Search

Alzheimer’s disease (AD) remains a global health challenge in the 21st century due to its increasing prevalence as the major cause of dementia. State-of-the-art artificial intelligence (AI)-based tests could potentially improve population-based strategies to detect and manage AD. Current retinal imaging demonstrates immense potential as a non-invasive screening measure for AD, by studying qualitative and quantitative changes in the neuronal and vascular structures of the retina that are often associated with degenerative changes in the brain. On the other hand, the tremendous success of AI, especially deep learning, in recent years has encouraged its incorporation with retinal imaging for predicting systemic diseases. Further development in deep reinforcement learning (DRL), defined as a subfield of machine learning that combines deep learning and reinforcement learning, also prompts the question of how it can work hand in hand with retinal imaging as a viable tool for automated prediction of AD. This review aims to discuss potential applications of DRL in using retinal imaging to study AD, and their synergistic application to unlock other possibilities, such as AD detection and prediction of AD progression. Challenges and future directions, such as the use of inverse DRL in defining reward function, lack of standardization in retinal imaging, and data availability, will also be addressed to bridge gaps for its transition into clinical use.

Purpose To evaluate the efficacy and safety of pegylated liposomal doxorubicin (PLD) in patients with human epidermal growth factor receptor 2 (HER2)-negative metastatic breast cancer (MBC) heavily pretreated with anthracycline and taxanes. Methods In this single-arm, phase II study, patients with HER2-negative MBC previously treated with anthracycline and taxanes as second- to fifth chemotherapy received PLD (Duomeisu®, generic doxorubicin hydrochloride liposome) 40 mg/m2 every 4 weeks until disease progression, unacceptable toxicity, or completion of six cycles. Primary endpoint was progression-free survival (PFS). Secondary endpoints included overall survival (OS), objective response rate (ORR), disease control rate (DCR), clinical benefit rate (CBR), and safety. Results Of 44 enrolled patients (median age, 53.5 years; range, 34–69), 41 and 36 were evaluable for safety and efficacy, respectively. In total, 59.1% (26/44) of patients had ≥ 3 metastatic sites, 86.4% (38/44) had visceral disease, and 63.6% (28/44) had liver metastases. Median PFS was 3.7 months (95% confidence interval [CI] 3.3–4.1) and median OS was 15.0 months (95% CI 12.1–17.9). ORR, DCR, and CBR were 16.7%, 63.9%, and 36.1%, respectively. The most common adverse events (AEs) were leukopenia (53.7%), fatigue (46.3%), and neutropenia (41.5%), with no grade 4/5 AEs. The most common grade 3 AEs were neutropenia (7.3%) and fatigue (4.9%). Patients experienced palmar-plantar-erythrodysesthesia (24.4%, 2.4% grade 3), stomatitis (19.5%, 7.3% grade 2), and alopecia (7.3%). One patient displayed a left ventricular ejection fraction decline of 11.4% from baseline after five cycles of PLD therapy. Conclusion PLD (Duomeisu®) 40 mg/m2 every 4 weeks was effective and well-tolerated in patients with HER2-negative MBC heavily pretreated with anthracycline and taxanes, revealing a potentially viable treatment option for this population. Trial registration Chinese Clinical Trial Registry: ChiCTR1900022568.

Iodide/triiodide redox shuttles as new additives are introduced to simultaneously passivate the cation and anion defects of methylammonium lead iodide-based perovskite solar cells.

