically, Hurst exponent alterations were spatially associated with GABAergic, opioidergic, serotonergic, and synaptic density distributions. Divergent group-by-treatment effects were observed in the anterior cingulate and medial prefrontal cortices, with ketamine-induced increases in TRD patients. CONCLUSION:These findings highlight that prefrontal-cingulate E/I imbalance, anchored to specific transcriptional and neurochemical substrates, may underlie the pathophysiology of MDD and the antidepressant effects of ketamine. The Hurst exponent offers a promising neuroimaging approach for probing E/I imbalance and identifying potential treatment targets in depression.

OBJECTIVE:Tinnitus and chronic pain, as two perceptual disorders frequently accompanied by negative emotions, significantly impair patients' daily functioning and overall well-being. However, the neurophysiological mechanisms that account for their frequently observed clinical similarities have yet to be fully elucidated. METHODS:The study enrolled a total of 56 participants, including 18 tinnitus patients and 19 chronic pain patients, alongside 19 healthy controls. We computed the fractional amplitude of low-frequency fluctuations (fALFF) and regional homogeneity (ReHo) from resting-state fMRI data to quantify localized brain activity intensity and neural synchronization. RESULTS:Statistical analysis suggested a significant elevation in fALFF within the calcarine cortex of tinnitus patients compared to healthy controls (p < 0.05, FDR-corrected). In contrast, chronic pain patients exhibited notable increases in both fALFF and ReHo in the inferior frontal gyrus (IFG). Direct comparisons between patient groups suggested that tinnitus patients had higher fALFF in the inferior temporal gyrus (ITG), supplementary motor area and fusiform gyrus, but lower fALFF in the right IFG, precuneus, and caudate nucleus. Regarding ReHo, patients with tinnitus exhibited significantly elevated ReHo values in the supplementary motor area, while demonstrating reduced ReHo in both the precuneus and caudate nucleus (all p < 0.05, FDR-corrected). CONCLUSION:The observed differences between tinnitus and chronic pain in brain regions including the prefrontal cortex, ITG, and fusiform gyrus provide neuroimaging correlates associated with the two conditions. These findings may inform future neuromodulation targets and support clinical differential diagnosis.

Oral squamous cell carcinoma (OSCC) remains a significant clinical challenge due to frequent recurrence, metastasis, and therapeutic resistance. Here, we establish a living biobank of OSCC patient-derived organoids (PDOs) comprising 46 lines using optimized culture medium. These PDOs are long-term passaged, cryopreserved, and recovered with stable viability and tumorigenicity. Comprehensive morphological, genomic, and transcriptomic analyses confirm that PDOs faithfully recapitulate the histopathological, genetic, and molecular features of parental tumors. These PDOs enable disease modeling, genetic manipulation, and drug screening. Through transcriptomic profiling and functional assays, we find that CDCP1 mediates cisplatin resistance by modulating Wnt/β-catenin signaling-driven stemness. Notably, we develop a pH-sensitive nanoparticle delivering siCDCP1, which effectively restores chemosensitivity and impairs tumor growth in cisplatin-resistant patient-derived xenograft (PDX) models with favorable safety profile. These findings establish PDOs as robust preclinical models for mechanistic explorations and therapeutics development and highlight CDCP1-targeting strategies as promising approaches to overcome cisplatin resistance in OSCC.

Background Protein kinases dysregulation is implicated in various cancer-related processes; however, its clinical utility and biological significance in head and neck squamous cell carcinoma (HNSCC) remain incompletely understood. This study aimed to establish a novel prognostic signature using kinase-related genes (KRGs) for prognostic and therapeutic prediction in HNSCC. Methods A KRG-related predictive signature was constructed using LASSO-Cox regression. A nomogram incorporating this signature and selected clinicopathological variables was developed through multivariate Cox regression. The predictive value of KRG signature for immune status and chemotherapeutic or immunotherapeutic responses was further evaluated. Finally, both in vitro and in vivo experiments were performed to validate the oncogenic role of ABL2 in HNSCC. Results Patients were stratified into high- and low-risk groups based on a seven-gene KRG signature (TRIB3, RPS6KA4, ABL2, FJX1, JAK3, PIM2, and NEK6), with high-risk patients showing significantly worse prognosis. Additionally, the high-risk group was characterized by an immunosuppressive status, reduced predicted immunotherapeutic response, and enhanced chemoresistance. In vitro and in vivo experimental validation further demonstrated that ABL2 knockdown significantly suppressed tumor proliferation, migration and invasion, while alleviating chemoresistance and reversing the immunosuppressive phenotype. Conclusion This study demonstrates that the KRG-derived signature and corresponding nomogram serves as a robust prognostic biomarker and a reliable predictor of therapeutic response in HNSCC.

