The widespread use of medical imaging in telemedicine and EHRs demands robust watermarking that preserves diagnostic quality. Conventional spread spectrum methods, despite their robustness, are limited by shared secret keys, geometric vulnerabilities, and weak resilience to generative AI attacks. This paper proposes a spread spectrum discrete wavelet transform (DWT) watermarking framework for medical images, which uses a deep perceptual masking network (JNDnet) to place the watermark where it remains invisible, a lightweight CNN to adaptively select resilient sub‑bands, and a neural detector that replaces fixed‑threshold correlation for improved extraction accuracy while remaining blind. Experiments on three medical datasets show imperceptibility (PSNR > 44 dB, SSIM > 0.98) and robust performance against common and generative AI attacks, with bit error rates below 6

Dynamic Facial Expression Recognition (DFER) in unconstrained video is challenging because a model must jointly capture per-frame facial appearance and the temporal evolution of expression under head-pose variation, occlusion, and low image quality. State-of-the-art accuracy on the DFEW benchmark is dominated by heavy Transformer or self-supervised models (85–150M parameters), which are impractical on the 8 GB consumer GPUs available to most researchers. We present LiteDFER-Hybrid, a compact hybrid CNN–Transformer that couples an ImageNet-pretrained EfficientNet-B0 spatial encoder, a TinyViT spatial-attention block over per-frame patch tokens, and a Transformer-based Temporal Sequence Encoder (TSE) with a learnable classification token. With only 7.5M parameters and 7.24 GFLOPs per clip, LiteDFER-Hybrid attains a mean UAR of 56.12% and WAR of 65.96% on DFEW under the official five-fold cross-validation protocol. This exceeds the seminal Former-DFER Transformer (UAR 53.69%, 18M parameters) and the Expression Snippet Transformer (UAR 53.43%, 43M parameters) on both metrics, while using 2.4×–5.7× fewer parameters, placing LiteDFER-Hybrid on the accuracy–efficiency Pareto front for sub-10M-parameter DFER. We additionally report a systematic negative result: five distinct strategies for fusing RGB with TV-L1 optical flow — naive joint training and four constrained two-stage variants — all regressed below the RGB-only baseline on in-the-wild DFEW (-2.6 to -5.7 percentage points UAR). We analyse this failure, identifying weight-sharing distribution conflict and nuisance in-the-wild motion as root causes. This reproducible cautionary finding, contrasting with optical flow’s success in controlled micro-expression settings, redirects future work on motion modelling for in-the-wild DFER.

Medical image watermarking must ensure data authenticity and integrity without compromising diagnostic quality. Conventional spread spectrum methods suffer from key dependency, geometric vulnerability, hand-crafted perceptual models, and generative AI susceptibility. This paper presents a hybrid DWT-based spread spectrum framework for medical imaging with three innovations: a learned perceptual masking network (JND_net) that replaces analytical JND models, a lightweight CNN (SubBandSelector) that dynamically selects resilient wavelet sub-bands, and a neural detector (DetectorNet) that supplants fixed-threshold detection, all integrated with dual-key security (2128 key space) and MAC authentication. Experiments on 1,200 medical images with radiologist validation (51.3% detection accuracy, chance level; diagnostic confidence unchanged, p = 0.34) show bit error rates below 0.03 under JPEG, noise, and filtering, outperforming seven baselines (p < 0.001). Under generative AI attacks, BER reaches 0.112 (diffusion) and 0.087 (inpainting). The framework operates in real time (3.7 ms on T4 GPU, 25 ms on CPU) under blind extraction.

This paper introduces a robust and efficient digital image watermarking technique designed for precise tamper detection and localization. The proposed approach integrates Singular Value Decomposition (SVD) and QR decomposition to generate two distinct authentication bits (Au1 and Au2), ensuring enhanced imperceptibility and security. These authentication bits are embedded into the cover image (CI) using a novel Scale-based Embedding Factor Quantization Index Modulation (SEF-QIM) technique. This embedding strategy guarantees resilience against various challenges, including geometric distortions, compression, and noise. Comprehensive experimental evaluations conducted on grayscale images demonstrate exceptional imperceptibility, achieving peak signal-to-noise ratio (PSNR) values exceeding 54 dB and 52 dB for scaling factors of 2 and 4, respectively. Additionally, the proposed method exhibits strong robustness against common attacks such as Gaussian noise, cropping, and rotation. The computational efficiency of the scheme ensures reduced time complexity for both embedding and extraction processes, making it a reliable and secure solution for image authentication in high-integrity digital media applications.

