Search

Abstract Background In solid-predominantly invasive lung adenocarcinoma (SPILAC), occult lymph node metastasis (OLNM) is pivotal for determining treatment strategies. This study seeks to develop and validate a fusion model combining radiomics and deep learning to predict OLNM preoperatively in SPILAC patients across multiple centers. Methods In this study, 1325 cT1a-bN0M0 SPILAC patients from six hospitals were retrospectively analyzed and divided into pathological nodal positive (pN+) and negative (pN-) groups. Three predictive models for OLNM were developed: a radiomics model employing decision trees and support vector machines; a deep learning model using ResNet-18, ResNet-34, ResNet-50, DenseNet-121, and Swin Transformer, initialized randomly or pre-trained on large-scale medical data; and a fusion model integrating both approaches using addition and concatenation techniques. The model performance was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC). Results All patients were assigned to four groups: training set (n = 470), internal validation set (n = 202), and independent test set 1 (n = 227) and 2 (n = 426). Among the 1325 patients, 478 (36%) had OLNM (pN+). The fusion model, combining radiomics with pre-trained ResNet-18 features via concatenation, outperformed others with an average AUC (aAUC) of 0.754 across validation and test sets, compared to aAUCs of 0.715 for the radiomics model and 0.676 for the deep learning model. Conclusion The radiomics-deep learning fusion model showed promising ability to generalize in predicting OLNM from CT scans, potentially aiding personalized treatment for SPILAC patients across multiple centers.

Abstract Background Chimeric antigen receptor T (CAR‐T) cell therapy has emerged as a potent treatment for relapsed or refractory multiple myeloma, demonstrating significant clinical efficacy. Despite these advances, treatment‐related toxicities, particularly infections, pose a significant challenge to patient safety. Methods This review synthesizes current knowledge on the mechanisms underlying post‐CAR‐T therapy infections, focusing on the interplay between immune dysfunction, host factors, and treatment‐induced toxicity. It provides a comprehensive analysis of the temporal and individual variability in infection characteristics and the confounding clinical presentation of cytokine release syndrome. Results The review identifies that patients receiving CAR‐T cells are at increased risk of concurrent infections due to the heterogeneity in infection characteristics across different time periods, individuals, and patient groups. It highlights the diagnostic and therapeutic complexities introduced by the overlapping symptoms of infection and cytokine release syndrome. Conclusion To enhance the infection control post‐CAR‐T therapy, this review proposes preventive strategies tailored to the early and long‐term management of patients. It underscores the need for a nuanced understanding of infection mechanisms and the importance of personalized prevention plans to improve clinical outcomes in multiple myeloma treatment.

This study aims to present a data-guided intelligent design framework for discovering high-strength, cost-effective, and low-carbon rice husk ash concrete (RHC). First, gradient boosting (GB) models were developed to predict compressive strength of RHC based on a dataset with 1150 samples. The CatBoost model outperforms XGBoost and LightGBM models on the test set, and its determination of coefficient (R2), root mean squared error (RMSE), and mean absolute error (MAE) are 0.984, 3.270 MPa, and 2.290 MPa respectively. With the best prediction model, SHapley Additive exPlanations (SHAP) was utilized to shed light on the model outputs. The curing age is the most dominant features effecting the compressive strength of RHC, and the next two are the contents of cement and water. Then, a computational design framework was proposed by integrating the machine learning-based model and the evolutionary algorithm. Considering compressive strength, materials cost, embodied carbon, and embodied energy, the multi-objective optimization was conducted leveraging the improved adaptive geometry estimation based many-objective evolutionary algorithm (AGE-MOEA II). The optimal solutions of the different design scenarios were achieved using the multi-criteria decision-making method. The research can provide a theoretical reference for the capability of developing RHC mixture using data-guided approachesto promote the sustainable development of concrete industry.

