Search

Face forgery detection aims to identify manipulated or altered facial images or videos created using artificial intelligence. Existing detection methods exhibit favorable performance on high-quality videos, but the videos in daily applications are commonly compressed into low-quality formats via social media. The detection difficulty is increased by the poor quality, indistinct detail features, and noises such as artifacts in these images or videos. To address this challenge, we propose a multi-scale dual-branch network for compressed face forgery, called MDCF-Net, effectively capturing cross-domain forgery features at various scales in compressed facial images. The MDCF-Net comprises two branches: an RGB domain branch utilizing Transformers to extract multi-scale fine-texture features from the original RGB images; a frequency domain branch designed to capture artifacts in low-quality videos by extracting global spectral features as a supplementary measure. Then, we introduce a feature fusion module (FFM) based on multi-head attention to merge diverse feature representations in a spatial-frequency complementary manner. Extensive comparative experiments on public datasets such as FaceForensics++, Celeb-DF, and WildDeepfake demonstrate the significant advantage of MDCF-Net in detecting highly compressed and low-quality forged images or videos, especially in achieving state-of-the-art performance on the FaceForensics++ low-quality dataset. Our approach presents a new perspective and technology for low-quality face forgery detection.

How to achieve fast and accurate small object detection holds crucial theoretical and practical significance. However, this task encounters substantial challenges due to scale differences among instances in the scene, along with the scarcity of inherent features and weak representation of small instances. To alleviate the above problems, we propose a novel attention-based cascaded adaptive feature pyramid fusion network, CA2Det, to effectively improve the small object detection performance. First, to prevent the information degradation of small-instance features during training, we introduce the efficient and lightweight Shuffle Attention mechanism to highlight the features of small instances. Second, to mitigate the information conflicts arising from the scale inconsistency among instances, we design a double-layer cascaded adaptive fusion pyramid module, CAFP, which can effectively suppress the information conflicts while enabling full information exchange across layers. Finally, we combine sparse convolution to achieve efficient high-resolution input, providing richer geometric information of the instances. Compared to the baseline network, on the COCO benchmark dataset and the popular UAV dataset VisDrone, both of which contain a large number of small instances, the proposed method improves the detection accuracy mAP values by 1.1% and 2.2%, respectively, while having a good real-time detection speed.

Abstract Background The duration of treatment (DOT) of the initial intervention and subsequent treatment is the key to determining the accuracy of anticancer-drug budget impact analysis (BIA) calculations. However, existing studies only use simple assumptions as a proxy for DOT, resulting in a high degree of bias. Objectives To enhance the accuracy and reliability of anticancer-drug BIA and solve the problem regarding DOT, we propose an alternative individual patient data (IPD)-based approach that reconstructs IPD from the published Kaplan Meier survival curves to estimate DOT. Methods We developed a four-step methodological framework for this new approach, taking the use of pembrolizumab in treating microsatellite-instability–high (MSI-H) advanced colorectal cancer as an example: (1) reconstructing the IPD; (2) calculating the total DOT of the initial intervention and subsequent treatment for each patient; (3) assigning a randomized time and DOT; and (4) multiple replacement sampling and calculation of the mean value. Results Using this approach, the average DOT for the initial intervention and subsequent treatment in each year of the BIA time horizon can be calculated and used to calculate the resources consumed and costs in each year. In our example, the average DOT for the initial intervention with pembrolizumab from the first to the fourth year was 4.90, 6.60, 5.24, and 5.06 months, respectively, while the average DOT for subsequent treatment was 0.75, 2.84, 2.99, and 2.50 months, respectively. Conclusions The reconstructed IPD-based approach can improve the accuracy and reliability of anticancer-drug BIA compared with conventional methods, and can be widely used, especially for anticancer drugs with excellent efficacy.