Secure authentication and traceability of medical audio data remain critical challenges in modern telemedicine systems and digital health record management.. This paper proposes a novel blind and robust audio watermarking scheme for medical applications. The method combines the Fractional Charlier Transform (FrCT) for optimized time-frequency decomposition, local entropy analysis with critical-band masking for intelligent coefficient selection, and adaptive dithered quantization index modulation (ADQIM) for imperceptible watermark embedding. The proposed scheme provides comprehensive encryption of metadata including patient information and acquisition context through AES-based cryptographic mechanisms, while maintaining imperceptibility and embedding robustness. Comprehensive experimental validation on a diverse medical audio corpus demonstrates that the method achieves a practical payload capacity of 71.8 bits per second, high audio transparency with an SNR of 38.2 dB and a PESQ score of 4.15, and strong resilience against various signal processing attacks with an average BER of 3.2 %. The approach provides a computationally efficient solution suitable for integration into operational telemedicine platforms and large-scale medical archiving systems, offering reliable authentication and integrity verification of medical audio records.

The exponential growth of telemedicine and digital health platforms has introduced serious challenges in maintaining the confidentiality, authenticity, and diagnostic integrity of medical images transmitted over insecure networks. This study specifically addresses these challenges by developing a blind and imperceptible watermarking architecture that ensures both data privacy and image reliability. The proposed method integrates four complementary techniques-Non-Subsampled Shearlet Transform (NSST) for multiscale feature extraction, QR decomposition for numerically stable embedding, Particle Swarm Optimization (PSO) for adaptive block selection, and Grad-CAM attention maps for perceptual guidance. Together, these components solve three longstanding issues in medical image protection: (1) preserving diagnostic quality while embedding sensitive data, (2) achieving robustness against signal and geometric distortions without reference to the original image, and (3) reducing computational complexity for real-time telemedicine integration. The watermark encodes both compressed patient metadata and biometric images using BCH error correction and XOR encryption. Experiments on colorized CT and X-ray datasets show high imperceptibility (PSNR = 45.21 dB, SSIM = 0.9864), strong robustness (NCC >= 0.897), and fast runtime (approximate to 2 s per image), confirming the method's suitability for secure and practical clinical deployment.

This paper proposes a volumetric blind watermarking framework for 3D medical images based on adaptive DWT-SVD embedding with multi-slice fusion. The method dynamically adjusts embedding strength per slice using two complementary criteria: entropy (to exploit perceptual masking in textured regions) and anatomical position (to prioritize centrally located slices that are less likely to be cropped). A content-dependent chaotic encryption, initialized from the SHA-256 hash of the entire volume, secures the watermark before embedding into the low-frequency DWT subbands via singular value modification. During extraction, a multi-slice fusion mechanism aggregates watermark estimates from all slices using positional weights, ensuring robust recovery even under localized attacks. Experiments on 20 brain MRI and 15 chest CT volumes (256 & times;256 & times; up to 160 slices) demonstrate high imperceptibility (average PSNR = 44.8dB for MRI, 43.6dB for CT; SSIM >0.997) and strong robustness (average BER = 0.044 across 15 distinct attack types including JPEG compression, Gaussian noise, median filtering, rotation, scaling, cropping, and slice drop-out). Compared to nine state-of-the-art methods, including transform-based, feature-based, and deep learning approaches, the proposed framework achieves the lowest average BER (0.044 versus 0.063 for the closest competitor MCANet) while maintaining competitive imperceptibility and requiring only CPU-based computation (approximate to 1.3 seconds per volume). These results position the proposed method as a practical, secure, and clinically viable solution for protecting patient data in telemedicine and PACS environments.