Diabetes mellitus (DM) is associated with systemic metabolic disturbances across multiple organs. Total-body 18F-fluorodeoxyglucose (F-FDG) PET/CT enables simultaneous quantification of glucose metabolism in numerous organs. This study aimed to characterize multi-organ 18F-FDG uptake patterns in type 2 DM patients compared with healthy controls and to explore associations with clinical variables including brain volume. Compared with controls, DM patients exhibited significantly lower SULmean in brain (− 15.3

RATIONALE AND OBJECTIVES:To investigate the value of time-dependent diffusion MRI (Td-dMRI) and macromolecular proton fraction (MPF) imaging in assessing the pathological grade of cervical cancer (CC). MATERIALS AND METHODS:A total of 92 CC patients, comprising 35 with high-grade (grade III) cancer and 57 with low-grade (grade I/II) cancer, who underwent Td-dMRI and MPF, were prospectively enrolled. Td-dMRI derived parameters including cellularity (cell density), diameter (tumor cell size), Dex (extracellular diffusivity), Vin (intracellular volume fraction), and three apparent diffusion coefficients (ADCPGSE, ADC17 Hz, ADC33Hz) and MPF derived parameter MPF (tissue macromolecular) were calculated and compared. Diagnostic performance was assessed via area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and decision curve analysis (DCA); internal validation was performed using 1000 bootstrap resamples to mitigate model optimism. Multiple pairwise AUC comparisons were adjusted using the DeLong test. RESULTS:Cellularity, Vin, and MPF were higher and diameter, Dex, ADCPGSE, ADC17 Hz, and ADC33 Hz were lower in high-grade group than in low-grade group (all P < 0.05). Cellularity, Vin, and MPF were independent predictors and their combination achieved optimal diagnostic efficacy (AUC = 0.960; 95% CI: 0.897-0.990; sensitivity = 97.14%; specificity = 82.46%), which was significantly higher than any individual parameter (AUC range: 0.685-0.923; all P < 0.05 after Bonferroni correction). Internal validation confirmed stable performance (AUC = 0.955; 95% CI: 0.939-0.960), and DCA demonstrated higher net benefit for patients. Vin strongly correlated with pathological nuclear fraction (r = 0.752, P < 0.001). CONCLUSION:Td-dMRI and MPF were effective methods of predicting pathological grade in CC, and the combination of cellularity, Vin, and MPF has the potential to serve as a new imaging marker, facilitating preoperative grading and personalized treatment.