PurposeIschemic stroke (IS) has a considerable impact on the health-related quality of life (HRQoL) of patients. A systematic review was conducted to summarize and synthesize the HRQoL reported from IS patients.MethodsAn electronic search was performed in PubMed, Web of Science, ScienceDirect, Embase, and Cochrane Library databases from inception to February 2022 for studies measuring utility values in IS patients. Basic information about the studies, patient characteristics, measurement of the utility values, and utility values were extracted and summarized. Utility values were pooled according to the time of evaluation, and disease severity was classified with modified Rankin Scale (mRS) scores. The quality of the studies was assessed according to key criteria recommended by the National Institute for Health and Care Excellence.ResultsA total of 39 studies comprising 30,853 participants were included in the study. Measured with EQ-5D-3L, the pooled utility values were 0.42 [95% confidential interval (CI): 0.13 to 0.71], 0.55 (95% CI: 0.43 to 0.68), 0.65 (95% CI: 0.52 to 0.78), 0.60 (95% CI: 0.43 to 0.78), and 0.67 (95% CI: 0.60 to 0.74) for patients diagnosed with IS within 1, 3, 6, 12, and 24 months or above among poststroke patients. Four studies reported utility values classified by mRS scores where synthesized estimates stratified by mRS scores ranged from 0.91 (95% CI: 0.85 to 0.97) for patients with an mRS score of 1 to−0.04 (95% CI:−0.18 to 0.11) for those with an mRS score of 5. As for the health dimension profiles, usual activity was the most impacted dimension, while self-care was the least impacted one.ConclusionThis study indicated that the utility values in IS patients kept increasing from stroke onset and became relatively stabilized at 6 months poststroke. Health utility values decreased significantly as mRS scores increased. These results facilitate economic evaluations in utility retrieval and selection. Further exploration was required regarding the factors that affect the HRQoL of IS patients.

Visual object detection is a computer vision technique based on deep learning. However, due to privacy issues, it is highly challenging to build an object detection model according to the current methods on the centrally stored training dataset. Federated learning is an effective approach to solving the challenge of training data collection by using distributed training. We propose FedVisionBC, a blockchain-based federated learning system for visual object detection that addresses the challenges of single point of failure, model poisoning attacks, and membership inference attacks in traditional federated learning. In the FedVisionBC system, we set up an aggregation node and a verification node instead of a central server to solve the single point of failure problem. We establish a security mechanism that uses encryption techniques, verification nodes, and smart contracts to resist model poisoning attacks. Experimental results show that FedVisionBC can accomplish the object detection task when the percentage of malicious clients is less than 60%. We also propose a new algorithm, ADPFedAvg, to prevent membership inference attacks, which relies on user-level differential privacy technology and the federated average algorithm. Experimental results indicate that ADPFedAvg can achieve a large-scale visual object detection model with differential privacy protection, while only a negligible cost in predictive accuracy.

ObjectiveThe aim of this study was to investigate the cost-effectiveness of serplulimab versus regorafenib in previously treated unresectable or metastatic microsatellite instability-high (MSI-H)/deficient mismatch repair (dMMR) colorectal cancer in China.MethodsFrom the perspective of China’s health-care system, a Markov model with three health states (progression free, progression, death) was developed for estimating the costs and health outcomes of serplulimab and regorafenib. Data for unanchored matching-adjusted indirect comparison (MAIC), standard parametric survival analysis, the mixed cure model, and transition probabilities calculation were obtained from clinical trials (ASTRUM-010 and CONCUR). Health-care resource utilization and costs were derived from government-published data and expert interviews. Utilities used to calculate quality-adjusted life years (QALYs) were obtained from clinical trials and literature reviews. The primary outcome was the incremental cost-effectiveness ratio (ICER) expressed as cost/QALY gained. Four scenarios were considered in scenario analysis: (a) using original survival data without conducting MAIC; (b) limiting the time horizon to the follow-up time of the clinical trial of serplulimab; (c) adopting a fourfold increase in the risk of death; and (d) applying utilities from two other sources. One-way sensitivity analysis and probabilistic sensitivity analysis were also performed to assess the uncertainty of the results.ResultsIn the base-case analysis, serplulimab provided 6.00 QALYs at a cost of $68,722, whereas regorafenib provided 0.69 QALYs at a cost of $40,106. Compared with that for treatment with regorafenib, the ICER for treatment with serplulimab was $5,386/QALY, which was significantly lower than the triple GDP per capita of China in 2021 ($30,036), which was the threshold used to define the cost-effectiveness. In the scenario analysis, the ICERs were $6,369/QALY, $20,613/QALY, $6,037/QALY, $4,783/QALY, and $6,167/QALY, respectively. In the probabilistic sensitivity analysis, the probability of serplulimab being cost-effective was 100% at the threshold of $30,036/QALY.ConclusionCompared with regorafenib, serplulimab is a cost-effective treatment for patients with previously treated unresectable or metastatic MSI-H/dMMR colorectal cancer in China.

