This paper surveys the emerging field of hybrid quantum–classical (HQC) architectures for medical image analysis, with a particular focus on COVID-19 classification as a rep-resentative application. Despite extensive progress in quantum machine learning, there is still a lack of structured taxonomy-driven analysis that systematically characterizes how quantum components are integrated within hybrid imaging pipelines. This study pro-vides a structured review of representative HQC models, categorizing existing approaches based on the functional role of quantum modules, including front-end processing, feature extraction, and classification layers. A comparative analysis is conducted to examine how these architectural design choices relate to reported performance across different ex-perimental settings. The review shows that hybrid models can achieve strong binary classification performance under simulation conditions, particularly when quantum cir-cuits are used as feature extractors. However, performance in multi-class scenarios remains inconsistent and highly sensitive to encoding strategies, circuit design, and optimization settings. In addition, most existing studies rely on simulator-based evaluations, limited or single-source datasets, and non-standardized benchmarking protocols, which restrict reproducibility and generalization. Real-hardware validation remains scarce, and no consistent evidence of task-level quantum advantage over optimized classical baselines has been demonstrated. Overall, current HQC-based approaches should be considered exploratory rather than clinically validated solutions. Future progress requires stand-ardized benchmarking, multi-institutional validation, hardware-aware evaluation, and the development of interpretable hybrid architectures that respect NISQ-era constraints.
This paper reviews hybrid quantum–classical (HQC) architectures for COVID-19-related respiratory medical-image analysis. To address the heterogeneity of existing studies, we propose an architecture-centric taxonomy based on the functional role and placement of the quantum module. Reviewed models are grouped into three archetypes: Archetype A, where quantum circuits act as patch-level quanvolutional preprocessors; Archetype B, where classical feature extractors are coupled with quantum classifier heads; and Archetype C, where quantum circuits generate intermediate features for downstream classical classifiers. Ten peer-reviewed journal studies were selected through a PRISMA-inspired search and analyzed across architecture, diagnostic performance, quantum resource reporting, validation rigor, computational scalability, and deployment feasibility. The review shows that HQC models often report promising binary COVID-19 screening results on CT or chest X-ray images, but multiclass respiratory classification remains less stable. Key limitations include simulator-dominated evaluation, limited external validation, unclear patient-wise splitting, incomplete reporting of qubit counts, circuit depth, and shots, and insufficient comparison with strong classical baselines. Overall, current HQC models should be viewed as exploratory quantum-augmented classical pipelines rather than clinically validated diagnostic systems. No conclusive task-level quantum advantage has yet been demonstrated for COVID-19 medical imaging. Future progress requires standardized benchmarking, transparent quantum-resource reporting, patient-wise and multi-center validation, hardware-aware evaluation, and interpretable hybrid designs compatible with NISQ-era constraints.
Deep neural networks (DNNs) have demonstrated remarkable performance in analyzing 3D point cloud data. However, their vulnerability to adversarial attacks-such as point dropping, shifting, and adding-poses a critical challenge to the reliability of 3D vision systems. These attacks can compromise the semantic and structural integrity of point clouds, rendering many existing defense mechanisms ineffective. To address this issue, a defense strategy named KNN-Defense is proposed, grounded in the manifold assumption and nearest-neighbor search in feature space. Instead of reconstructing surface geometry or enforcing uniform point distributions, the method restores perturbed inputs by leveraging the semantic similarity of neighboring samples from the training set. KNN-Defense is lightweight and computationally efficient, enabling fast inference and making it suitable for real-time and practical applications. Empirical results on the ModelNet40 dataset demonstrated that KNN-Defense significantly improves robustness across various attack types. In particular, under point-dropping attacks-where many existing methods underperform due to the targeted removal of critical points-the proposed method achieves accuracy gains of 20.1%, 3.6%, 3.44%, and 7.74% on PointNet, PointNet++, DGCNN, and PCT, respectively. These findings suggest that KNN-Defense offers a scalable and effective solution for enhancing the adversarial resilience of 3D point cloud classifiers. (An open-source implementation of the method, including code and data, is available at https://github.com/nimajam41/3d-knn-defense).
Retinal vessel segmentation remains difficult along thin, low-contrast branches. We evaluate a conditional diffusion refiner for probability maps from a lightweight U-Net with reverse attention (LU-Net+RA). A frozen mask autoencoder produces a spatially downsampled multi-channel grid; a global RGB embedding conditions denoising diffusion implicit model (DDIM) updates, followed by frozen decoding. The refiner adds 527,233 parameters to the 1,946,961-parameter backbone. Matched evaluation on CHASE-DB1, DRIVE and HRF uses five training seeds and five sampling seeds. Refined mean Dice was 0.746, 0.790 and 0.666, against coarse values of 0.811, 0.792 and 0.732. Exploratory paired intervals exclude zero in the negative direction. Conditioning variants changed Dice by less than 0.0002. Deterministic encoded-grid refinement and autoencoder fine-tuning narrowed the gap but exceeded the baseline only on DRIVE; spatial residual refinement changed Dice by −0.0001, +0.0058 and +0.0074. Matched out-of-fold training did not establish absence of input shift. Increasing DDIM updates from 50 to 100 doubled refinement time with negligible observed Dice change. An earlier single-run result was not reproduced. Under this controlled protocol, diffusion refinement did not improve LU-Net+RA.