BACKGROUND:Dynamic positron emission tomography (PET) is a powerful tool for clinical tumor diagnosis. However, the conventional dynamic scanning duration takes about 60 m i n $min$ , which is inconvenient for patients and limits the widespread application of this technology. PURPOSE:This study aims to develop an innovative method to achieve shortened high-quality K i $K_i$ parametric imaging based on the Patlak model. METHODS:We proposed a population-based input function integral estimation (PBIF-IE) method. The core of this method is to construct a linear regression model between the early-stage integral ( S e a r l y $S_{early}$ ) and the late-stage mean ( M l a t e $M_{late}$ ) of the image-derived input functions (IDIFs) in the training dataset. The goal is to estimate S e a r l y $S_{early}$ in the testing datasets using M l a t e $M_{late}$ and the linear regression model when only late-stage dynamic sequences are available. To verify the effectiveness and stability of the model, we set up three testing datasets, A, B, and C, each with different framing protocols for K i $K_i$ parametric imaging analysis. Furthermore, to explore the potential of our proposed method in shortening scan duration, we evaluated the K i $K_i$ parameter results obtained by our proposed method under three scan durations (30, 20, and 10 m i n $min$ ) using multiple quantitative metrics, including the peak signal-to-noise ratio (PSNR), the structural similarity index (SSIM), and the relative error (RE). RESULTS:Through multi-center data studies, we demonstrate the effectiveness of the PBIF-IE method. Extensive experimental results demonstrate that the PBIF-IE method outperforms other methods in both S e a r l y $S_{early}$ estimation and K i $K_i$ parametric imaging. The 30 m i n $min$ dynamic scanning protocol can obtain K i $K_i$ parameter images that are highly consistent with the 60 m i n $min$ scanning protocol, while the 20 m i n $min$ dynamic scanning protocol is sufficient for preliminary tumor localization. CONCLUSIONS:Based on the experimental results, the PBIF-IE method outperforms other existing methods for shortened K i $K_i$ parametric imaging. In future research, we plan to explore how the number of dynamic sequences used in the training dataset affects the model construction. This will help us further optimize the parametric imaging process.

Parkinson's disease (PD) is an incurable neurological disorder that often begins insidiously with sleep disturbances and somatic symptoms, progressing to whole-body motor and cognitive symptoms1-5. Dysfunction of the somato-cognitive action network (SCAN)-which is thought to control action execution6,7 by coordinating arousal, organ physiology and whole-body motor plans with behavioural motivation-is a potential contributor to the diverse clinical manifestations of PD. To investigate the role of the SCAN in PD pathophysiology and treatments (medications, deep-brain stimulation (DBS), transcranial magnetic stimulation (TMS) and MRI-guided focused ultrasound stimulation (MRgFUS)), we built a large (n = 863), multimodal, multi-intervention clinical imaging dataset. Resting-state functional connectivity revealed that the substantia nigra and all PD DBS targets (subthalamic nucleus, globus pallidus and ventral intermediate thalamus) are selectively connected to the SCAN rather than to effector-specific motor regions. Importantly, PD was characterized by specific hyperconnectivity between the SCAN and the subcortex. We therefore followed six PD cohorts undergoing DBS, TMS, MRgFUS and levodopa therapy using precision resting-state functional connectivity and electrocorticography recording. Efficacious treatments reduced SCAN-to-subcortex hyperconnectivity. Targeting the SCAN instead of effector regions doubled the efficacy of TMS treatments. Focused ultrasound treatment benefits increased when the target was closer to the thalamic SCAN sweet spot. Thus, SCAN hyperconnectivity is central to PD pathophysiology and its alleviation is a hallmark of successful neuromodulation. Targeting functionally defined subcortical SCAN nodes may improve existing therapies (DBS, MRgFUS), whereas cortical SCAN targets offer effective non-invasive or minimally invasive neuromodulation for PD.

Cross-modal brain MRI translation bridges structural and functional information across imaging modalities, but existing approaches like GANs and diffusion models face trade-offs between synthesis quality and computational efficiency. In the work, we propose Short-Flow, the first flow matching-based framework for efficient medical image translation, which formulates the task as deterministic distribution transport via an ordinary differential equation, wherein a time-dependent velocity field directly mapping source to target modality. We further introduce a two-time flow map mechanism that encodes the average velocity between arbitrary temporal states, enabling one-step inference without retraining while retaining the option of multi-step refinement. The inherent invertibility of the flow yields a single unified network for symmetric forward/reverse translation, eliminating the need for dual-model redundancy. Extensive validations on IXI and UPenn-GBM datasets shows that Short-Flow achieves superior synthesis quality while delivering orders-of-magnitude faster inference compared to prior diffusion models, making it suitable for real-time clinical deployment.

