SMDRnet: Saliency multiscale dense residual network for multimodal medical image fusion.

Foundation models (FMs), which are large-scale architectures pretrained on diverse datasets, are rapidly reshaping artificial intelligence (AI). In medical imaging, these models provide unified methods for learning rich visual and multimodal representations, facilitating accurate diagnoses, efficient clinical workflows, and equitable access to healthcare. This review surveys the FMs in medical imaging and summarizes key insights for future research. It situates medical FMs in the broader context of generalist AI, categorizing them from vision-derived adaptations to modality-specific and emerging general-purpose systems. This review systematically compares core architectures, pre-training objectives, and clinical performance, highlighting improvements in adaptability, robustness, and data efficiency. Persistent challenges, such as limited data availability, privacy regulations, interpretability issues, and real-world deployment constraints, were examined alongside practical solutions, including self-supervised learning, federated training, and parameter-efficient fine-tuning. Future directions discussed include multimodal integration, lightweight inference suitable for edge devices, and rigorous validation that adheres to regulatory standards. By consolidating the current knowledge and identifying open research questions, this review offers clear guidance for researchers and clinicians aiming to integrate FMs into routine medical practice, laying the groundwork for subsequent detailed explorations. • Foundation models improves generalization in medical imaging tasks. • Vision-, modality-, and general-purpose models are systematically compared. • FMs mitigate data scarcity via self-supervised and federated learning. • Clinical applications include diagnosis, prognosis, and workflow automation. • Future work includes multimodal fusion, deployment, and validation.

Generative Artificial Intelligence (GAI) offers promising solutions to long-standing challenges in developing medical imaging methods and applications, including data scarcity, privacy concerns, and class imbalance. However, limited consolidation of publicly accessible synthetic datasets and trained GAI checkpoints restricts reproducibility and benchmarking. This systematic review aims to identify and evaluate such resources and assess their utility in clinical imaging applications. We systematically searched PubMed, IEEE Xplore, and Scopus for studies published between January 2017 and June 2024. Eligible studies generated or used synthetic medical image datasets and publicly released either the dataset or the trained GAI model. Extracted data included imaging modality, dataset characteristics, model architecture, public availability, and evaluation strategy. Of 941 screened records, 35 studies met inclusion criteria, comprising 37 publicly available resources spanning radiology (59%), pathology (16%), ophthalmology (14%), and dermatology (11%). Generative models included generative adversarial networks (73%), diffusion models (21%), autoencoders (3%), and hybrid architectures (8%). As some studies employed multiple model types, these categories are not mutually exclusive. Fifteen (43%) studies provided trained model checkpoints, enabling the generation of task-specific synthetic data. Evaluation methods included quantitative metrics, clinical expert assessment, and downstream performance in classification, segmentation, or detection tasks. Although the reviewed resources support diverse downstream applications, publicly available synthetic datasets and trained models remain scarce. Evaluation strategies vary widely, and the absence of standardized benchmarks limits cross-study comparisons and reliability assessment. To support reproducibility and responsible use of GAI in medical imaging, future work should prioritize the public release of curated synthetic resources, clearer guidance on model selection, and standardized, multi-dimensional evaluation frameworks.

Deep learning has been extensively applied to medical imaging tasks over the past years, achieving outstanding results. However, the obscure reasoning of the models and the lack of supportive evidence causes both clinicians and patients to distrust the models' predictions, hindering their adoption in clinical practice. In recent years, the research community has focused on developing explanations capable of revealing a model's reasoning. Among various types of explanations, example-based explanations emerged as particularly intuitive for medical practitioners. Despite the intuitiveness and wide development of example-based explanations, no work provides a comprehensive review of existing example-based explainability works in the medical image domain. In this work, we review works that provide example-based explanations for medical imaging tasks, reflecting on their strengths and limitations. We identify the absence of objective evaluation metrics, the lack of clinical validation and privacy concerns as the main issues that hinder the deployment of example-based explanations in clinical practice. Finally, we reflect on future directions contributing towards the deployment of example-based explainability in clinical practice.

MedSapiens: Taking a pose to rethink medical imaging landmark detection.

Teaching anatomy in the Bantu context: Traditions, cultural issues and educational perspectives in Gabon.

OMICS data in the diagnosis of diabetic retinopathy: A comparison between transcriptome data and DNA methylation data.

