Efficacy and safety of olverembatinib in adult BCR::ABL1-positive ALL with T315I mutation or relapsed/refractory disease.

Third-generation tyrosine kinase inhibitors (TKIs) have much potential for the treatment of BCR::ABL1-positive leukaemia, particularly that harbouring the ABL1 T315I mutation. Olverembatinib (HQP1351), a novel third-generation TKI, has favourable efficacy and safety profiles in chronic myeloid leukaemia. Here, we present the clinical findings from 31 BCR::ABL1-positive acute lymphoblastic leukaemia (ALL) patients who received olverembatinib. Among the 14 patients with overt relapsed/refractory (R/R) disease (including 10 with the T315I mutation), 71.4% achieved an overall response. Of the other 17 patients with minimal residual disease (MRD)-positive ALL (including 14 with the T315I mutation), 60.0% and 47.1% achieved MRD flow negativity and complete molecular remission, respectively. With a median follow-up time of 16.3 months, the median event-free survival and overall survival were 3.9 and 8.3 months respectively, in overt R/R patients, and 11.5 and 18.4 months in MRD-positive patients. Allogeneic haematopoietic stem cell transplantation further improved outcomes among responders. The safety profile was generally manageable. This study suggests that olverembatinib-based therapy is another promising option for BCR::ABL1-positive ALL in addition to ponatinib, especially for patients with MRD-positive disease and a single T315I mutation.

Magnesium Nanocomposite Hydrogel Reverses the Pathologies to Enhance Mandible Regeneration.

The healing of bone defects after debridement in medication-related osteonecrosis of the jaw (MRONJ) is a challenging medical condition with impaired angiogenesis, susceptible infection, and pro-inflammatory responses. Magnesium (Mg) nanocomposite hydrogel is developed to specifically tackle multiple factors involved in MRONJ. Mg-oxide nanoparticles tune the gelation kinetics in the reaction between N-hydroxysuccinimide-functionalized hyperbranched poly (ethylene glycol) and proteins. This reaction allows an enhanced mechanical property after instant solidification and, more importantly, also stable gelation in challenging environments such as wet and hemorrhagic conditions. The synthesized hydrogel guides mandible regeneration in MRONJ rats by triggering the formation of type H vessels, activating Osterix+ osteoprogenitor cells, and generating anti-inflammatory microenvironments. Additionally, this approach demonstrates its ability to suppress infection by inhibiting specific pathogens while strengthening stress tolerance in the affected alveolar bone. Furthermore, the enhanced osteogenic properties and feasibility of implantation of the hydrogel are validated in mandible defect and iliac crest defect created in minipigs, respectively. Collectively, this study offers an injectable and innovative bone substitute to enhance mandible defect healing by tackling multiple detrimental pathologies.

Polydopamine-Cloaked Nanoarchitectonics of Prussian Blue Nanoparticles Promote Functional Recovery in Neonatal and Adult Ischemic Stroke Models.

Ischemic stroke is a devastating disease and one of the leading causes of mortality worldwide. Overproduction of reactive oxygen species and inflammatory response contribute to secondary damage following ischemic insult. Nanozymes with robust anti-oxidative stress properties possess therapeutic possibility for ischemic insult. However, insufficiency of nanozyme accumulation in the neuronal mitochondria hindered their application. Herein, we constructed polydopamine-coated Prussian blue nanoparticles (PB@PDA NPs) to realize the targeting neuronal mitochondria for ischemic stroke, with the properties of antioxidant and anti-inflammation. After administration, much higher accumulation of PB@PDA NPs in the brain was observed compared to that in the PB NP group. Moreover, PB@PDA NPs effectively attenuated brain infarct than that of PB NPs in neonatal mice following hypoxia-ischemia (HI) insult. PB@PDA NPs mainly colocated with neuronal mitochondria in vivo and in vitro. Apart from attenuating oxidative stress, PB@PDA NPs also suppressed neuronal apoptosis and counteracted inflammation, which effectively promote a short- and long-term functional recovery in HI mice. Further, the therapeutic efficacy of PB@PDA NPs was also found in adult ischemic mice via tail vein injection. Collectively, these findings illustrate that PB@PDA NPs via system injection accumulate in neuronal mitochondria and are beneficial for ischemic stroke.

Paclobutrazol Enhanced Stem Lodging Resistance of Direct-Seeded Rice by Affecting Basal Internode Development.