Spatiotemporally controlled reactive oxygen species (ROS) generation remains pivotal for advancing photodynamic immunotherapy. Herein, we present tumor microenvironment (TME)-responsive nano-immunoregulator featuring a CaCO3 shell encapsulating a mesoporous silica core modified with folic acid/chlorin e6 (FA/Ce6) and co-loaded with dBET6 and maleimide (MA). Acidic TME triggers CaCO3 shell degradation, neutralizing tumor acidity to repolarize macrophages toward M1 type, enhancing antigen presentation and T cells immune response while exposing the BM@MFC core. The resulting BM@MFC further endocytosed by tumor cells via FA-mediated tumor targeting, followed by MA-mediated glutathione depletion amplifies Ce6-generated ROS for potent photodynamic therapy (PDT). Concurrently, dBET6 degrades BRD4 to inhibit metastasis, downregulate PD-L1, and synergize with PDT to induce immunogenic cell apoptosis, thereby activating dendritic cells (DCs) and T cells for enhanced photodynamic immunotherapy. Both in vitro/vivo results indicate that BM@MFCC have excellent performance to inhibit primary/metastatic tumors, while single-cell RNA sequencing elucidates BM@MFCC can induce immunosuppressive TME remodeling, including increased CD8+ T cells, cDCs, and NK cells; M2-to-M1 macrophage polarization; and enhanced intercellular communication. Tumor cell profiling confirms BM@MFCC will downregulate oncogenic pathways and activate immune signatures. Collectively, this TME responsive nano-immunoregulator demonstrates superior therapeutic outcomes and provides a promising strategy for precision cancer photodynamic immunotherapy.

Supplementary Materials and Methods

Integrated positron emission tomography/magnetic resonance (PET/MRI) may have the potential to evaluate lymph node metastasis (LNM) status and Ki-67 proliferation index in patients with non-small cell lung cancer. This study enrolled 92 pathologically confirmed NSCLC patients. Quantitative analysis of metabolic parameters, such as maximum standardized uptake value (SUVmax), metabolic tumor volume (MTV), total lesion glycolysis (TLG), and IVIM parameters including apparent diffusion coefficient (ADC), true diffusion coefficient (D), pseudo diffusion coefficient (D*), perfusion fraction (f), and distributed diffusion coefficient (DDC). The predictive performance of each parameter for LNM was assessed using receiver operating characteristic (ROC) curve analysis, and a multivariate logistic regression model was constructed to establish the optimal combined predictive model. Spearman correlation analysis was used to explore the relationship between imaging parameters and Ki-67 expression. LNM prediction: The LNM-positive group exhibited lower ADC, D, and DDC (all P < 0.05) compared to the LNM-negative group. Multivariate analysis identified MTV and DDC as independent predictors of LNM. The combined model (MTV + DDC) achieved an AUC of 0.821 (sensitivity 79.49

Positron emission tomography (PET) provides functional information by capturing tracer uptake and is widely used for disease assessment. Accurate segmentation of regions of interest is essential for quantitative analysis and clinical decision-making. However, PET images often exhibit low spatial resolution, high noise, and blurred boundaries due to partial volume effects, which hampers precise delineation. To address this, we propose TDPSR-Net, a PET image segmentation network that integrates topographic distance priors (TDP) and spatial regularization techniques. Our method automatically generates marker points for computing topographic distances, and the network jointly extracts features from both PET images and the resulting TDP maps. To enhance spatial coherence and boundary consistency, we introduce a Soft Threshold Dynamics of Sigmoid (STD-Sigmoid) layer that imposes spatial regularization on the network output, and we further establish a theoretical connection between the proposed formulation and a Potts-type model. We evaluate TDPSR-Net on multiple datasets, including liver, hippocampus, and lung cancer tumor segmentation, and the results demonstrate consistently high accuracy and robustness across diverse datasets, highlighting the potential of TDPSR-Net for a wide range of clinical applications.