Relapsed and refractory multiple myeloma (RRMM) and B-cell leukemia/lymphoma with extramedullary disease (EMD) have poor prognosis and high mortality, lack of effective therapeutic approaches. We reported for the first time that 6 patients with malignant hematological diseases with EMD received chimeric antigen receptor (CAR)-T treatment combined with pomalidomide, and CAR-T cells were treated with pomalidomide in vitro to determine its killing activity and cytokine secretion. Three patients with RRMM were given B cell maturation antigen (BCMA)-CAR-T therapy. All 3 patients with B-cell leukemia/lymphoma received CD19/22-CAR-T sequential infusion. There were no treatment-related deaths. The maximum overall response rate (ORR) was 100%. Median follow-up was 211.5 days (75–407 days). Three patients (50%) experienced cytokine release syndrome, all of which were grade 1, and no neurotoxicity was observed. In vitro experiments showed that the killing activity did not differ significantly between BCMA-CAR-T cells with and without pomalidomide (10, 25, or 50 μg/mL) in 8226/U266 cell cocultures (P > .05). Tumor necrosis factor (TNF)-α and interferon (IFN)-γ secretion was significantly higher from 8226 and Raji cells cocultured with BCMA-CAR-T and cluster of differentiation (CD)19-CAR-T cells (P < .05). Based on the cocultures, adding pomalidomide significantly promoted IFN-γ and TNF-α secretion (P < .05). Based on the above clinical and in vitro studies demonstrating the co-administration of pomalidomide with CAR-T cell treatment demonstrated favorable tolerability and therapeutic effectiveness in RRMM or B-cell leukemia/lymphoma.

Abstract The outbreak of coronavirus disease 2019 (COVID-19) posed an unprecedented challenge on public health systems. Despite the measures put in place to contain it, COVID-19 is likely to continue experiencing sporadic outbreaks for some time, and individuals will remain susceptible to recurrent infections. Chimeric antigen receptor (CAR)-T recipients are characterized by durable B-cell aplasia, hypogammaglobulinemia and loss of T-cell diversity, which lead to an increased proportion of severe/critical cases and a high mortality rate after COVID-19 infection. Thus, treatment decisions have become much more complex and require greater caution when considering CAR T-cell immunotherapy. Hence, we reviewed the current understanding of COVID-19 and reported clinical experience in the management of COVID-19 and CAR-T therapy. After a panel discussion, we proposed a rational procedure pertaining to CAR-T recipients with the aim of maximizing the benefit of CAR-T therapy in the post COVID-19 pandemic era.

The application of waste rubber powder (WRP) for asphalt pavement could achieve the harmless and resourceful utilization of solid waste, but the storage stability of waste rubber powder modified asphalt (RA) is one of the main problems restricting its application. Existing studies have demonstrated that graphene could enhance the storage stability of RA, but graphene's size effect on the modifying effect and its corresponding mechanism are still uncertain. In this research, the effects of graphene microstructural properties (i.e., molecular size and layer number) on the storage stability of RA were investigated by storage stability testing, dynamic shear rheometry (DSR) testing and fluorescence microscopy (FM) testing, in combination with molecular dynamics simulation (MD). The experimental results indicated that graphene improved the storage stability of RA significantly, with few-layer graphene being more effective in enhancing it. MD was used to investigate the graphene size effect on RA in compatibility, intermolecular binding energy and structural stability of the system. The simulation results revealed that small-sized graphene molecules were more compatible with RA. Meanwhile, few-layer, small-sized graphene can provide higher binding energy and better enhancement of storage stability of RA. The number of graphene layers mainly influences the binding energy rather than solubility parameters. The relative concentration distribution results demonstrated that graphene facilitated the spatial distribution of asphaltenes, rubber components and light components. This research provides theoretical support for the rational selection of microstructural properties of graphene to improve the modified asphalt storage stability performance.

Snow cover is critical to the ground thermal regime because it affects surface conditions and the energy balance. Prior work inherently included confounding effects from vegetation and the soil when estimating the snow’s effect by using the differences between air and ground surface or soil temperature. Here we use the Kudryavtsev model including a snow module to isolate the effect of only the snow cover during winter, based on observational sites across the Northern Hemisphere. We find that snow’s damping of annual mean ground surface temperature averages 5.06 ± 3.15 °C and ranges 0 °C–10 °C, while damping of the annual ground surface temperature amplitude is 7.95 ± 4.95 °C, ranging 0 °C–20 °C. Greater insulation occurs in the high-latitudes. The insulating effect is primarily driven by snow depth, but also the combination of air temperature and its amplitude, snow depth, and duration. Snow cover duration and low air temperature enhance snow’s insulating ability. These observational results that isolate the damping effect of only the snow could play a significant role in better understanding the energy change between air and the ground.