Protecting medical images in AI‑driven healthcare workflows is challenging because conventional watermarks are destroyed by routine deep learning operations such as classification, segmentation, or denoising. In this paper, a novel framework called AI‑Pipeline Persistent Latent Watermarking (APPL‑WM) is proposed. A binary watermark is embedded into multi‑scale latent features extracted by a hybrid encoder, and an invertible neural network guarantees perfect reversibility of the original image. A differentiable simulator of a realistic medical AI pipeline (CNN classifier, U‑Net segmenter, diffusion denoiser, and JPEG compression) is employed during training to enforce robustness. Experimental results on three medical imaging datasets show that the proposed method achieves high imperceptibility (PSNR of 42.35 dB, SSIM of 0.991) and retains 94.22 %-bit accuracy after the full AI pipeline, significantly outperforming state‑of‑the‑art methods (which drop below 72 %). The framework provides a robust and reversible solution for copyright protection and integrity verification of medical images processed by AI systems.

The increasing digitization of data offers significant advantages, particularly in digital images, which continue to grow rapidly due to advances in image acqui sition systems. However, this transition also presents significant security risks, necessitating robust methods to protect digital images from illegal manipulation. While existing watermarking approaches for medical imaging often struggle to balance robustness, computational efficiency, and diagnostic integrity, this study introduces a robust blind watermarking technique that integrates cryptographic hashing, deep learning, and frequency-domain embedding to address these lim itations. The methodology leverages the Haar wavelet transform to embed a binary watermark, generated from personal information using the secure hashing algorithm SHA-256, into the HH2 sub-band of the red channel. A convolutional neural network-based model optimized watermark insertion and facilitated blind extraction. Comprehensive evaluations across diverse medical imaging modalities, including MRI, ultrasound, and CT scans, demonstrate excellent performance with peak signal-to-noise ratio (PSNR) values exceeding 41 dB achieving up to 43.13 dB for fundus images, representing a 3.1% improvement over the best exist ing method and structural similarity (SSIM) values around 0.94. The proposed method shows significant speed improvements of approximately 75% compared to state-of-the-art techniques, with embedding and extraction times of 0.4430 sec onds and 0.1800 s, respectively, compared to 1.5 s total processing time of recent deep learning methods. These findings underscore the technique’s potential for telemedicine applications, where maintaining high image quality and security is paramount for accurate diagnosis.

The Flexible Representation of Quantum Images (FRQI) is widely used for encoding images into normalized quantum states for secure processing. However, it suffers from high computational and qubit complexity, limiting its scalability. In addition, quantum comparator models, often employed for qubit comparison, remain vulnerable to geometric attacks such as scaling and rotation and require a slow embedding and extraction rate. To address these challenges, this paper proposes a novel quantum watermarking technique that integrates Quantum Image Representation (QIR) with the Novel Enhanced Quantum Representation (NEQR), offering improved efficiency, robustness, and security for quantum image processing (QIP). The proposed technique initiates watermark qubit (QB) generation by integrating pixel position bits (PPB) with an arithmetic average (AA)-based logic, implemented through a quantum circuit comprising reverse parallel adder (RPA), dividing-4 (D4) models, and controlled-NOT (CNOT) gates. For watermark embedding, the proposed circuit employs RPA, quantum equal (QE) models, CNOT gates, and Pauli-X gates, embedding the watermark via a least significant bit (LSB) strategy within the NEQR-encoded image. The extraction process mirrors the embedding architecture, utilizing reverse parallel subtractor (RPS), QE models, CNOT, and Pauli-X gates to ensure accurate watermark retrieval. Experimental results demonstrate high robustness by achieving the highest peak-signal-to-noise-ratio(PSNR) values of 59.97 dB (dB) along with a structural similarity index matrix (SSIM) of 0.9999. The proposed framework is validated using IBM Qiskit-based quantum circuit simulation on classical computing hardware and is not intended to demonstrate fault-tolerant quantum hardware execution or provable quantum computational advantage. Rather, the work focuses on simulation-level reversible quantum-circuit modelling for secure image watermarking using NEQR-based image representation and quantum-compatible reversible logic structures. Furthermore, the proposed technique is subjected to a range of image processing attacks, quantum noise and decoherence to rigorously evaluate its resilience and performance. In addition, an AI-assisted attack classification module is incorporated to automatically identify the type of distortion applied to the attacked watermarked images. A lightweight Random Forest classifier is trained to classify different attack categories, including no attack, Gaussian noise, salt-and-pepper noise, median filtering, histogram equalization, JPEG compression, cropping, rotation, and text addition. The proposed AI-assisted module achieves an overall classification accuracy of 96.56%, demonstrating its effectiveness in automatic attack identification and further strengthening the applicability of the proposed quantum watermarking framework for secure image authentication. Resource analysis shows that the watermark generation circuit requires 354 QBs (including 320 for RPA, 10 for D4, and 24 for CNOT), while both embedding and extraction consume 400 QBs each. Theoretical evaluation and empirical