The objectives of this study were to explore the mechanism of stem mechanical strength in direct-seeded rice (DSR) as affected by paclobutrazol, especially its related endogenous hormone and cell wall component changes in culm tissue and response to the application of paclobutrazol. Field experiments were conducted in Changchun County, Jilin Province, China, by using two japonica rice varieties, Jiyujing and Jijing305, with soaking seeds in paclobutrazol at concentrations of (0 mg L-1, S0; 50 mg L-1; S1; 100 mg L-1; S2; 150 mg L-1, S3; 200 mg L-1, S4) in 2021 and 2022. The results suggest that the application of paclobutrazol increased the grain yield and reduced the lodging rate of DSR. Compared with the S0 treatments, soaking the seeds in paclobutrazol treatments rapidly shortened the length of the basal internode by decreasing the endogenous indole acetic acid (IAA) and gibberellin A3 (GA3) contents in culm tissue. The larger breaking strength (M) was attributed to a higher section modulus (SM) and bending stress (BS). The higher mechanical tissue thickness in culm tissue under paclobutrazol treatments, which was raised by higher endogenous zeatin and zeatin riboside (Z+ZR) content in culm tissue, increased the culm diameter, culm wall thickness, and section modulus (SM) of the internode. Compared with the S0 treatments, soaking the seeds in paclobutrazol treatments increased the cellulose content, lignin content, activities of lignin-related enzymes, and expression of key genes in lignin biosynthesis, as well as resulted in a higher bending stress (BS) to enhance the culm breaking strength (M).

Hexokinase HK3-mediated O-GlcNAcylation of EP300: a key regulator of PD-L1 expression and immune evasion in ccRCC.

Clear cell renal cell carcinoma (ccRCC) demonstrates enhanced glycolysis, critically contributing to tumor development. Programmed death-ligand 1 (PD-L1) aids tumor cells in evading T-cell-mediated immune surveillance. Yet, the specific mechanism by which glycolysis influences PD-L1 expression in ccRCC is not fully understood. Our research identified that the glycolysis-related gene (GRG) HK3 has a unique correlation with PD-L1 expression. HK3 has been identified as a key regulator of O-GlcNAcylation in ccRCC. O-GlcNAcylation exists on the serine 900 (Ser900) site of EP300 and can enhance its stability and oncogenic activity by preventing ubiquitination. Stably expressed EP300 works together with TFAP2A as a co-transcription factor to promote PD-L1 transcription and as an acetyltransferase to stabilize PD-L1 protein. Furthermore, ccRCC exhibits interactive dynamics with tumor-associated macrophages (TAMs). The uridine 5'-diphospho-N-acetylglucosamine (UDP-GlcNAc), which serves as a critical substrate for the O-GlcNAcylation process, facilitates TAMs polarization. In ccRCC cells, HK3 expression is influenced by IL-10 secreted by M2 TAMs. Our study elucidates that HK3-mediated O-GlcNAcylation of EP300 is involved in tumor immune evasion. This finding suggests potential strategies to enhance the efficacy of immune checkpoint blockade therapy.

Programmed cell death factor 4-mediated hippocampal synaptic plasticity is involved in early life stress and susceptibility to depression.

Early life stress (ELS) increases the risk of depression later in life. Programmed cell death factor 4 (PDCD4), an apoptosis-related molecule, extensively participates in tumorigenesis and inflammatory diseases. However, its involvement in a person's susceptibility to ELS-related depression is unknown. To examine the effects and underlying mechanisms of PDCD4 on ELS vulnerability, we used a "two-hit" stress mouse model: an intraperitoneal injection of lipopolysaccharide (LPS) into neonatal mice was performed on postnatal days 7-9 (P7-P9) and inescapable foot shock (IFS) administration in adolescent was used as a later-life challenge. Our study shows that compared with mice that were only exposed to the LPS or IFS, the "two-hit" stress mice developed more severe depression/anxiety-like behaviors and social disability. We detected the levels of PDCD4 in the hippocampus of adolescent mice and found that they were significantly increased in "two-hit" stress mice. The results of immunohistochemical staining and Sholl analysis showed that the number of microglia in the hippocampus of "two-hit" stress mice significantly increased, with morphological changes, shortened branches, and decreased numbers. However, knocking down PDCD4 can prevent the number and morphological changes of microglia induced by ELS. In addition, we confirmed through the Golgi staining and immunohistochemical staining results that knocking down PDCD4 can ameliorate ELS-induced synaptic plasticity damage. Mechanically, the knockdown of PDCD4 exerts neuroprotective effects, possibly via the mediation of BDNF/AKT/CREB signaling. Combined, these results suggest that PDCD4 may play an important role in the ELS-induced susceptibility to depression and, thus, may become a therapeutic target for depressive disorders.