This work aims to reveal the compatibility mechanism of different plasticizers with the base asphalt using molecular dynamics (MD) simulation. For this purpose, four kinds of plasticizers were chosen, namely dioctyl phthalate (DOP), dioctyl adipate (DOA), acetyl tributyl citrate (ATBC), and trioctyl trimellitate (TOTM). The compatibility between those plasticizers and the asphalt was evaluated by various MD simulation parameters, including electric dipole moment, solubility parameters, and binding energy, and the simulation results were verified using macroscopic tests. The results show that the four plasticizers are effectively miscible with asphalt within the temperature range of 130 °C to 170 °C. The optimal temperatures for DOP, DOA, ATBC, and TOTM in modified asphalt preparation are 140 °C, 130 °C, 140 °C, and 150 °C respectively. The saturate component is more miscible with weak polar plasticizers, while the other three asphalt components are more miscible with strong polar plasticizers. TOTM has the best stability in combination with asphalt and ATBC has the worst. Aromatics and saturates have better torsional and bending deformation ability, and are more likely to form stable blending systems with strong polar plasticizers. Asphaltenes and resins have more stable benzene ring structures, they have greater binding energy with DOA, which is chain-variable and weakly polarized.

IntroductionThis research explored the clinical application of grade ≥ 3 infection predictive models for the newly diagnosed multiple myeloma (NDMM) population.MethodsIt evaluated 306 patients with NDMM based on three different predictive models. The relationship between the grade ≥ 3 infection rates in NDMM and the scores was analyzed retrospectively. The cumulative incidence of early grade ≥ 3 infection was estimated using the Kaplan–Meier method and log-rank test to assess the statistical significance of the difference. To compare the predictive performance in the prediction of infection, the Receiver Operating Characteristic Curve (ROC) curve was used to show the area under the curve (AUC), and DeLong’s test was used to analyze the difference in AUC.ResultsThe incidence of grade ≥ 3 infection within the first 4 months of NDMM was 40.20%. Concerning the FIRST score (predictors: ECOG, β2-microglobulin, hemoglobin, and lactate dehydrogenase), GEM-PETHEMA score (predictors: albumin, male sex, ECOG, and non-IgA type MM), and Infection Risk model of Multiple Myeloma (IRMM) score (predictors: ECOG, serum β2-microglobulin, globulin, and hemoglobin), the probability of early grade ≥ 3 infection in the different groups showed statistically significant differences (low-risk vs. high-risk: 25.81% vs. 50.00%, p

Background: Chimeric antigen receptor (CAR) T-cell therapy carries the risk of inducing severe and life-threatening toxicities such as cytokine release syndrome (CRS), neurotoxicity, and infection. Although CRS and infections have similar symptoms, their treatment strategies differ, and early diagnosis is very important. For CRS and infections, the fastest detection time currently takes more than 24 h, so a quick and simple method to identify a fever after CAR T-cell infusion is urgently needed. Methods: We enrolled 27 patients with recurrent fever treated with different types of CAR T-cells, including cluster of differentiation (CD) 7, CD19, CD22, and CD19-CD22 bicistronic CAR T-cells, and evaluated the infection events occurring in these patients. We detailed the morphology of CAR T-cells in peripheral blood smears (PBS) and reported the infection events, CAR transgene copy number, and inflammatory indicators within the first month after treatment. Results: Similar morphological characteristics were observed in the PBS of different CAR T-cells, namely, enlarged cell bodies, deep outside and shallow inside basophilic blue cytoplasm, and natural killer (NK) cell-like purplish red granules. There were ten infections in nine of the twenty-seven patients (33%). The percentage of atypical lymphocytes in PBS was significantly associated with CAR transgene copy number and absolute lymphocyte count in all patients. The atypical lymphocyte percentage was significantly higher in the non-infection group. Conclusions: In conclusion, the unique morphology of CAR T-cells in PBS can be used to evaluate CAR T-cell kinetics and provide reliable evidence for the rapid early identification of fever after CAR T-cell infusion. Clinical Trial Registrations: ChiCTR-OPN-16008526; ChiCTR-OPN-16009847; ChiCTR2000038641; NCT05618041; NCT05388695.