Proxalutamide is a novel second-generation non-steroidal androgen receptor (AR) antagonist. This study aimed to evaluate the preliminary efficacy and safety of proxalutamide in patients with AR-positive metastatic breast cancer (ARIn this open-label, dose-expansion, multicentre phase Ib trial, patients with ARForty-five patients with three median lines (range, 1-13) prior systemic therapy were enrolled (cohort A, n = 30; cohort B, n = 15). Among 39 evaluable patients, DCR at 8 and 16 weeks was 25.6% (95% confidence interval [CI], 11.9-39.4%), with 26.9% in cohort A and 23.1% in cohort B. No patient achieved partial response or complete response. Proxalutamide 200 mg/day was determined as RP2D. The 6-month progression-free survival (PFS) rate was 19.6% (95% CI, 10.2-37.5%). In the triple-negative subgroup, DCR at 8 weeks was 38.5%, with median PFS of 9.1 months (95% CI, 7.8-NA) in those who achieved response at 8 weeks (n = 5). Most common grade 3/4 adverse events were aspartate aminotransferase increase (8.9%) and γ-glutamyltransferase increase (8.9%). By biomarker analysis, patients with moderate AR expression of IHC (26%-75%), PIK3CA pathogenic mutations, or60 ng/ml cell-free DNA yield showed longer PFS.Proxalutamide showed promising anti-tumour activity with good tolerability in patients with heavily pretreated ARThis clinical study was prospectively registered at chinadrugtrials.org.cn (Identifier: CTR20170757) and clinical trials.gov (Identifier: NCT04103853).

The C≡N stretching vibration is a versatile infrared (IR) reporter that is useful for a wide range of applications. Aiming to further expand its spectroscopic utility, herein, we show that, using 4-cyanoindole and 4-cyano-7-azaindole as examples, photoexcitation can significantly shift the frequency (ν

To add to the limited data that exist on the selection of drugs to prevent mother-to-child transmission (MTCT) of hepatitis B virus (HBV).This is a prospective cohort study that enrolled mothers with HBV-DNA ≥2 × 10A total of 270 mothers received TDF treatment and 275 mothers received LdT treatment. The predelivery decline in HBV-DNA in the TDF group was higher than the LdT group (3.92 ± 0.93 log IU/ml vs. 3.76 ± 0.94 log IU/ml, P = 0.043). In the primary analysis, the MTCT rate of the TDF group was comparable to that of the LdT group, both in the intention-to-treat analysis (1.5% [4/275] vs. 1.8% [5/273], P 0.99) and the per-protocol analysis (0% in both groups, P 0.99). The alanine aminotransferase elevation rates in the TDF group were lower than in the LdT group (17.3% vs. 27.4%, P = 0.005). Less anorexia and more arthralgia were observed in the LdT group than the TDF group.TDF and LdT are both effective in preventing MTCT of HBV, but they may cause different adverse events. TDF is more effective in reducing HBV viral load and had fewer alanine aminotransferase abnormalities than LdT.

As a global health threat, NASH has been confirmed to be a chronic progressive liver disease that is strongly associated with obesity. However, no approved drugs or efficient therapeutic strategies are valid, mainly because its complicated pathological processes is underestimated.We identified the RING-type E3 ubiquitin transferase-tripartite motif-containing protein 31 (TRIM31), a member of the E3 ubiquitin ligases family, as an efficient endogenous inhibitor of transforming growth factor-beta-activated kinase 1 (mitogen-activated protein kinase kinase kinase 7; MAP3K7), and we further confirmed that TRIM31 is an MAP3K7-interacting protein and promotes MAP3K7 degradation by enhancing ubiquitination of K48 linkage in hepatocytes. Hepatocyte-specific Trim31 deletion blocks hepatic metabolism homeostasis, concomitant with glucose metabolic syndrome, lipid accumulation, up-regulated inflammation, and dramatically facilitates NASH progression. Inversely, transgenic overexpression, lentivirus, or adeno-associated virus-mediated Trim31 gene therapy restrain NASH in three dietary mice models. Mechanistically, in response to metabolic insults, TRIM31 interacts with MAP3K7 and conjugates K48-linked ubiquitination chains to promote MAP3K7 degradation, thus blocking MAP3K7 abundance and its downstream signaling cascade activation in hepatocytes.TRIM31 may serve as a promising therapeutic target for NASH treatment and associated metabolic disorders.

