We investigate the use of quantum computing algorithms on real quantum hardware to tackle the computationally intensive task of feature selection for light-weight medical image datasets. Feature selection is often formulated as a k of n selection problem, where the complexity grows binomially with increasing k and n. Quantum computers, particularly quantum annealers, are well-suited for such problems, which may offer advantages under certain problem formulations. We present a method to solve larger feature selection instances than previously demonstrated on commercial quantum annealers. Our approach combines a linear Ising penalty mechanism with subsampling and thresholding techniques to enhance scalability. The method is tested in a toy problem where feature selection identifies pixel masks used to reconstruct small-scale medical images. We compare our approach against a range of feature selection strategies, including randomized baselines, classical supervised and unsupervised methods, combinatorial optimization via classical and quantum solvers, and learning-based feature representations. The results indicate that quantum annealing-based feature selection is effective for this simplified use case, demonstrating its potential in high-dimensional optimization tasks. However, its applicability to broader, real-world problems remains uncertain, given the current limitations of quantum computing hardware. While learned feature representations such as autoencoders achieve superior reconstruction performance, they do not offer the same level of interpretability or direct control over input feature selection as our approach.

Reconstructing high-quality images from corrupted measurements remains a fundamental challenge in medical imaging. Recently, deep unrolling (DUN) methods have emerged as a promising solution, combining the interpretability of traditional iterative algorithms with the powerful representation capabilities of deep learning. However, their performance is often limited by weak information flow between iterative stages and a constrained ability to capture global features across stages-limitations that tend to worsen as the number of iterations increases.
Approach: In this work, we propose a memory-enhanced and multi-domain learning-based deep unrolling network for interpretable, high-fidelity medical image reconstruction. First, a memory-enhanced module is designed to adaptively integrate historical outputs across stages, reducing information loss. Second, we introduce a cross-stage spatial-domain learning transformer (CS-SLFormer) to extract both local and non-local features within and across stages, improving reconstruction performance. Finally, a frequency-domain consistency learning (FDCL) module ensures alignment between reconstructed and ground truth images in the frequency domain, recovering fine image details.
Main Results: Comprehensive experiments evaluated on three representative medical imaging modalities (PET, MRI, and CT) show that the proposed method consistently outperforms state-of-the-art (SOTA) approaches in both quantitative metrics and visual quality. Specifically, our method achieved a PSNR of 37.835 dB and an SSIM of 0.970 in 1 $\%$ dose PET reconstruction.
Significance: This study expands the use of model-driven deep learning in medical imaging, demonstrating the potential of memory-enhanced deep unrolling frameworks for high-quality reconstructions.

Denoising reconstructed Computed Tomography (CT) images without access to raw projection data remains a significant difficulty in medical imaging, particularly when utilizing sharp or medium reconstruction kernels that generate high-frequency noise. This work introduces an innovative method that integrates quaternion mathematics with bilateral filtering to resolve this issue. The proposed Quaternion Bilateral Filter (QBF) effectively maintains anatomical structures and mitigates noise caused by the kernel by expressing CT scans in quaternion form, with the red, green, and blue channels encoded together. Compared to conventional methods that depend on raw data or grayscale filtering, our approach functions directly on reconstructed sharp kernel images. It converts them to mimic the quality of soft-kernel outputs, obtained with kernels such as B30f, using paired data from the same patients. The efficacy of the QBF is evidenced by both full-reference metrics (Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE)) and no-reference perceptual metrics (Naturalness Image Quality Evaluator (NIQE), Blind Referenceless Image Spatial Quality Evaluator (BRISQUE), and Perception-based Image Quality Evaluator (PIQE)). The results indicate that the QBF demonstrates improved denoising efficacy compared to traditional Bilateral Filter (BF), Non-Local Means (NLM), wavelet, and Convolutional Neural Network (CNN)-based processing, achieving an SSIM of 0.96 and a PSNR of 36.3 on B50f reconstructions. Additionally, segmentation-based visual validation verifies that QBF-filtered outputs maintain essential structural details necessary for subsequent diagnostic tasks. This study emphasizes the importance of quaternion-based filtering as a lightweight, interpretable, and efficient substitute for deep learning models in post-reconstruction CT image enhancement.