Canonical pathways for validating steroid-associated osteonecrosis in mice.

The present study aimed to establish and evaluate a preclinical model of steroid-associated osteonecrosis (SAON) in mice. Sixteen 24-week-old male C57BL/6 mice were used to establish SAON by two intraperitoneal injections of lipopolysaccharide (LPS), followed by three subcutaneous injections of methylprednisolone (MPS). Each injection was conducted on working day, with an interval of 24 h. Six cycles of injections were conducted. Additional twelve mice (age- and gender-matched) were used as normal controls. At 2 and 6 weeks after completing induction, bilateral femora and bilateral tibiae were collected for histological examination, micro-CT scanning, and bulk RNA sequencing. All mice were alive until sacrificed at the indicated time points. The typical SAON lesion was identified by histological evaluation at week 2 and week 6 with increased lacunae and TUNEL+ osteocytes. Micro-CT showed significant bone degeneration at week 6 in SAON model. Histology and histomorphometry showed significantly lower Runx2+ area, mineralizing surface (MS/BS), mineral apposition rate (MAR), bone formation rate (BFR/BS), type H vessels, Ki67+ (proliferating) cells, and higher marrow fat fraction, osteoclast number and TNFα+ areas in SAON group. Bulk RNA-seq revealed changed canonical signaling pathways regulating cell cycle, angiogenesis, osteogenesis, and osteoclastogenesis in the SAON group. The present study successfully established SAON in mice with a combination treatment of LPS and MPS, which could be considered a reliable and reproducible animal model to study the pathophysiology and molecular mechanism of early-stage SAON and to develop potential therapeutic approaches for the prevention and treatment of SAON.

PD-1 downregulation enhances CAR-T cell antitumor efficiency by preserving a cell memory phenotype and reducing exhaustion.

BACKGROUND: Despite the encouraging outcome of chimeric antigen receptor T cell (CAR-T) targeting B cell maturation antigen (BCMA) in managing relapsed or refractory multiple myeloma (RRMM) patients, the therapeutic side effects and dysfunctions of CAR-T cells have limited the efficacy and clinical application of this promising approach. METHODS: In this study, we incorporated a short hairpin RNA cassette targeting PD-1 into a BCMA-CAR with an OX-40 costimulatory domain. The transduced PD-1KD BCMA CAR-T cells were evaluated for surface CAR expression, T-cell proliferation, cytotoxicity, cytokine production, and subsets when they were exposed to a single or repetitive antigen stimulation. Safety and efficacy were initially observed in a phase I clinical trial for RRMM patients. RESULTS: Compared with parental BCMA CAR-T cells, PD-1KD BCMA CAR-T cell therapy showed reduced T-cell exhaustion and increased percentage of memory T cells in vitro. Better antitumor activity in vivo was also observed in PD-1KD BCMA CAR-T group. In the phase I clinical trial of the CAR-T cell therapy for seven RRMM patients, safety and efficacy were initially observed in all seven patients, including four patients (4/7, 57.1%) with at least one extramedullary site and four patients (4/7, 57.1%) with high-risk cytogenetics. The overall response rate was 85.7% (6/7). Four patients had a stringent complete response (sCR), one patient had a CR, one patient had a partial response, and one patient had stable disease. Safety profile was also observed in these patients, with an incidence of manageable mild to moderate cytokine release syndrome and without the occurrence of neurological toxicity. CONCLUSIONS: Our study demonstrates a design concept of CAR-T cells independent of antigen specificity and provides an alternative approach for improving the efficacy of CAR-T cell therapy.

Downregulation of UBB potentiates SP1/VEGFA-dependent angiogenesis in clear cell renal cell carcinoma.