T cell heterogeneity presents a challenge for accurate cell identification, understanding their inherent plasticity, and characterizing their critical role in adaptive immunity. Immunologists have traditionally employed techniques such as flow cytometry to identify T cell subtypes based on a well-established set of surface protein markers. With the advent of single-cell RNA sequencing (scRNA-seq), researchers can now investigate the gene expression profiles of these surface proteins at the single-cell level. The insights gleaned from these profiles offer valuable clues and a deeper understanding of cell identity. However, CD45RA, the isoform of CD45 which distinguish between naïve/central memory T cells and effector memory/effector memory cells re-expressing CD45RA T cells, cannot be well profiled by scRNA-seq due to the difficulty in mapping short reads to genes. In order to facilitate cell type annotation in T cell scRNA-seq analysis, we employed machine learning and trained a CD45RA+/-classifier on single-cell mRNA count data annotated with known CD45RA antibody levels provided by cellular indexing of transcriptomes and epitopes sequencing (CITE-seq) data. Among all algorithms we tested, the trained support vector machine (SVM) with a radial basis function (RBF) kernel with optimized hyperparameters achieved a 99.96% accuracy on an unseen dataset. The multilayer Perceptron (MLP) classifier, the second most predictive method overall, also achieved a decent accuracy of 99.74%. Our simple yet robust machine learning approach provides a valid inference on the CD45RA level, assisting the cell identity annotation and further exploring the heterogeneity within human T cells.

The gut plays a critical role in maintaining human health by facilitating the absorption of nutrients, regulating metabolism, and interacting with the immune system and gut microbiota. The co-culture of two human colorectal cancer cell lines, Caco-2 and HT29, on Transwell is commonly used as anin vitrogut mimic in studies of intestine absorption pharmacokinetic, gut mechanics, and gut-microbes interplay given the similar morphology, expression of transporters and enzymes, and barrier function. However, to sufficiently evaluate the translatability of insights from such a system to human physiological contexts, a detailed survey of cell type heterogeneity in the system and a holistic comparison with human physiology needs to be conducted rather than by the presence of a few well-studied proteins. Single-cell RNA sequencing provides high-resolution expression profiles of cells in the co-culture, enabling the heterogeneity to be characterized and the similarity to human epithelial cells to be evaluated. 16019 genes transcriptional profile in 13784 cells were acquired and compared to human epithelial cells (GSE185224). We identified the intestinal stem cell, transit amplifying, enterocyte, goblet cell, and enteroendocrine-like cells together with differentiating HT29 cells in the system based on the expression of canonical markers in healthy adult human epithelial cells. The epithelium-like co-culture was fetal intestine-like, with less variety of gene expression compared to the human gut. Transporters for major types of substance (lipid, amino acid, ion, water, etc.) were found transcribed in the majority of the enterocytes-like cells in the system. However, some of the well-studied transporters were absent. Toll-like receptors were not highly expressed in the sample, yet the treatment of LPS still caused a mild change in TEER and gene expression, possibly by the interaction with CD14. Overall, the Caco-2/HT29 co-culture is a cost-effective epithelium model for drug permeability testing or mechanical simulation, but its discrepancy with the real epithelium phenotype-wise is not negligible. As a result, its response to biological factors might not provide transferrable knowledge to the study of human gut physiology, especially the innate immune aspect.

With the global pandemic of the new coronavirus disease (COVID-19), a safe, effective, and affordable mass-produced vaccine remains the current focus of research. Herein, we designed an adjuvant-protein conjugate vaccine candidate, in which the TLR7 agonist (TLR7a) was conjugated to S1 subunit of SARS-CoV-2 spike protein, and systematically compared the effect of different numbers of built-in TLR7a on the immune activity for the first time. As the number of built-in TLR7a increased, a bell-shaped reaction was observed in three TLR7a-S1 conjugates, with TLR7a(10)-S1 (with around 10 built-in adjuvant molecules on one S1 protein) eliciting a more potent immune response than TLR7a(2)-S1 and TLR7a(18)-S1. This adjuvant-protein conjugate strategy allows the built-in adjuvant to provide cluster effects and prevents systemic toxicity and facilitates the co-delivery of adjuvant and antigen. Vaccination of mice with TLR7a(10)-S1 triggered a potent humoral and cellular immunity and a balanced Th1/Th2 immune response. Meanwhile, the vaccine induces effective neutralizing antibodies against SARS-CoV-2 and all variants of concern (B.1.1.7/alpha, B.1.351/beta, P.1/gamma, B.1.617.2/delta, and B.1.1.529/omicron). It is expected that the adjuvant-protein conjugate strategy has great potential to construct a potent recombinant protein vaccine candidate against various types of diseases.