OBJECTIVE:This study aimed to construct a novel signature utilizing DNA damage response-related genes (DRGs) for head and neck squamous cell carcinoma (HNSCC) prognostication and therapeutic prediction. DESIGN:A prognostic signature specific to HNSCC was developed using univariate Cox regression, Kaplan-Meier survival analysis, and least absolute shrinkage and selection operator (LASSO)-penalized multivariate Cox regression analyses. A nomogram incorporating this signature along with selected clinicopathological factors was obtained through multivariate Cox regression. The effectiveness of this DRG signature in predicting immune status and chemotherapeutic or immunotherapeutic responses was evaluated. Gene function was assessed via a pharmacological approach in vitro. RESULTS:The 10-gene DRG signature/nomogram demonstrated well prognostic performance across various independent cohorts. Pharmacological inhibition of PLK1 from the DRG signature led to cell death and apoptosis which were probably resulted from impaired DNA damage response. Higher DRG signature scores were negatively correlated with the abundance of tumor-infiltrating immune cells and linked to several chemotherapeutic drugs, radiotherapy, and immunotherapy sensitivities. CONCLUSIONS:Our findings indicated that this novel DRG-derived signature/nomogram serves as a robust prognostic biomarker and viable therapeutic response predictor in HNSCC.

Pancreatic ductal adenocarcinoma (PDAC), one of the deadliest solid malignancies, is often detected at a late and inoperable stage. Retrospective reviews of prediagnostic CT scans, when conducted by expert radiologists aware that the patient later developed PDAC, frequently reveal lesions that were previously overlooked. To help detecting these lesions earlier, we developed an automated system named ePAI (early Pancreatic cancer detection with Artificial Intelligence). It was trained on data from 1,598 patients from a single medical center. In the internal test involving 1,009 patients, ePAI achieved an area under the receiver operating characteristic curve (AUC) of 0.939-0.999, a sensitivity of 95.3%, and a specificity of 98.7% for detecting small PDAC less than 2 cm in diameter, precisely localizing PDAC as small as 2 mm. In an external test involving 7,158 patients across 6 centers, ePAI achieved an AUC of 0.918-0.945, a sensitivity of 91.5%, and a specificity of 88.0%, precisely localizing PDAC as small as 5 mm. Importantly, ePAI detected PDACs on prediagnostic CT scans obtained 3 to 36 months before clinical diagnosis that had originally been overlooked by radiologists. It successfully detected and localized PDACs in 75 of 159 patients, with a median lead time of 347 days before clinical diagnosis. Our multi-reader study showed that ePAI significantly outperformed 30 board-certified radiologists by 50.3% (P < 0.05) in sensitivity while maintaining a comparable specificity of 95.4% in detecting PDACs early and prediagnostic. These findings suggest its potential of ePAI as an assistive tool to improve early detection of pancreatic cancer.

TOF-MRA intracranial vessel segmentation is critical in clinical practice but challenged by limited annotations and significant cross-modality domain shifts. To address these issues, this study proposes DGRTA, an unsupervised domain adaptation framework that integrates cross-modality pseudo-label generation, a dual-gated pseudo-label refinement strategy (DGR), and a topology aware weighting mechanism (TA). Initially, rigid and non-rigid registration are used to transfer CTA predictions to TOF-MRA to generate initial pseudo-labels. DGR then refines these labels using prediction probabilities and image intensity, enhancing sensitivity and specificity, while TA leverages Persistence Diagrams (PD) to quantify topological discrepancies and dynamically adjust loss weights. Experiments on 185 paired CTA/TOF-MRA cases demonstrated that DGRTA consistently improved performance across four backbone architectures (UNet, Attention UNet, UNETR, Swin UNETR). The Attention UNet DGRTA achieved the best results, with a Dice of 0.810, clDice of 0.800, and an AHD of 0.353 mm on the validation set, significantly outperforming the baseline model (p < 0.001). DGRTA offers a feasible solution that reduces reliance on extensive manual annotations, underscoring the potential of unsupervised cross-modality segmentation in various vascular imaging applications.