AI-based automatic contouring streamlines radiotherapy by reducing contouring time but requires rigorous validation and ongoing daily monitoring. This study assessed how software updates affect contouring accuracy and examined how image quality variations influence AI performance. Two patient cohorts were analyzed. The software updates cohort (40 CT scans: 20 thorax, 10 pelvis, 10 H&N) compared six versions of Limbus AI contouring software. The image quality cohort (20 patients: H&N, pelvis, brain, thorax) analyzed 12 reconstructions per patient using Standard, iDose, and IMR algorithms, with simulated noise and spatial resolution (SR) degradations. AI performance was assessed using Volumetric Dice Similarity Coefficient (vDSC) and 95 % Hausdorff Distance (HD95%) with Wilcoxon tests for significance. In the software updates cohort, vDSC improved for re-trained structures across versions (mean DSC ≥ 0.75), with breast contour vDSC decreasing by 1 % between v1.5 and v1.8B3 (p > 0.05). Median HD95% values were consistently <4 mm, <5 mm, and <12 mm for H&N, pelvis, and thorax contours, respectively (p > 0.05). In the image quality cohort, no significant differences were observed between Standard, iDose, and IMR algorithms. However, noise and SR degradation significantly reduced performance: vDSC ≥ 0.9 dropped from 89 % at 2 % noise to 30 % at 20 %, and from 87 % to 70 % as SR degradation increased (p < 0.001). AI contouring accuracy improved with software updates and showed robustness to minor reconstruction variations, but it was sensitive to noise and SR degradation. Continuous validation and quality control of AI-generated contours are essential. Future studies should include a broader range of anatomical regions and larger cohorts.

Segmentation of anatomical structures and pathologies in medical images is essential for modern disease diagnosis, clinical research, and treatment planning. While significant advancements have been made in deep learning-based segmentation techniques, many of these methods still suffer from limitations in data efficiency, generalizability, and interactivity. As a result, developing robust segmentation methods that require fewer labeled datasets remains a critical challenge in medical image analysis. Recently, the introduction of foundation models like CLIP and Segment-Anything-Model (SAM), with robust cross-domain representations, has paved the way for interactive and universal image segmentation. However, further exploration of these models for data-efficient segmentation in medical imaging is an active field of research. In this paper, we introduce MedCLIP-SAMv2, a novel framework that integrates the CLIP and SAM models to perform segmentation on clinical scans using text prompts, in both zero-shot and weakly supervised settings. Our approach includes fine-tuning the BiomedCLIP model with a new Decoupled Hard Negative Noise Contrastive Estimation (DHN-NCE) loss, and leveraging the Multi-modal Information Bottleneck (M2IB) to create visual prompts for generating segmentation masks with SAM in the zero-shot setting. We also investigate using zero-shot segmentation labels in a weakly supervised paradigm to enhance segmentation quality further. Extensive validation across four diverse segmentation tasks and medical imaging modalities (breast tumor ultrasound, brain tumor MRI, lung X-ray, and lung CT) demonstrates the high accuracy of our proposed framework. Our code is available at https://github.com/HealthX-Lab/MedCLIP-SAMv2.

Due to the inductive bias of convolutions, CNNs perform hierarchical feature extraction efficiently in the field of medical image segmentation. However, the local correlation assumption of inductive bias limits the ability of convolutions to focus on global information, which has led to the performance of Transformer-based methods surpassing that of CNNs in some segmentation tasks in recent years. Although combining with Transformers can solve this problem, it also introduces computational complexity and considerable parameters. In addition, narrowing the encoder-decoder semantic gap for high-quality mask generation is a key challenge, addressed in recent works through feature aggregation from different skip connections. However, this often results in semantic mismatches and additional noise. In this paper, we propose a novel segmentation method, X-UNet, whose backbones employ the CFGC (Collaborative Fusion with Global Context-aware) module. The CFGC module enables multi-scale feature extraction and effective global context modeling. Simultaneously, we employ the CSPF (Cross Split-channel Progressive Fusion) module to progressively align and fuse features from corresponding encoder and decoder stages through channel-wise operations, offering a novel approach to feature integration. Experimental results demonstrate that X-UNet, with fewer computations and parameters, exhibits superior performance on various medical image datasets.The code and models are available on https://github.com/XSJ0410/X-UNet.