Background. Last two decades, many association studies on genetic variants and chronic obstructive pulmonary disease (COPD) risk have been published. But results from different studies are inconsistent. Therefore, we performed this article to systematically evaluate results from previous meta-analyses and genome-wide association studies (GWASs). Material and Methods. Firstly, we retrieved meta-analyses in PubMed, Embase, and China National Knowledge Infrastructure and GWASs in PubMed and GWAS catalog on or before April 7th, 2022. Then, data were extracted and screened. Finally, two main methods—Venice criteria and false-positive report probability test—were used to evaluate significant associations. Results. As a result, eighty-eight meta-analyses and 5 GWASs were deemed eligible for inclusion. Fifty variants in 26 genes obtained from meta-analyses were significantly associated with COPD risk. Cumulative epidemiological evidence of an association was graded as strong for 10 variants in 8 genes (GSTM1, CHRNA, ADAM33, SP-D, TNF-α, VDBP, HMOX1, and HHIP), moderate for 6 variants in 5 genes (PI, GSTM1, ADAM33, TNF-α, and VDBP), and weak for 40 variants in 23 genes. Five variants in 4 genes showed convincing evidence of no association with COPD risk in meta-analyses. Additionally, 29 SNPs identified in GWASs were proved to be noteworthy based on the FPRP test. Conclusion. In summary, more than half (52.38%) of genetic variants reported in previous meta-analyses showed no association with COPD risk. However, 13 variants in 9 genes had moderate to strong evidence for an association. This article can serve as a useful reference for further studies.

Long noncoding RNAs (lncRNA) are dysregulated in various human cancers and control tumor development and progression. However, the upstream mechanisms underlying their dysregulation remain unclear. Here, we demonstrate that the expression of hepatocyte nuclear factor 1 homeobox A antisense RNA 1 (HNF1A-AS1) is significantly upregulated in gastric cancer tissues. Overexpression of HNF1A-AS1 enhanced cell proliferation and promoted cell-cycle progression, whereas knockdown of HNF1A-AS1 elicited the opposite effects. Early growth response protein 1 (EGR1) directly bound the HNF1A-AS1 promoter region and activated its transcription. Overexpression of EGR1 enhanced cell proliferation and promoted cell-cycle promotion, similar to the function of HNF1A-AS1. HNF1A-AS1 functioned as competing endogenous RNA (ceRNA) by binding to miR-661, upregulating the expression of cell division cycle 34 (CDC34), which is a direct target of miR-661. EGR1 and HNF1A-AS1 enhanced the expression of cyclin-dependent kinase 2 (CDK2), CDK4, and cyclin E1 but inhibited the expression of p21 by promoting CDC34-mediated ubiquitination and degradation of p21. Taken together, these findings suggest that EGR1-activated HNF1A-AS1 regulates various pro- and antigrowth factors to promote the development of gastric cancer, implicating it as a possible target for therapeutic intervention in this disease.Significance: This study provides novel insights into mechanisms by which the noncoding RNA HNF1A-AS1 contributes to gastric cancer progression through modulation of the cell cycle. Cancer Res; 78(20); 5877–90. ©2018 AACR.

Figure S1 showed that the overexpression and knockdown efficiency of HNF1A-AS1 in both GC cells and the proliferation of GC cells transfected with SI-HNF1A-AS1-2 were inhibited. Figure S2 indicated that the migration, invasion and cell cycle progression of GC cells transfected with SI-HNF1A-AS1-2 were inhibited. Table S1 showed sequences of primers.