Watermarking proves to be an effective technique for safeguarding crucial medical information. In this research, we propose a robust and imperceptible watermarking method designed to enhance the security of telemedicine-transmitted medical electrocardiogram (ECG) data. Embedding a mark in medical ECGs enables precise patient identification, reduces the risk of confusion during scans, and helps prevent diagnostic errors that could have adverse consequences. To ensure the security of ECG signals exchanged in telemedicine, our approach involves a frequency-domain watermarking method that conceals electronic patient records within the corresponding ECG signals. In this methodology, the signal undergoes a conversion into a 2D image, followed by a three-layer transform to extract the frequency content of the medical image. The low-frequency subbands undergo Schur decomposition, and the watermark bits are subsequently incorporated into the values of the upper triangular matrix. According to experimental results, these proposed techniques maintain a significant level of watermarked ECG quality while demonstrating high resistance to standard attacks. Experimental results show that the proposed SWT–Schur-based watermarking scheme achieves an average PSNR of 44.56 dB and an NCC higher than 0.95 under most common signal processing attacks. The average embedding capacity is 0.27 bits per pixel (BPP), while preserving the diagnostic quality of the ECG signals.

Nowadays, modern medical information systems enable medical data, including medical images, to be transferred quickly and easily, making healthcare more accessible and efficient. However, this advancement also introduces significant security risks, necessitating robust methods to protect images from unauthorized manipulation. While existing watermarking approaches for medical imaging often struggle to balance robustness, computational efficiency, and diagnostic integrity, this study introduces a robust blind watermarking technique that integrates cryptographic hashing, deep learning, and frequency-domain embedding to address these limitations. The methodology leverages the Haar wavelet transform to embed a binary watermark, generated from personal information using the Secure Hash Algorithm (SHA-256), into the HH2 sub-band of the red channel. A convolutional neural network-based model was employed to optimize watermark insertion and enable blind extraction. Extensive evaluations across various medical imaging modalities confirm the effectiveness of the proposed watermarking scheme, with PSNR values exceeding 41 dB and SSIM around 0.94. Compared to existing methods, it achieves a 3.11% improvement in PSNR, along with 41.23% faster embedding and extraction times. Under no-attack conditions, the system attains a bit error rate (BER) of 0 and a normalized correlation (NC) value of 1, indicating perfect watermark recovery, with a 0.05% improvement in NC observed under attack scenarios. These results highlight the method’s suitability for telemedicine applications, where image quality, speed, and security are critical.

Medical deepfakes and adversarial manipulations threaten AI-based diagnosis and telemedicine security. Existing watermarking methods for image authentication do not reliably distinguish clinically neutral transformations from semantic tampering, offer no interpretability of detected alterations, and often exceed latency constraints of clinical workflows. We propose Proactive Forensic Fragile Watermarking (PFF-WM), a framework that embeds two complementary watermarks: a fragile watermark in wavelet detail coefficients (pixel-level sensitivity) and a semi-robust watermark in the DCT domain whose embedding strength is modulated by a multi-scale attention map that prioritises diagnostically relevant regions. At the receiver side, a stacked autoencoder trained exclusively on authentic images detects manipulations via reconstruction error, a lightweight refinement network produces a tamper localisation mask, and a gradient-based explainability layer estimates the clinical impact of any alteration. Experiments on CheXpert (resampled to 512 & times; 512), LiTS, and ISIC 2019 show that PFF-WM achieves 97.6% detection accuracy and an AUC of 0.989, with tamper localisation IoU of 83.2%, with a false positive rate below 0.5% under each tested non-geometric benign transformation (JPEG, resizing, contrast, blur) and below 0.7% under chained non-geometric transformations. Geometric transformations (rotation, translation) are a recognised limitation, reaching 18.7% FPR at 15 degrees rotation.The method shows competitive or superior performance against existing watermarking-based forensic methods under the tested conditions, although direct comparability is limited for methods designed for different resolutions or modalities. Inference time is 44 ms per 512 & times; 512 image, making it computationally feasible for real-time verification under the tested conditions.