Abstract Background Pancreatic ductal adenocarcinoma (PDAC) is a lethal cancer. Approximately 80% of patients initially diagnosed with locally advanced or metastatic disease survive only 4–11 months after diagnosis. Tremendous efforts have been made toward understanding the biology of PDAC. Results In this study, we first utilized next-generation sequencing technique and existing microarray datasets to identify significant differentially expressed genes between PDAC and non-tumor adjacent tissue. By comparing top significant survival genes in PDAC Gene Expression Profiling Interactive Analysis database and PDAC transcriptome data from patients, our integrated analysis discovered five potential central genes (i.e., MYEOV, KCNN4, FAM83A, S100A16, and DDX60L). Subsequently, we analyzed the cellular functions of the potential novel oncogenes MYEOV and DDX60L, which are highly expressed in PDAC cells. Notably, the knockdown of MYEOV and DDX60L significantly inhibited the metastasis of cancer cells and induced apoptosis. Further RNA sequencing analyses showed that massive signaling pathways, particularly the TNF signaling pathway and nuclear factor-kappa B (NF-κB) signaling pathway, were affected in siRNA-treated cancer cells. The siDDX60L and siMYEOV significantly inhibited the expression of chemokine CXCL2, which may potentially affect the tumor microenvironment in PDAC tissues. Conclusions The present findings identified the novel oncogene DDX60L, which was highly expressed in PDAC. Transcriptome profiling through siRNA knockdown of DDX60L uncovered its functional roles in the PDAC in humans.

The application of rubber powder to modify asphalt has become one of the main ways of resourceful and harmless utilization of waste tire rubber. However, rubber powder is prone to degradation during service, and this deficiency limits its further development. To solve the problem of insufficient aging resistance of RA, multi-layer graphene with different sizes was adopted to modify it. The aging resistance of modified asphalt was evaluated by high-temperature dynamic shear test(DSR), low-temperature bending beam test (BBR), and Brookfield rotational viscosity test (RV), and the microscopic mechanism was investigated by Fourier infrared spectroscopy (FTIR) and Four-component analysis(SARA). Furthermore, the aging model of modified asphalt was established by combining the content of asphalt components and the number of oxidation functional groups using MD technology, to reveal the size effect and anti-aging mechanism of graphene from a molecular perspective. In the molecular simulation section, the radial distribution function (RDF), mean square displacement (MSD), glass transition temperature (TG), and viscosity of the model were compared and analyzed. Macroscopic tests showed that graphene could improve the rheological properties and viscosity of RA, and the graphene with a smaller size has a more obvious improvement of asphalt anti-aging properties. Comparing the carbonyl index and carboxyl index of the four components before and after asphalt aging, it can be found that, graphene mainly relies on the inhibition of the oxidation of saturated phenols to improve the aging resistance of asphalt. In further simulation analysis, the glass transition behavior and viscosity simulation of the asphalt model confirmed the above conclusion and further revealed that the modification mechanism of graphene is related to its size. The high surface of small-sized graphene adsorbs more light components and forms a ''micelle core'', which inhibits the tendency of resin, aromatic, and asphaltene rearrangement during aging. In addition, graphene increases the energy within asphalt molecules and decreases the energy between asphalt molecules, which leads to more active intermolecular motion and promotes uniform dispersion of asphalt micelles in the dispersion medium. This research can offer a theoretical guideline for optimizing the use of graphene in modified asphalt.

The main objective of this study was to investigate the alterations in the gut microbiota (GM) of pulmonary fibrosis (PF) mice induced by bleomycin (BLM) with its underlying mechanisms. BLM was docked with the targets of TGF-β/SMAD and caspase-3 pathways using the molecular docking technique. HE staining and Masson staining were applied to observe the histopathological changes in the pulmonary tissues. Detection of the apoptotic signals was conducted by flow cytometry and TUNEL staining. The mRNA expression of targets involved in the TGF-β/SMAD and caspase-3 signaling pathways in lungs was determined by qPCR. Immunohistochemistry (IHC) assay was used to detect the expression levels of cleaved caspase-3 and BAX proteins in mice lung tissues. 16S rDNA sequencing analysis was used to investigate the changes of GM in the fecal samples of mice in each group. The results showed that the apoptosis rate of pulmonary cells in the BLM group distinctly increased, with the expression levels of crucial target pro-apoptotic gene caspase-3, BAX with the corresponding protein, cleaved caspase-3, BAX were apparently elevated. This was accompanied by a significant increase in pro-fibrotic targets level such as TGF-β, fibronectin, collagen I, and collagen III. The mechanisms of PF induced by BLM were related to apoptosis of lung tissue cells such as alveolar epithelial cells and destroyed alveolar structure and excessive production of extracellular matrix (ECM), which may be bound up with activating TGF-β/SMAD and caspase-3 pathways. As for the GM, it was found that, after BLM induced PF in mice, the micro ecological balance of the GM was destroyed; the distance of PCo1 and Pco2 was significantly elongated, and the relative abundance of some intestinal probiotics like Catenibacterium and Lactobacillus (L. johnsonii and L. gasseri) dramatically lowered while the relative abundance of Verrucomicrobiales and Enterobacteriales substantially increased. Therefore, GM changes associated with PF in mouse models induced by BLM and the concept of “gut-lung axis” might provide an optional therapeutic strategy for PF.