Background This study aimed to investigate the relationship between chronotypes and aggression in adolescents. Methods A cross-sectional study was conducted on 755 primary and secondary school students aged 11–16 years in rural areas of Ningxia Province, China. The Chinese version of the Buss-Perry Aggression Questionnaire (AQ-CV) and the Chinese version Morningness-Eveningness Questionnaire (MEQ-CV) were used to assess the aggressive behavior and chronotypes of the study subjects. The Kruskal-Wallis test was then used to compare the differences in aggression among adolescents with different chronotypes, and Spearman correlation analysis to determine the relationship between chronotypes and aggression. Further linear regression analysis was used to investigate the effects of chronotype, personality traits, family environment, and class environment on adolescent aggression. Results There were significant differences in chronotypes between different age groups and different sexes. Spearman correlation analysis showed that the MEQ-CV total score was negatively correlated with the AQ-CV total score (r = -0.263) and score of each AQ-CV subscale. In Model 1, chronotypes were negatively associated with aggression when controlling for age and sex, and evening-type adolescents might be more likely to exhibit aggressive behavior (b = -0.513, 95% CI: [-0.712, -0.315], P P P Conclusion Compared to morning-type adolescents, evening-type adolescents were more likely to exhibit aggressive behavior. Given social expectations for MT adolescents, adolescents should be actively guided to develop a good circadian rhythm that may be more conducive to their physical and mental development.

Mitochondria damage is an important cause of heart dysfunction after severe burn. However, the pathophysiological process remains unclear. This study aims to examine the mitochondrial dynamics in the heart after severe burn, and the role of μ-calpain, a cysteine protease, in this scenario. Rats were subjected to thermal injury treatment, and calpain inhibitor MDL28170 was given intravenously 1 h before burn injury. Compared with the sham group, the rats in the burn group displayed weakened heart performance and decreased mean arterial pressure, which was accompanied by a diminishment of mitochondria function. Furthermore, severe burn increased the level of calpain in mitochondria, reflected by immunofluorescence staining and activity test. In contrast, treatment with MDL28170 diminished these responses to a severe burn. Severe burn induced morphological defects of mitochondria, and decreased the abundance of mitochondria. Of note, severe burn increased the percentage of the mitochondria with bigger size, but decreased the smaller ones. Analysis of dynamic proteins revealed that severe burn caused an increase of fission protein DRP1 in the mitochondria and a decrease of inner membrane fusion protein OPA1. Similarly, these alterations were also blocked by MDL28170. Last but not the least, inhibition of calpain yielded the emergence of more elongated mitochondria along with membrane invagination in the middle of the longitude, an indicator of the fission process. Overall, these results for the first time provide evidence that mitochondria recruitment of calpain confers heart dysfunction after severe burn, which involves mitochondrial dynamics damage.

Gene therapy has shown great potential to treat various diseases by repairing the abnormal gene function. However, a great challenge in bringing the nucleic acid formulations to the market is the safe and effective delivery to the specific tissues and cells. To be excited, the development of ionizable drug delivery systems (IDDSs) has promoted a great breakthrough as evidenced by the approval of the BNT162b2 vaccine for prevention of coronavirus disease 2019 (COVID-19) in 2021. Compared with conventional cationic gene vectors, IDDSs can decrease the toxicity of carriers to cell membranes, and increase cellular uptake and endosomal escape of nucleic acids by their unique pH-responsive structures. Despite the progress, there remain necessary requirements for designing more efficient IDDSs for precise gene therapy. Herein, we systematically classify the IDDSs and summarize the characteristics and advantages of IDDSs in order to explore the underlying design mechanisms. The delivery mechanisms and therapeutic applications of IDDSs are comprehensively reviewed for the delivery of pDNA and four kinds of RNA. In particular, organ selecting considerations and high-throughput screening are highlighted to explore efficiently multifunctional ionizable nanomaterials with superior gene delivery capacity. We anticipate providing references for researchers to rationally design more efficient and accurate targeted gene delivery systems in the future, and indicate ideas for developing next generation gene vectors.