Reconstructing high-fidelity MR images from undersampled k-space data remains a challenging problem in MRI. While Mamba variants for vision tasks offer promising long-range modeling capabilities with linear-time complexity, their direct application to MRI reconstruction inherits two key limitations: (1) insensitivity to high-frequency anatomical details; and (2) reliance on redundant multi-directional scanning. To address these limitations, we introduce High-Fidelity Mamba (HiFi-Mamba), a novel dual-stream Mamba-based architecture comprising stacked W-Laplacian (WL) and HiFi-Mamba blocks. Specifically, the WL block performs fidelity-preserving spectral decoupling, producing complementary low- and high-frequency streams. This separation enables the HiFi-Mamba block to focus on low-frequency structures, enhancing global feature modeling. Concurrently, the HiFi-Mamba block selectively integrates high-frequency features through adaptive state-space modulation, preserving comprehensive spectral details. To eliminate the scanning redundancy, the HiFi-Mamba block adopts a streamlined unidirectional traversal strategy that preserves long-range modeling capability with improved computational efficiency. Extensive experiments on standard MRI reconstruction benchmarks demonstrate that HiFi-Mamba consistently outperforms state-of-the-art CNN-based, Transformer-based, and other Mamba-based models in reconstruction accuracy while maintaining a compact and efficient model design.

Federated learning (FL) is a decentralized machine learning paradigm in which multiple clients collaboratively train a shared model without sharing their local private data. However, real-world applications of FL frequently encounter challenges arising from the non-identically and independently distributed (non-IID) local datasets across participating clients, which is particularly pronounced in the field of medical imaging, where shifts in image feature distributions significantly hinder the global model's convergence and performance. To address this challenge, we propose FedMP, a novel method designed to enhance FL under non-IID scenarios. FedMP employs stochastic feature manifold completion to enrich the training space of individual client classifiers, and leverages class-prototypes to guide the alignment of feature manifolds across clients within semantically consistent subspaces, facilitating the construction of more distinct decision boundaries. We validate the effectiveness of FedMP on multiple medical imaging datasets, including those with real-world multi-center distributions, as well as on a multi-domain natural image dataset. The experimental results demonstrate that FedMP outperforms existing FL algorithms. Additionally, we analyze the impact of manifold dimensionality, communication efficiency, and privacy implications of feature exposure in our method.

Medical imaging, particularly retinal fundus photography, plays a crucial role in early disease detection and treatment for various ocular disorders. However, the development of robust diagnostic systems using deep learning remains constrained by the scarcity of expertly annotated data, which is time-consuming and expensive. Self-Supervised Learning (SSL) has emerged as a promising solution, but existing models fail to effectively incorporate critical domain knowledge specific to retinal anatomy. This potentially limits their clinical relevance and diagnostic capability. We address this issue by introducing an anatomically aware SSL framework that strategically integrates domain expertise through specialized masking of vital retinal structures during pretraining. Our approach leverages vessel and optic disc segmentation maps to guide the SSL process, enabling the development of clinically relevant feature representations without extensive labeled data. The framework combines a Vision Transformer with dual-masking strategies and anatomically informed loss functions to preserve structural integrity during feature learning. Comprehensive evaluation across multiple datasets demonstrates our method's competitive performance in diverse retinal disease classification tasks, including diabetic retinopathy grading, glaucoma detection, age-related macular degeneration identification, and multi-disease classification. The evaluation results establish the effectiveness of anatomically-aware SSL in advancing automated retinal disease diagnosis while addressing the fundamental challenge of limited labeled medical data.

Computer-aided medical image analysis is crucial for disease diagnosis and treatment planning, yet limited annotated datasets restrict medical-specific model development. While vision-language models (VLMs) like CLIP offer strong generalization capabilities, their direct application to medical imaging analysis is impeded by a significant domain gap. Existing approaches to bridge this gap, including prompt learning and one-way modality interaction techniques, typically focus on introducing domain knowledge to a single modality. Although this may offer performance gains, it often causes modality misalignment, thereby failing to unlock the full potential of VLMs. In this paper, we propose \textbf{NEARL-CLIP} (i\underline{N}teracted qu\underline{E}ry \underline{A}daptation with o\underline{R}thogona\underline{L} Regularization), a novel cross-modality interaction VLM-based framework that contains two contributions: (1) Unified Synergy Embedding Transformer (USEformer), which dynamically generates cross-modality queries to promote interaction between modalities, thus fostering the mutual enrichment and enhancement of multi-modal medical domain knowledge; (2) Orthogonal Cross-Attention Adapter (OCA). OCA introduces an orthogonality technique to decouple the new knowledge from USEformer into two distinct components: the truly novel information and the incremental knowledge. By isolating the learning process from the interference of incremental knowledge, OCA enables a more focused acquisition of new information, thereby further facilitating modality interaction and unleashing the capability of VLMs. Notably, NEARL-CLIP achieves these two contributions in a parameter-efficient style, which only introduces \textbf{1.46M} learnable parameters.