ABSTRACT Purpose To develop SAFE, a self‐calibrated framework to estimate B 1 + and B 0 field inhomogeneities directly from conventional magnetic resonance fingerprinting (MRF) acquisitions and improve its quantification accuracy. Methods SAFE utilized a two‐step approach. First, two physics‐informed image markers are extracted from the MRF data to create a magnitude and a phase image that are highly correlated with B 1 + and B 0 inhomogeneities, respectively. Second, a deep learning (DL) network is applied to map these markers to quantitative field maps. SAFE was tested on 3D Spiral Projection Imaging MRF brain acquisition at 3T, where the network was trained and validated across a multi‐site, multi‐vendor dataset ( N = 358) from healthy volunteers and clinical population. The capability of SAFE to be applied to unseen MRF sequences with different signal preparations and flip‐angle trains through adaptively retraining without the need for additional training data was also tested. Results SAFE achieved normalized‐root‐mean‐square‐error within 3% against gold‐standard field‐calibration scans on both B 1 + and B 0 maps ( N = 32), with high performance remaining on datasets from scanners that the training data were acquired from. T1 and T2 biases were shown to be effectively corrected on healthy volunteers, a patient with brain tumor, and a pediatric subject. Tissue quantification accuracy after SAFE‐estimated field correction was validated on a large patient cohort ( N = 86). SAFE was also demonstrated to be adaptable to different MRF sequences without acquiring additional training data. Conclusion By combining physics‐informed image markers with DL, the proposed SAFE framework enables calibration‐free estimation of B 1 + and B 0 field inhomogeneities at 3T for whole brain MRF.

Lung cancer (LC) is increasingly recognized as a systemic disease, and these systemic alterations may differ across histological subtypes. Smoking is a major risk factor for LC, yet its impact on systemic metabolism across LC subtypes remains underexplored. We studied 125 LC patients and 200 healthy controls from three centers who underwent static and dynamic total-body 18F-FDG PET/CT. Thirty-seven regions of interest spanning the central nervous system and peripheral organs were analyzed. Regional standardized uptake values normalized by lean body mass and time-activity curves were harmonized using ComBat and adjusted for demographic covariates. Population-level metabolic networks were constructed from bootstrapped Pearson correlations of regional uptake to characterize smoking status-associated systemic alterations across LC subtypes. Individual-level metabolic networks were derived from correlations of residual time-activity curves after third-order polynomial fitting, from which organ-specific abnormality strength metrics were calculated. In addition, tumor habitats were defined from PET/CT intensity and entropy features, and their associations with smoking status and organ-level abnormalities were assessed using multivariable regression analyses. Population- and individual-level network analyses showed that smoking status was associated with altered systemic metabolic coordination in a subtype-dependent manner, with the liver consistently emerging as a central hub of metabolic disruption. Tumor habitat analysis further revealed a smoking-associated shift toward hypometabolic tumor microenvironments. These findings provide quantitative insight into smoking status-related systemic metabolic reprogramming in LC and highlight the potential value of imaging-based characterization across histological subtypes.

Accurate diagnosis of brain disorders (BDs) is challenging in clinical practice. Most existing deep learning-based methods perform diagnosis only in a one-step manner, ignoring the step-wise, multi-level diagnosis processes as performed by radiologists. This oversight often leads to a high risk of misdiagnosis, especially for long-tail or challenging BDs. In this work, we introduce a Hierarchical Prompt and Prototype Learning (HP2L) framework for BD diagnosis, which emulates multi-level diagnostic procedures. HP2L explicitly captures hierarchical relationships among 23 BDs and groups them into three diagnostic levels: coarse classes (e.g., vascular lesions), intermediate classes (e.g., hemorrhage), and fine-grained classes (e.g., chronic hemorrhage). HP2L integrates three key innovations: (1) Hierarchical Prompting Vision Transformer (ViT) backbone, which performs coarse-to-fine feature extraction for step-wise BD classification; (2) Prompt Learning, which employs optimizable prompt tokens that encode diagnostic knowledge, guiding the classification at each level of the hierarchy; (3) Prototype Learning, which enriches the prompt token with BD-specific prototypes by injecting diagnostic information to enhance diagnosis performance. Extensive evaluations on 54,360 subjects across six multi-center datasets show that HP2L consistently outperforms state-of-the-art methods, achieving a balanced accuracy of 88.43% for both common and long-tail BDs, 8.42 percentage points higher than the best-performing benchmark. Furthermore, HP2L improves interpretability by aligning its predictions and attention visualizations with the clinical hierarchical reasoning process. The code and a portion of data (more data will be released after the decision of the paper) are available under: code, data.