Background and Purpose: Colorectal cancer (CRC) is an exceptionally deadly disease, whereas therapeutic drugs for CRC have presented a serious shortage over the past few decades. Natural products have become an inexhaustible source of anticancer drugs. Natural alkaloid N-hydroxyapiosporamide (NHAP) exerts antitumor effects, but its effect and mechanism in CRC remain unclear. This study aimed to reveal the antitumor target of NHAP and identify NHAP as a promising leading compound for CRC. Experimental Approach: The inhibitory effects of NHAP on growth of CRC cells were determined by SRB and colony formation assays. Various biochemical methods were used to investigate molecular mechanisms of action for NHAP, including western blotting, RT-PCR, flow cytometry, immunofluorescent staining, cell transfection and luciferase assay, RNA-seq analysis, ELISA and immunoprecipitation analysis. Mouse endotoxin shock model, CRC xenograft model and azoxymethane model were used to assess the in vivo anti-tumor effect of NHAP against CRC. Key Results: NHAP exhibited potent cytotoxicity, induced both apoptosis and autophagic cell death of CRC cells, and inhibited the NF-κB signaling pathway by blocking the interaction of TAK1-TRAF6 proteins. NHAP also markedly inhibited CRC tumour growth in vivo without obvious toxicity and possessed superior pharmacokinetic characteristics. Conclusion and Implications: These findings identify, for the first time, that natural alkaloid NHAP is a novel NF-κB inhibitor with potent anti-tumor activity against CRC in vitro and in vivo. This study clarifies the anti-tumor target of NHAP against CRC, which will contribute to the future development of NHAP as a novel therapeutic leading compound for CRC.

Electrostatic interactions are key determinants of protein structure, dynamics, and function. Since protein electrostatics are nonuniform, assessment of the internal electric fields (EFs) of proteins requires spatial resolution at the amino acid residue level. In this regard, vibrational Stark spectroscopy, in conjunction with various unnatural amino acid-based vibrational probes, has become a common method for site-specific interrogation of protein EFs. However, application of this method is often limited to proteins with relatively high solubility, due to the intrinsically low oscillator strength of vibrational transitions. Therefore, it would be useful to develop an alternative method that can overcome this limitation. To this end, we show that, using solvatochromic study and molecular dynamics simulations, the frequency of maximum emission intensity of the fluorophore of 4-cyanotryptophan (4CN-Trp), 3-methyl-1

Supplemental material, sj-pdf-1-imr-10.1177_03000605221121952 for Spinal subdural hematoma in a patient with immune thrombocytopenic purpura following microvascular decompression: a rare case report by Daoqing Su, Zonglei Chong, Ran Ran, Xiaoxiao Peng, Lei Hou, Qiang Zong and Hongxing Li in Journal of International Medical Research

Additional file 1. Multiple linear regression analysis showed that operative time and number of reloads correlated with surgical method.

Supplemental material, sj-pdf-1-imr-10.1177_03000605221121952 for Spinal subdural hematoma in a patient with immune thrombocytopenic purpura following microvascular decompression: a rare case report by Daoqing Su, Zonglei Chong, Ran Ran, Xiaoxiao Peng, Lei Hou, Qiang Zong and Hongxing Li in Journal of International Medical Research

Multispectral and Hyperspectral Image Fusion (MHIF) is a practical task that aims to fuse a high-resolution multispectral image (HR-MSI) and a low-resolution hyperspectral image (LR-HSI) of the same scene to obtain a high-resolution hyperspectral image (HR-HSI). Benefiting from powerful inductive bias capability, CNN-based methods have achieved great success in the MHIF task. However, they lack certain interpretability and require convolution structures be stacked to enhance performance. Recently, Implicit Neural Representation (INR) has achieved good performance and interpretability in 2D tasks due to its ability to locally interpolate samples and utilize multimodal content such as pixels and coordinates. Although INR-based approaches show promise, they require extra construction of high-frequency information (\emph{e.g.,} positional encoding). In this paper, inspired by previous work of MHIF task, we realize that HR-MSI could serve as a high-frequency detail auxiliary input, leading us to propose a novel INR-based hyperspectral fusion function named Implicit Neural Feature Fusion Function (INF). As an elaborate structure, it solves the MHIF task and addresses deficiencies in the INR-based approaches. Specifically, our INF designs a Dual High-Frequency Fusion (DHFF) structure that obtains high-frequency information twice from HR-MSI and LR-HSI, then subtly fuses them with coordinate information. Moreover, the proposed INF incorporates a parameter-free method named INR with cosine similarity (INR-CS) that uses cosine similarity to generate local weights through feature vectors. Based on INF, we construct an Implicit Neural Fusion Network (INFN) that achieves state-of-the-art performance for MHIF tasks of two public datasets, \emph{i.e.,} CAVE and Harvard. The code will soon be made available on GitHub.

