This study introduces a novel approach for integrating sensitive patient information within medical images with minimal impact on their diagnostic quality. Utilizing the mask region-based convolutional neural network for identifying regions of minimal medical significance, the method embeds information using discrete cosine transform-based steganography. The focus is on embedding within "insignificant areas", determined by deep learning models, to ensure image quality and confidentiality are maintained. The methodology comprises three main steps: neural network training for area identification, an embedding process for data concealment, and an extraction process for retrieving embedded information. Experimental evaluations on the CHAOS dataset demonstrate the method's effectiveness, with the model achieving an average intersection over union score of 0.9146, indicating accurate segmentation. Imperceptibility metrics, including peak signal-to-noise ratio, were employed to assess the quality of stego images, with results showing high capacity embedding with minimal distortion. Furthermore, the embedding capacity and payload analysis reveal the method's high capacity for data concealment. The proposed method outperforms existing techniques by offering superior image quality, as evidenced by higher peak signal-to-noise ratio values, and efficient concealment capacity, making it a promising solution for secure medical image handling.
(© 2024. The Author(s).)

Classifying images has become a straightforward and accessible task, thanks to the advent of Deep Neural Networks. Nevertheless, not much attention is given to the privacy concerns associated with sensitive data contained in images. In this study, we propose a solution to this issue by exploring an intersection between Machine Learning and cryptography. In particular, Fully Homomorphic Encryption (FHE) emerges as a promising solution, as it enables computations to be performed on encrypted data. We therefore propose a Residual Network implementation based on FHE which allows the classification of encrypted images, ensuring that only the user can see the result. We suggest a circuit which reduces the memory requirements by more than [Formula: see text] compared to the most recent works, while maintaining a high level of accuracy and a short computational time. We implement the circuit using the well-known Cheon-Kim-Kim-Song (CKKS) scheme, which enables approximate encrypted computations. We evaluate the results from three perspectives: memory requirements, computational time and calculations precision. We demonstrate that it is possible to evaluate an encrypted ResNet20 in less than five minutes on a laptop using approximately 15[Formula: see text]GB of memory, achieving an accuracy of 91.67% on the CIFAR-10 dataset, which is almost equivalent to the accuracy of the plain model (92.60%).

The decoupling of control and forwarding layers brings Software-Defined Networking (SDN) the network programmability and global control capability, but it also poses SDN security risks. The adversaries can use the forwarding and control decoupling character of SDN to forge legitimate traffic, launching saturation attacks targeted at SDN switches. These attacks can cause the overflow of switch flow tables, thus making the switch cannot forward benign network traffic. How to effectively detect saturation attack is a research hotspot. There are only a few graph-based saturation attack detection methods. Meanwhile, the current graph generation methods may take useless or misleading information to the attack detection, thus decreasing the attack detection accuracy. To solve the above problems, this paper proposes TITAN, a bidirecTional forwardIng graph-based saturaTion Attack detectioN method. TITAN defines flow forwarding rules and topology information, and designs flow statistical features. Based on these definitions, TITAN generates nodes of the bi-forwarding graph based on the flow statistics features and edges of the bi-forwarding graph based on the network traffic routing paths. In this way, each traffic flow in the network is transformed into a bi-directional forwarding graph. Then TITAN feeds the above bidirectional forwarding graph into a Graph Convolutional Network (GCN) to detect whether the flow is a saturation attack flow. The experimental results show that TITAN can effectively detect saturation attacks in SDNs with a detection accuracy of more than 97%.
Competing Interests: The authors have declared that no competing interests exist.
(Copyright: © 2024 Ran et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.)

Pre-trained models such as BERT have made great achievements in natural language processing tasks in recent years. In this paper, we investigate the privacy-preserving pre-training based neural network inference in a two-server framework based on additive secret sharing technique. Our protocol allows a resource-restrained client to request two powerful servers to cooperatively process the natural processing tasks without revealing any useful information about its data. We first design a series of secure sub-protocols for non-linear functions used in BERT model. These sub-protocols are expected to have broad applications and of independent interest. Based on the building sub-protocols, we propose SecBERT, a privacy-preserving pre-training based neural network inference protocol. SecBERT is the first cryptographically secure privacy-preserving pre-training based neural network inference protocol. We show security, efficiency and accuracy of SecBERT protocol through comprehensive theoretical analysis and experiments.
Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(Copyright © 2024 Elsevier Ltd. All rights reserved.)