Clear cell renal cell carcinoma (ccRCC) presents a unique profile characterized by high levels of angiogenesis and robust vascularization. Understanding the underlying mechanisms driving this heterogeneity is essential for developing effective therapeutic strategies. This study revealed that ubiquitin B (UBB) is downregulated in ccRCC, which adversely affects the survival of ccRCC patients. UBB exerts regulatory control over vascular endothelial growth factor A (VEGFA) by directly interacting with specificity protein 1 (SP1), consequently exerting significant influence on angiogenic processes. Subsequently, we validated that DNA methyltransferase 3 alpha (DNMT3A) is located in the promoter of UBB to epigenetically inhibit UBB transcription. Additionally, we found that an unharmonious UBB/VEGFA ratio mediates pazopanib resistance in ccRCC. These findings underscore the critical involvement of UBB in antiangiogenic therapy and unveil a novel therapeutic strategy for ccRCC.

Blood culture quality and turnaround time of clinical microbiology laboratories in Chinese Teaching Hospitals: A multicenter study.

PURPOSE: Blood culture (BC) remains the gold standard for the diagnosis of bloodstream infections. Improving the quality of clinical BC samples, optimizing BC performance, and accelerating antimicrobial susceptibility test (AST) results are essential for the early detection of bloodstream infections and specific treatments. METHODS: We conducted a retrospective multicenter study using 450,845 BC specimens from clinical laboratories obtained from 19 teaching hospitals between 1 January 2021 and 31 December 2021. We evaluated key performance indicators (KPIs), turnaround times (TATs), and frequency distributions of processing in BC specimens. We also evaluated the AST results of clinically significant isolates for four different laboratory workflow styles. RESULTS: Across the 10 common bacterial isolates (n = 16,865) and yeast isolates (n = 1011), the overall median (interquartile range) TATs of AST results were 2.67 (2.05-3.31) and 3.73 (2.98-4.64) days, respectively. The specimen collections mainly occurred between 06:00 and 24:00, and specimen reception and loadings mainly between 08:00 and 24:00. Based on the laboratory workflows of the BCs, 16 of the 19 hospitals were divided into four groups. Time to results (TTRs) from specimen collection to the AST reports were 2.35 (1.95-3.06), 2.61 (1.98-3.32), 2.99 (2.60-3.87), and 3.25 (2.80-3.98) days for groups I, II, III, and IV, respectively. CONCLUSION: This study shows the related BC KPIs and workflows in different Chinese hospitals, suggesting that laboratory workflow optimization can play important roles in shortening time to AST reports and initiation of appropriate timely treatment.

Study on the Behavior of Saturated Cardanol-Based Surfactants at the Crude Oil/Water Interface through Molecular Dynamics Simulations.

Cardanol is a green biosurfactant with broad application prospects, which is expected to be used to enhance oil recovery (EOR). This paper designed two types of surfactants (extended and nonextended), including six kinds of nonionic and anion-nonionic surfactants. The position changes of PO and EO chains and the effects of different hydrophilic groups on the interface properties were studied with molecular dynamics simulations by constructing a model of crude oil (containing four components) and water molecules. The results of interfacial tension and solvent-accessible surface area showed that the interfacial properties of sulfate were better than those of sulfonates and nonionic surfactants. Meanwhile, the interface properties of nonextended surfactants were better than those of extended surfactants. The gyration radius (Rg) and tilt angle data demonstrated that when EO chains were located between hydrophobic groups and PO chains (nonextended surfactants), the adsorption capacity of surfactants at crude oil and water interfaces could be effectively improved. The radial distribution function of the hydrophilic group and hydrophobic group of surfactants with water molecules and four components of the crude oil molecule, respectively, explained that surfactants (8EO8POSO4) had better emulsification performance when the intermolecular interactions between crude oil and water two phases were relatively balanced. This study provides a theoretical reference for the design of oil-displacement surfactants and the mechanism analysis of emulsification properties.

Characterizing Soil Profile Salinization in Cotton Fields Using Landsat 8 Time-Series Data in Southern Xinjiang, China.