Denosing diffusion model, as a generative model, has received a lot of attention in the field of image generation recently, thanks to its powerful generation capability. However, diffusion models have not yet received sufficient research in the field of image fusion. In this article, we introduce diffusion model to the image fusion field, treating the image fusion task as image-to-image translation and designing two different conditional injection modulation modules (i.e., style transfer modulation and wavelet modulation) to inject coarse-grained style information and fine-grained high-frequency and low-frequency information into the diffusion UNet, thereby generating fused images. In addition, we also discussed the residual learning and the selection of training objectives of the diffusion model in the image fusion task. Extensive experimental results based on quantitative and qualitative assessments compared with benchmarks demonstrates state-of-the-art results and good generalization performance in image fusion tasks. Finally, it is hoped that our method can inspire other works and gain insight into this field to better apply the diffusion model to image fusion tasks. Code shall be released for better reproducibility.

The proliferation of deep learning (DL) has led to the emergence of privacy and security concerns. To address these issues, secure Two-party computation (2PC) has been proposed as a means of enabling privacy-preserving DL computation. However, in practice, 2PC methods often incur high computation and communication overhead, which can impede their use in large-scale systems. To address this challenge, we introduce RRNet, a systematic framework that aims to jointly reduce the overhead of MPC comparison protocols and accelerate computation through hardware acceleration. Our approach integrates the hardware latency of cryptographic building blocks into the DNN loss function, resulting in improved energy efficiency, accuracy, and security guarantees. Furthermore, we propose a cryptographic hardware scheduler and corresponding performance model for Field Programmable Gate Arrays (FPGAs) to further enhance the efficiency of our framework. Experiments show RRNet achieved a much higher ReLU reduction performance than all SOTA works on CIFAR-10 dataset., Comment: This is work is a updated version of arXiv:2209.09424, the original version has been withdrawn

Semi-supervised medical image segmentation offers a promising solution for large-scale medical image analysis by significantly reducing the annotation burden while achieving comparable performance. Employing this method exhibits a high degree of potential for optimizing the segmentation process and increasing its feasibility in clinical settings during translational investigations. Recently, cross-supervised training based on different co-training sub-networks has become a standard paradigm for this task. Still, the critical issues of sub-network disagreement and label-noise suppression require further attention and progress in cross-supervised training. This paper proposes a cross-supervised learning framework based on dual classifiers (DC-Net), including an evidential classifier and a vanilla classifier. The two classifiers exhibit complementary characteristics, enabling them to handle disagreement effectively and generate more robust and accurate pseudo-labels for unlabeled data. We also incorporate the uncertainty estimation from the evidential classifier into cross-supervised training to alleviate the negative effect of the error supervision signal. The extensive experiments on LA and Pancreas-CT dataset illustrate that DC-Net outperforms other state-of-the-art methods for semi-supervised segmentation. The code will be released soon., Comment: 13 pages, 4 figures, 5 tables. Code will come soon

Two-party computation (2PC) is promising to enable privacy-preserving deep learning (DL). However, the 2PC-based privacy-preserving DL implementation comes with high comparison protocol overhead from the non-linear operators. This work presents PASNet, a novel systematic framework that enables low latency, high energy efficiency & accuracy, and security-guaranteed 2PC-DL by integrating the hardware latency of the cryptographic building block into the neural architecture search loss function. We develop a cryptographic hardware scheduler and the corresponding performance model for Field Programmable Gate Arrays (FPGA) as a case study. The experimental results demonstrate that our light-weighted model PASNet-A and heavily-weighted model PASNet-B achieve 63 ms and 228 ms latency on private inference on ImageNet, which are 147 and 40 times faster than the SOTA CryptGPU system, and achieve 70.54% & 78.79% accuracy and more than 1000 times higher energy efficiency., DAC 2023 accepeted publication, short version was published on AAAI 2023 workshop on DL-Hardware Co-Design for AI Acceleration: RRNet: Towards ReLU-Reduced Neural Network for Two-party Computation Based Private Inference