Abstract Psychiatric disorders such as anxiety and depression precipitated by substance use occurred during both use and withdrawal. Exosomes play significant roles in biological functions and regulate numerous physiological and pathological processes in various diseases, in particular substance use disorders (SUDs) and other psychiatric disorders. To better understand the role of exosomal miRNAs in the pathology of symptoms of anxiety and depression in patients with SUDs, we first isolated circulating exosomes from heroin-dependent patients (HDPs) and methamphetamine-dependent patients (MDPs) and identified exosomal miRNAs that were differentially expressed between patients and healthy controls (HCs). Furthermore, the correlations between exosomal DE-miRNAs and symptoms of anxiety and depression which were measured using Hamilton-Anxiety (HAM-A)/Hamilton-Depression (HAM-D) Rating Scales in the participants. Notably, the expression level of exosomal hsa-miR-16-5p, hsa-miR-129-5p, hsa-miR-363-3p, and hsa-miR-92a-3p showed significantly negative correlations with HAM-A scores in both HDPs and MDPs. But all of the 4 DE-miRNAs lost significant correlations with HAM-D scores in HDPs. Functional annotation analyses showed that the target genes of the DE-miRNAs were mainly enriched for “synapse”, “cell adhesion”, “focal adhesion” and “MHC class II protein complex”. Our study suggests that a set of circulating exosomal miRNAs were associated with anxiety and depression in SUD patients and may have clinical utility as diagnostic and prognostic biomarkers.

Methamphetamine (MA) abuse results in neurotoxic outcomes, including increased anxiety and depression. Studies have reported an association between MA exposure and anxiety, nonetheless, the underlying mechanism remains elusive. In the present study, we developed a mouse model of anxiety-like behavior induced by MA administration. RNA-seq was then performed to profile the gene expression patterns of hippocampus (HIPP), and the differentially expressed genes (DEGs) were significantly enriched in signaling pathways related to psychiatric disorders and mitochondrial function. Based on these, mitochondria was hypothesized to be involved in MA-induced anxiety. Quercetin, as a mitochondrial protector, was used to investigate whether to be a potential treatment for MA-induced anxiety; accordingly, it alleviated anxiety-like behavior and improved mitochondrial impairment in vivo. Further experiments in vitro suggested that quercetin alleviated the dysfunction and morphological abnormalities of mitochondria induced by MA, via decreasing the levels of reactive oxygen species (ROS), mitochondrial membrane potential (MMP), and increasing the oxygen consumption rate (OCR) and ATP production. Moreover, the study examined the effect of quercetin on astrocytes activation and neuroinflammation, and the results indicated that it significantly attenuated the activation of astrocytes and reduced the levels of IL-1β, TNFα but not IL-6. In light of these findings, quantitative evidence is presented in the study supporting the view that MA can evoke anxiety-like behavior via the induction of mitochondrial dysfunction. Quercetin exerted antipsychotic activity through modulation of mitochondrial function and neuroinflammation, suggesting its potential for further therapeutic development in MA-induced anxiety.