The rapid integration of digital technologies into modern healthcare has led to an unprecedented exchange of medical imaging data across clinical and remote platforms, raising critical concerns about patient privacy and data exposure. While conventional encryption techniques ensure secure storage and transmission, protected images become fully vulnerable once decrypted. Moreover, many existing protection schemes fail to guarantee faithful reversibility of the original diagnostic content. We propose a key-controlled, entropy-guided Gaussian perturbation framework for reversible privacy preservation. The method performs local entropy analysis to identify sensitive regions, then injects spatially adaptive Gaussian noise using cryptographically secure pseudo-random sequences from a 256-bit key. High-entropy pathological structures receive amplified perturbation while anatomical contexts are preserved. Exact reversibility is achieved through deterministic inversion. Experiments on ChestX-ray14, BraTS 2021, and OCTID demonstrate: imperceptibility (PSNR 41.8 dB, SSIM 0.971), protection against ResNet-50/U-Net (ASR 8.2%, SRR 28.5%), near-exact reconstruction (MAE 0.14), and real-time processing (57 ms/image). Cross-dataset generalization and JPEG compression robustness confirm practical viability. This training-free method enables secure telemedicine and research data sharing without compromising diagnostic integrity.

Protecting medical images in interconnected healthcare systems requires maintaining both diagnostic integrity and secure verification. This study presents a watermarking framework combining deep feature extraction, chaotic cryptography, adaptive frequency-domain embedding, and AI-assisted extraction. Deep convolutional networks identify perceptually tolerant regions for content-aware embedding while preserving vital diagnostic areas. Hybrid chaos-based encryption secures watermark data against unauthorized recovery. The embedding operates in the fractional discrete cosine transform (FDCT) domain with adaptive coefficient selection and multi-objective optimization balancing imperceptibility, robustness, and capacity. During extraction, a convolutional autoencoder refines recovered watermarks, maintaining fidelity under compression, noise, and geometric distortions. Experimental validation on medical datasets demonstrates high embedding capacity (0.07309 BPP), excellent visual similarity (PSNR = 46.85 dB, SSIM = 0.9995), and strong resilience against attacks (NCC ≥ 0.94), ensuring secure and compliant medical image transmission.

Securing and tracing medical audio data is crucial in telemedicine and digital archiving. This paper presents a blind and irreversible audio watermarking scheme designed to satisfy imperceptibility, robustness, and embedding capacity requirements for sensitive medical applications. The method integrates the Fractional Charlier Transform (FrCT) for adaptive time-frequency analysis, local entropy analysis with the Watson perceptual model for intelligent coefficient selection, and adaptive logarithmic quantization index modulation (LQIM) for embedding. It securely incorporates patient and acquisition metadata, ensuring confidentiality and integrity via cryptographic and error-correction techniques. Experiments demonstrate a payload of 67.3 bits per second, high audio transparency (SNR > 36 dB, PESQ > 4.0), and robustness against various signal processing attacks (average BER 4.5

As medical imaging becomes increasingly central to telemedicine and electronic health records; robust integrity verification is required to preserve diagnostic quality while enabling precise tamper detection. Existing quantum image watermarking methods lack post-quantum key security and deliberate fragility. This paper proposes a quantum-inspired watermarking framework for medical images with post-quantum key security and deliberate fragility based on entangled DCT coefficients.Watermarks are carried by collective coefficient sums, making any alteration detectable. Post-quantum security is ensured through SHA3-256 key derivation. Experiments on three medical datasets (ChestX-ray14, BraTS, OCTID) show excellent imperceptibility (PSNR 44.2 dB, SSIM > 0.98), high tamper sensitivity (BER > 30%), and precise localization (accuracy > 80%). Unlike prior methods, the framework integrates post-quantum security, deliberate fragility, and tamper localization. Furthermore, a lightweight Random Forest classifier is trained on global and block-wise BER statistics to automatically distinguish intact images from those subjected to benign processing or malicious tampering, enhancing the system's practical utility as an AI-assisted integrity verification tool for medical images in future quantum-threat environments..