Soil salinization is a major obstacle to land productivity, crop yield and crop quality in arid areas and directly affects food security. Soil profile salt data are key for accurately determining irrigation volumes. To explore the potential for using Landsat 8 time-series data to monitor soil salinization, 172 Landsat 8 images from 2013 to 2019 were obtained from the Alar Reclamation Area of Xinjiang, northwest China. The multiyear extreme dataset was synthesized from the annual maximum or minimum values of 16 vegetation indices, which were combined with the soil conductivity of 540 samples from soil profiles at 0~0.375 m, 0~0.75 m and 0~1.00 m depths in 30 cotton fields with varying degrees of salinization as investigated by EM38-MK2. Three remote sensing monitoring models for soil conductivity at different depths were constructed using the Cubist method, and digital mapping was carried out. The results showed that the Cubist model of soil profile electrical conductivity from 0 to 0.375 m, 0 to 0.75 m and 0 to 1.00 m showed high prediction accuracy, and the determination coefficients of the prediction set were 0.80, 0.74 and 0.72, respectively. Therefore, it is feasible to use a multiyear extreme value for the vegetation index combined with a Cubist modeling method to monitor soil profile salinization at a regional scale.

Data-Efficient Learning via Minimizing Hyperspherical Energy.

Deep learning on large-scale data is currently dominant nowadays. The unprecedented scale of data has been arguably one of the most important driving forces behind its success. However, there still exist scenarios where collecting data or labels could be extremely expensive, e.g., medical imaging and robotics. To fill up this gap, this paper considers the problem of data-efficient learning from scratch using a small amount of representative data. First, we characterize this problem by active learning on homeomorphic tubes of spherical manifolds. This naturally generates feasible hypothesis class. With homologous topological properties, we identify an important connection - finding tube manifolds is equivalent to minimizing hyperspherical energy (MHE) in physical geometry. Inspired by this connection, we propose a MHE-based active learning (MHEAL) algorithm, and provide comprehensive theoretical guarantees for MHEAL, covering convergence and generalization analysis. Finally, we demonstrate the empirical performance of MHEAL in a wide range of applications for data-efficient learning, including deep clustering, distribution matching, version space sampling, and deep active learning.

Photoacoustic/ultrasonic dual-mode imaging for monitoring angiogenesis and synovial erosion in rheumatoid arthritis.

Rheumatoid arthritis (RA) is a chronic autoimmune disease characterized by the formation of new vessels, synovial proliferation and destruction of articular cartilage. However, characteristic early diagnostic and therapeutic monitoring methods are still lacking. We report a study using a photoacoustic/ultrasound (PA/US) dual-mode imaging for RA disease. By establishing a collagen-induced (CIA) RA mouse model to classify disease states based on a subjective grading system, PA/US imaging allows real-time assessment of synovial erosion and vascular opacification within the knee joint in different disease states at high spatial resolution. The system also quantitatively monitors subcutaneous vascular physiology and morphology in the hind paw of mice, measuring the area and photoacoustic signal intensity of vascular proliferation and showing a positive correlation with disease grading. Compared to traditional subjective scoring of arthritis severity, the PA/US imaging is more sensitive i.e., vascular signals and synovial erosion can be observed early in the course of arthritis.

Inversion Symmetry and Exotic Interlayer Exciton Behavior in Twisted Trilayer MoS2 Produced by Vapor Deposition.

Two-dimensional materials (2DMs) that are stacked vertically with a certain twist angle provide new degrees of freedom for designing novel physical properties. Twist-related properties of homogeneous bilayer and heterogeneous bilayer 2DMs, such as excitons and phonons, have been described in many pioneering works. However, twist-related properties of homogeneous trilayer 2DMs have been rarely reported. In this work, trilayer MoS2 with the twisted angle of 12° by optimized vapor deposition rather than the conventional mechanical stacking method was successfully fabricated. The inversion symmetry of trilayer MoS2 is changed by twist. Phonons and excitons produced by twist have an enormous influence on the interlayer interaction of trilayer MoS2, making trilayer MoS2 appear to have exotic optical properties. Compared with monolayer MoS2, the phonon vibration and photoluminescence intensity of trilayer MoS2 with one-interlayer-twisted are significantly improved, and the second harmonic generation response in the non-twist region of trilayer MoS2 is ∼3 times that of monolayer MoS2. In addition, interlayer coupling, inversion symmetry, and exciton behavior of the twist region show regional differences. This work provides a new way for designing twist and exploring the influence of twist on the structures of 2DMs with few layers.

SphereFace Revived: Unifying Hyperspherical Face Recognition.