Anomaly detection (AD) plays a crucial role in various domains, including cybersecurity, finance, and healthcare, by identifying patterns or events that deviate from normal behavior. In recent years, significant progress has been made in this field due to the remarkable growth of deep learning models. Notably, the advent of self-supervised learning has sparked the development of novel AD algorithms that outperform the existing state-of-the-art approaches by a considerable margin. This paper aims to provide a comprehensive review of the current methodologies in self-supervised anomaly detection. We present technical details of the standard methods and discuss their strengths and drawbacks. We also compare the performance of these models against each other and other state-of-the-art anomaly detection models. Finally, the paper concludes with a discussion of future directions for self-supervised anomaly detection, including the development of more effective and efficient algorithms and the integration of these techniques with other related fields, such as multi-modal learning.
Competing Interests: Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Hadi Hojjati reports financial support was provided by Quebec Research Fund Nature and Technology. Narges Armanfard reports financial support was provided by Natural Sciences and Engineering Research Council of Canada. Hadi Hojjati reports financial support was provided by AGE-WELL NCE Inc.
(Copyright © 2024 Elsevier Ltd. All rights reserved.)

Objective: Develop and investigate the potential of a remote, computer-mediated and synchronous text-based triage, which we refer to as InSort, for quickly highlighting persons of interest after an insider attack.
Background: Insiders maliciously exploit legitimate access to impair the confidentiality and integrity of organizations. The globalisation of organisations and advancement of information technology means employees are often dispersed across national and international sites, working around the clock, often remotely. Hence, investigating insider attacks is challenging. However, the cognitive demands associated with masking insider activity offer opportunities. Drawing on cognitive approaches to deception and understanding of deception-conveying features in textual responses, we developed InSort, a remote computer-mediated triage.
Method: During a 6-hour immersive simulation, participants worked in teams, examining password protected, security sensitive databases and exchanging information during an organized crime investigation. Twenty-five percent were covertly incentivized to act as an 'insider' by providing information to a provocateur.
Results: Responses to InSort questioning revealed insiders took longer to answer investigation relevant questions, provided impoverished responses, and their answers were less consistent with known evidence about their behaviours than co-workers.
Conclusion: Findings demonstrate InSort has potential to expedite information gathering and investigative processes following an insider attack.
Application: InSort is appropriate for application by non-specialist investigators and can be quickly altered as a function of both environment and event. InSort offers a clearly defined, well specified, approach for use across insider incidents, and highlights the potential of technology for supporting complex time critical investigations.

Modern healthcare systems contain a large amount of sensitive information related to a patient in textual and visual form. Surgical videos and diagnosis data such as ultrasound, Computed Tomography (CT) scan, and Magnetic Resonance Imaging (MRI) are examples of healthcare video data. The secure storage and transmission of medical data have become an important issue in medical applications. To handle this challenge, chaos based cryptosystems are widely used these days. The work in this paper proposes a novel 2D Cosine-Sine map that exploits the existing Sine map and cosine transformation in its mathematical computation. The performance assessment of the suggested map indicates that it possesses a broader chaotic range, more dynamic and hyperchaotic nature when compared to existing chaotic maps. The proposed work, also, combines this novel 2D Cosine-Sine map with dynamic DNA encoding to propose a new scheme to encrypt medical videos. The approach consists primarily of four distinct phases. In the initial stage, the video is divided into frames, and for each frame, a chaotic sequence is generated using a 2D Cosine-sine map. During the second stage, we proceed to pixel permutation for each frame by utilizing the chaotic sequence. The third step is diffusion, in this, we integrate the 2D Cosine-Sine map with dynamic DNA encoding and apply double DNA operations to substitute the pixel values. The last step is DNA decoding to get the frames back in binary format. DNA encoding/decoding rules and operands of DNA operations are not fixed. They are selected dynamically using a random key for each frame. The dynamic selection of encoding/decoding rules and operands is a unique feature that enhances the scheme's security. Simulation results and security analysis proves that the proposed medical video encryption scheme can resist different types of attacks.
(© 2023. International Federation for Medical and Biological Engineering.)