This work introduces a non-blind watermarking framework for color images to address tamper detection, particularly under geometric transformations. The proposed scheme fuses two watermarks, a personal signature and a biometric fingerprint, into a unified composite watermark embedded into the chrominance component of the cover image using a multi-level transform domain approach, discrete wavelet transforms (DWTs), discrete cosine transforms (DCTs), and singular value decomposition (SVD). By leveraging the rotation-invariant properties of scale-invariant feature transform (SIFT) and oriented FAST and rotated BRIEF (ORB) descriptors, the framework ensures robust tamper detection without requiring alignment, thus mitigating the limitations of conventional detection techniques vulnerable to transformation-induced tamper obfuscation (TITO). Extensive experimentation demonstrates that the method maintains high perceptual fidelity, achieving PSNR values ranging from 50 to 55 dB for embedding strength factor mu (0.01-0.04) and SSIM indices near 1 across multiple benchmark images. Furthermore, the scheme exhibits notable resilience to a range of image processing attacks and geometric distortion. Comparative evaluation reveals its superiority over existing grayscale, color, SIFT-based and DWT-DCT-SVD-based watermarking techniques, affirming its applicability in scenarios demanding secure, imperceptible, and transformation-invariant image watermarking.

With the growing reliance on telemedicine, ensuring the authenticity and confidentiality of medical images has become a critical challenge. This study presents a blind watermarking framework for medical images that balances imperceptibility, robustness, and computational efficiency. The proposed method combines Dual-Tree Complex Wavelet Transform (DT-CWT) for multi-directional decomposition, QR decomposition for numerically stable embedding, and Glowworm Swarm Optimization (GSO) for adaptive coefficient selection. To maintain visual fidelity, watermark insertion is restricted to the luminance channel and guided by a Human Visual System (HVS)-based quantization strategy. The embedded payload includes both compressed patient metadata and a biometric facial image, protected by BCH error correction and lightweight encryption. Experimental evaluations on a public dataset of RGB CT and X-ray images demonstrate strong imperceptibility (PSNR 44.67 dB, SSIM 0.9853), which is competitive with published values for comparable watermarking methods. The scheme achieves high robustness under noise, filtering, compression, and moderate geometric distortions (NCC >= 0.879), with competitive or improved robustness metrics (up to 15.3% higher NCC under certain attacks compared to published results), while maintaining efficient processing (1.32 s embedding, 1.08 s extraction per image). These results confirm the method's suitability for real-world telemedicine applications requiring secure and visually intact medical image transmission.

Federated Learning (FL) is a cutting-edge method in the medical imaging field that allows hospitals to collaboratively build models without revealing patient data. Nevertheless, FL is still vulnerable to numerous security and privacy issues, including, but not limited to, data poisoning, Byzantine attacks, and inference attacks. The existing literature has only partly dealt with this topic by focusing either on particular threats or on mitigation strategies, thus leaving the overall comprehension of the problems and their solutions in medical imaging as inadequate. The main threats to FL are systematically classified in this systematic review, with two major vulnerable assets, medical data and model parameters, being pointed out. We review existing countermeasures based on cryptographic techniques, secure aggregation, perturbation methods, and security protocols with an emphasis on their efficiency in ensuring patient privacy and model integrity. We also discuss the impact of FL in medical imaging, where it serves as a tool for privacy preservation and has the potential to improve diagnostic accuracy. Data heterogeneity, communication overhead, and lack of standardization are key challenges that are considered, as well as potential future research paths to explore for solving these problems. Overall, the systematic review reveals that, although federated learning provides enormous privacy-preserving benefits in medical imaging, its actual implementation needs to be very cautious regarding the merging of very strong security measures and the use of standard protocols so that the weaknesses are reduced, and the reliability of the diagnostics is increased.