Objective: We conducted a large-scale meta-analysis and subgroup analysis to compare the effect of fixed-dose combination (FDC) therapy with that of free-equivalent combination (FEC) therapy on medication adherence.Methods: Studies published in Web of Science, PubMed, Cochrane Library, ScienceDirect, and Embase up to May 2022 were identified according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The primary assessed outcomes were the medication possession ratio (MPR) and proportion of days covered (PDC). We investigated the probability of being adherent to the prescribed treatment (MPR or PDC ≥80%) or the average estimate of these two parameters. Studies reporting such results were included in this meta-analysis. The summary measures were reported as the risk ratio (RR) and the weighted mean difference (MD) with 95% of confidence interval (CI) using the random-effects model of DerSimonian and Laird. The quality of the cohort studies was assessed using the Newcastle-Ottawa scale.Results: Of the 1,814 screened studies, 61 met the predefined inclusion criteria. The meta-analysis of the results showed that compared to FEC, FDC significantly improved the medication compliance of patients by 1.29 times (95% CI:1.23–1.35, p < 0.00001). I2 of 99% represent high heterogeneity across studies. The mean difference in medication adherence between FDC and FEC was 0.10 (95% CI: 0.06–0.14, p < 0.00001) with an I2 estimate of 100%. Subgroup analyses were performed for studies that reported adherence outcomes according to disease type, period of evaluation and compliance indicators. A sensitivity analysis was conducted to exclude the results of low-quality studies, as well as studies in which there was ambiguity in the method of calculating the estimator.Conclusion: Analysis of the assessed parameters for the intention-to-treat and subgroup populations suggests that FDC can improve adherence to treatment and its advantages over FEC may increase over time. Further research is needed to better understand how medical conditions affect the impact of reduced pill burden on adherence, particularly in diseases other than cardiovascular disease and type 2 diabetes mellitus.

At present, most of the perceptual hash methods for image copyright protection rely on manually designed feature extraction and mapping, whose detection accuracy is insufficient. Some schemes based on deep learning are designed to consider a limited variety of content retention operations, which are not enough to deal with the increasingly severe situation of image copyright protection. In response to this situation, a novel Convolutional Neural Network (CNN)-based perceptual image hashing scheme is introduced in this paper. In this scheme, the training images are classified according to their original images and a Hadamard matrix is used to generate a hash center for each class in Hamming space. Then the Convolution Neural Network learns the feature extraction process of the image automatically, constrained by central quantization and distinct quantization, so that the hash code of each image converges to the hash center of their class, and generates the final hash sequence. The proposed scheme can successfully strike a balance between perceived robustness and discrimination capacity. Based on the test results on large-scale test sets, $F_{1}$ scores, equal error rate (EER) and receiver operating characteristic (ROC) curves demonstrate the superiority of our scheme compared with some state-of-the-art schemes.

A blockchain voting system based on the feedback mechanism and Wilson score is proposed to solve the problem of the malicious votes behavior. Firstly, the relatively accurate supporting rate and ranking for candidates are obtained using the Wilson score. Secondly, different feedback coefficients are calculated according to the above parameters. Finally, the account points are adjusted according to the feedback coefficients. The feedback mechanism is designed in the voting smart contract, and the smart contract is deployed on the blockchain to ensure the enforcement of the feedback mechanism. A fully functional smart contract is designed and briefly verified in this paper. The experiment is conducted under the K-out-of-L type of weighted voting. Experimental results show that the Wilson score can accurately modify the candidates’ supporting rates, and the feedback mechanism can effectively suppress malicious votes.