This paper addresses the deep face recognition problem under an open-set protocol, where ideal face features are expected to have smaller maximal intra-class distance than minimal inter-class distance under a suitably chosen metric space. To this end, hyperspherical face recognition, as a promising line of research, has attracted increasing attention and gradually become a major focus in face recognition research. As one of the earliest works in hyperspherical face recognition, SphereFace explicitly proposed to learn face embeddings with large inter-class angular margin. However, SphereFace still suffers from severe training instability which limits its application in practice. In order to address this problem, we introduce a unified framework to understand large angular margin in hyperspherical face recognition. Under this framework, we extend the study of SphereFace and propose an improved variant with substantially better training stability - SphereFace-R. Specifically, we propose two novel ways to implement the multiplicative margin, and study SphereFace-R under three different feature normalization schemes (no feature normalization, hard feature normalization and soft feature normalization). We also propose an implementation strategy - "characteristic gradient detachment" - to stabilize training. Extensive experiments on SphereFace-R show that it is consistently better than or competitive with state-of-the-art methods.

The categorizations of vasculogenic mimicry in clear cell renal cell carcinoma unveil inherent connections with clinical and immune features.

Background: Clear cell renal cell carcinoma (ccRCC) stands as the prevailing variant kidney cancer in humans. Unfortunately, patients with disseminated RCC at diagnosis often have a diminished prognosis. Rapid tumor growth necessitates efficient blood supply for oxygen and nutrients, involving the circulation of blood from vessels to tumor tissues, facilitating tumor cell entry into the extracellular matrix. Vasculogenic mimicry (VM) significantly contributes to tumor growth and metastasis. Within this investigation, we identified vasculogenic mimicry-related genes (VMRGs) by analyzing data from 607 cases of kidney renal clear cell carcinoma (KIRC) in The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). These findings offer insights into ccRCC progression and metastasis. Method: We identified VMRGs-related subtypes using consistent clustering methods. The signature of the VMRGs was created using univariate Cox regression and LASSO Cox regression analyses. To evaluate differences in immune cell infiltration, we employed ssGSEA. Afterwards, we created an innovative risk assessment model, known as the VM index, along with a nomogram to forecast the prognosis of ccRCC. Additionally, we verified the expression of an important gene related to VM, peroxiredoxin 2 (PRDX2), in tissue samples. Furthermore, we assessed the sensitivity to drugs in various groups by utilizing the pRRophetic R package. Results: Significant predictors of survival rates in both high- and low-risk groups of KIRC patients were identified as VMRGs. The independent prognostic factors for RCC were confirmed by both univariate and multivariate Cox regression analyses, validating VMRG risk signatures. Differences were observed in drug sensitivity, immune checkpoint expression, and responses to immune therapy between patients classified into high- and low-VMRG-risk groups. Our nomograms consistently demonstrated precise predictive capabilities. Finally, we experimentally verified PRDX2 expression levels and their impact on prognosis. Conclusion: The signature predicts patient prognosis and therapy response, laying the groundwork for future clinical strategies in treating ccRCC patients.

Erratum: Author Correction: Advancing COVID-19 diagnosis with privacy-preserving collaboration in artificial intelligence.

[This corrects the article DOI: 10.1038/s42256-021-00421-z.].

Advancing COVID-19 Diagnosis with Privacy-Preserving Collaboration in Artificial Intelligence.

Artificial intelligence (AI) provides a promising substitution for streamlining COVID-19 diagnoses. However, concerns surrounding security and trustworthiness impede the collection of large-scale representative medical data, posing a considerable challenge for training a well-generalised model in clinical practices. To address this, we launch the Unified CT-COVID AI Diagnostic Initiative (UCADI), where the AI model can be distributedly trained and independently executed at each host institution under a federated learning framework (FL) without data sharing. Here we show that our FL model outperformed all the local models by a large yield (test sensitivity /specificity in China: 0.973/0.951, in the UK: 0.730/0.942), achieving comparable performance with a panel of professional radiologists. We further evaluated the model on the hold-out (collected from another two hospitals leaving out the FL) and heterogeneous (acquired with contrast materials) data, provided visual explanations for decisions made by the model, and analysed the trade-offs between the model performance and the communication costs in the federated training process. Our study is based on 9,573 chest computed tomography scans (CTs) from 3,336 patients collected from 23 hospitals located in China and the UK. Collectively, our work advanced the prospects of utilising federated learning for privacy-preserving AI in digital health.