The rapid replacement of PSTN with VOIP networks indicates the definitive phase-out of the PBX/PABX with smartphone-based VOIP technology that uses WLAN connectivity for local communication; however, security remains a key issue, regardless of the communication coverage area. Session initiation protocol (SIP) is one of the most widely adopted VOIP connection establishment protocols but requires added security. On the Internet, different security protocols, such as HTTPS (SSL/TLS), IPSec, and S/MIME, are used to protect SIP communication. These protocols require sophisticated infrastructure and some pose a significant overhead that may deteriorate SIP performance. In this article, we propose the following: i) avoid using Internet bandwidth and complex Internet protocols for local communication within an organization, but harness WLAN connectivity, ii) use multi-threaded or multicore computer systems to handle concurrent calls instead of installing hardware-based SIP servers, and iii) run each thread in a separate core. Cryptography is a key tool for securely transmitting confidential data for long- and short-range communication, and the Diffie-Hellman (DH) protocol has consistently been a popular choice for secret key exchanges. Primarily, used for symmetric key sharing, it has been proven effective in generating public/private key pairs, sharing public keys securely over public channels, and subsequently deriving shared secret keys from private/public keys. This key exchange scheme was proposed to safeguard VOIP communication within WLANs, which rely on the SIP for messaging and multimedia communication. For ensuring an efficient implementation of SIP, the system was rigorously analyzed using the M/M/1 and M/M/c queuing models. We analyze the behavior of SIP servers with queuing models with and without end-to-end security and increase users' trust in SIP security by providing a transparent sense of end-to-end security as they create and manage their private and public keys instead of relying on the underlying SIP technology. This research implements instant messaging, voice conversation, and secret key generation over DH while implementing and observing the role of multi-threading in multiqueue systems that serve incoming calls. By increasing the number of threads from one to two, the SIP response time improved from 20.23809 to 0.08070 min at an arrival rate of 4250 calls/day and a service rate of three calls/min. Similarly, by adding one to seven threads, the queue length was reduced by four calls/min. Implementing secure media streaming and reliable AES-based signaling for session confidentiality and integrity introduces a minor 8-ms tradeoff in SIP service performance. However, the advantages of implementing added security outweigh this limitation.
Competing Interests: The authors have declared that no competing interests exist
(Copyright: © 2024 Bhatti et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.)

Objective: Convolutional neural network (CNN), a classical structure in deep learning, has been commonly deployed in the motor imagery brain-computer interface (MIBCI). Many methods have been proposed to evaluate the vulnerability of such CNN models, primarily by attacking them using direct temporal perturbations. In this work, we propose a novel attacking approach based on perturbations in the frequency domain instead.
Methods: For a given natural MI trial in the frequency domain, the proposed approach, called frequency domain channel-wise attack (FDCA), generates perturbations at each channel one after another to fool the CNN classifiers. The advances of this strategy are two-fold. First, instead of focusing on the temporal domain, perturbations are generated in the frequency domain where discriminative patterns can be extracted for motor imagery (MI) classification tasks. Second, the perturbing optimization is performed based on differential evolution algorithm in a black-box scenario where detailed model knowledge is not required.
Results: Experimental results demonstrate the effectiveness of the proposed FDCA which achieves a significantly higher success rate than the baselines and existing methods in attacking three major CNN classifiers on four public MI benchmarks.
Conclusion: Perturbations generated in the frequency domain yield highly competitive results in attacking MIBCI deployed by CNN models even in a black-box setting, where the model information is well-protected.
Significance: To our best knowledge, existing MIBCI attack approaches are all gradient-based methods and require details about the victim model, e.g., the parameters and objective function. We provide a more flexible strategy that does not require model details but still produces an effective attack outcome.