Abstract Consider positive solutions and multiple positive solutions for a discrete nonlinear third-order boundary value problem {Δ3u(t−1)=a(t)f(t,u(t)),t∈[1,T−2]Z,Δu(0)=u(T)=0,Δ2u(η)−αΔu(T−1)=0, $$ \textstyle\begin{cases} \Delta ^{3}u(t-1)=a(t)f(t,u(t)), \quad t\in [1,T-2]_{\mathbb{Z}},\\ \Delta u(0)=u(T)=0,\qquad \Delta ^{2} u(\eta )-\alpha \Delta u(T-1)=0, \end{cases} $$ which has the sign-changing Green’s function. Here T>8 $T>8$ is a positive integer, [1,T−1]Z={1,2,…,T−2} $[1,T-1]_{\mathbb{Z}}=\{1,2,\dots ,T-2\}$, α∈[0,1T−1) $\alpha \in [0, \frac{1}{T-1})$, a:[0,T−2]Z→(0,+∞) $a:[0,T-2]_{\mathbb{Z}}\to (0,+\infty )$, f:[1,T−2]Z×[0,∞)→[0,∞) $f:[1,T-2]_{ \mathbb{Z}}\times [0,\infty )\to [0,\infty )$ is continuous.

Accurately identifying tumors from MRI scans is of the utmost importance for clinical diagnostics and when making plans regarding brain tumor treatment. However, manual segmentation is a challenging and time-consuming process in practice and exhibits a high degree of variability between doctors. Therefore, an axial attention brain tumor segmentation network was established in this paper, automatically segmenting tumor subregions from multi-modality MRIs. The axial attention mechanism was employed to capture richer semantic information, which makes it easier for models to provide local–global contextual information by incorporating local and global feature representations while simplifying the computational complexity. The deep supervision mechanism is employed to avoid vanishing gradients and guide the AABTS-Net to generate better feature representations. The hybrid loss is employed in the model to handle the class imbalance of the dataset. Furthermore, we conduct comprehensive experiments on the BraTS 2019 and 2020 datasets. The proposed AABTS-Net shows greater robustness and accuracy, which signifies that the model can be employed in clinical practice and provides a new avenue for medical image segmentation systems.

Heroin addiction and withdrawal influence multiple physiological functions, including immune responses, but the mechanism remains largely elusive. The objective of this study was to investigate the molecular inflammatory interactome, particularly the cytokines and transcriptome regulatory network in heroin addicts undergoing withdrawal, compared to healthy controls (HCs). Twenty-seven cytokines were simultaneously assessed in 41 heroin addicts, including 20 at the acute withdrawal (AW) stage and 21 at the protracted withdrawal (PW) stage, and 38 age- and gender-matched HCs. Disturbed T-helper(Th)1/Th2, Th1/Th17, and Th2/Th17 balances, characterized by reduced interleukin (IL)-2, elevated IL-4, IL-10, and IL-17A, but normal TNF-α, were present in the AW subjects. These imbalances were mostly restored to the baseline at the PW stage. However, the cytokines TNF-α, IL-2, IL-7, IL-10, and IL-17A remained dysregulated. This study also profiled exosomal long non-coding RNA (lncRNA) and mRNA in the plasma of heroin addicts, constructed co-expression gene regulation networks, and identified lncRNA-mRNA-pathway pairs specifically associated with alterations in cytokine profiles and Th1/Th2/Th17 imbalances. Altogether, a large amount of cytokine and exosomal lncRNA/mRNA expression profiling data relating to heroin withdrawal was obtained, providing a useful experimental and theoretical basis for further understanding of the pathogenic mechanisms of withdrawal symptoms in heroin addicts.

Deep neural networks have been widely used in detection tasks based on optical remote sensing images. However, in recent studies, deep neural networks have been shown to be vulnerable to adversarial examples. Adversarial examples are threatening in both the digital and physical domains. Specifically, they make it possible for adversarial examples to attack aerial remote sensing detection. To defend against adversarial attacks on aerial remote sensing detection, we propose a cascaded adversarial defense framework, which locates the adversarial patch according to its high frequency and saliency information in the gradient domain and removes it directly. The original image semantic and texture information is then restored by the image inpainting method. When combined with the random erasing algorithm, the robustness of detection is further improved. Our method is the first attempt to defend against adversarial examples in remote sensing detection. The experimental results show that our method is very effective in defending against real-world adversarial attacks. In particular, when using the YOLOv3 and YOLOv4 algorithms for robust detection of single-class targets, the AP60 of YOLOv3 and YOLOv4 only drop by 2.11% and 2.17%, respectively, under the adversarial example.