Tomek links-based SMOTE method for class imbalance in blood cell classification with dual path sliding window attention model.

Detection of diabetic retinopathy 5-stages by denseNet, a CNN model

Foundation models for brain imaging: A systematic review.

Read like a radiologist: Efficient vision-language model for 3D medical imaging interpretation.

Learning with less supervision: A survey of label-efficient learning for medical image analysis.

Mapping Artificial Intelligence Research in Oral and Maxillofacial Surgery: A Bibliometric Analysis.

Anomaly detection models based on autoencoders often suffer from unstable training, inconsistent reproducibility, and sensitivity to manually selected hyperparameters. These challenges limit their reliability across domains with varying data scales and feature distributions. This paper proposes an adaptive anomaly detection framework that combines a dual autoencoder architecture with a fully automated training-stability pipeline. The dual encoders capture global structural patterns and fine-grained local variations, enabling robust detection of subtle and heterogeneous anomalies. Unlike existing automated approaches which focus primarily on architecture search, our method targets the overlooked problem of training reproducibility and dataset-adaptive optimization. The proposed framework integrates (i) dynamic seed initialization for bias-resistant reproducibility, (ii) adaptive batch size estimation through a statistically derived logarithmic scaling function, (iii) iterative epoch optimization with multi-attempt checkpointing, and (iv) automated dropout tuning via Keras Tuner. To improve decision reliability, we further incorporate Monte Carlo dropout to estimate predictive uncertainty and reduce false-positive classifications. Extensive experiments on medical imaging, industrial inspection, 2D/3D textures, and surveillance datasets demonstrate consistent improvements over 11 state-of-the-art anomaly detection methods, yielding AUROC gains of up to 12.4% and significantly higher recall–precision stability. Ablation studies confirm that each optimization component contributes meaningfully to performance and reproducibility. • The framework eliminates manual tuning by integrating dynamic seed initialization, empirical batch size estimation, and attempt-level early stopping with model checkpointing. • Incorporates input gradient heatmaps and reconstruction error visualizations to explain model predictions, helping domain experts understand anomaly decisions. • Uses a min-distance ROC method to automatically determine optimal classification thresholds. • Evaluated on diverse datasets from medical, industrial, and surveillance domains. • Achieves better or comparable results to existing methods, while being more reproducible, scalable, and less sensitive to hyperparameter selection.

Automated dental report generation faces significant challenges in multimodal fusion, often resulting in suboptimal semantic quality and risks of hallucination, where AI generates clinically unsupported content. Current approaches that rely on simple feature concatenation or bidirectional attention mechanisms fail to effectively capture visual-textual relationships in medical imaging. This study aims to develop MedFusionT5, a unidirectional cross-modal alignment framework that (1) achieves superior clinical report quality through focused attention between visual patches and clinical text representations, and (2) ensures exceptional factual consistency by minimising hallucination rates. We implemented a novel architecture that integrates vision transformer (ViT) for patch-based visual feature extraction with Bio_ClinicalBERT for clinical text encoding. The core innovation is a unidirectional multihead attention alignment module that selectively maps textual embeddings to relevant visual patches before multimodal fusion. A T5-base decoder then generates diagnostic reports from the aligned representations. We evaluated performance on 700 dental panoramic radiographs using comprehensive metrics, including BLEU, ROUGE, CIDEr, clinical precision/recall, and specialised hallucination analysis, comparing against both concatenation and coattention baselines. MedFusionT5 demonstrated superior performance across all evaluated metrics. Compared to the coattention baseline, CIDEr increased by 122% (5.65 vs 2.54) and by 320% over simple concatenation. BLEU-4 reached 0.865, outperforming both baselines, while maintaining the lowest hallucination rate at 2.42% (39% reduction vs coattention, 46% vs concatenation). The model achieved an optimal balance between precision (0.982) and recall (0.923), with 90% of reports exhibiting near-zero hallucination. Notably, MedFusionT5 showed consistent quality independent of report length (r = -0.022), unlike coattention's length-dependent performance (r = +0.795). MedFusionT5 establishes a new state-of-the-art in automated dental report generation, demonstrating that unidirectional cross-modal alignment achieves superior semantic quality and clinical precision while minimising hallucinations. This work identifies unidirectional attention as the optimal alignment strategy for medical AI, providing a foundation for trustworthy clinical deployment where both accuracy and reliability are paramount.

Development and validation of stability prediction models for intracranial aneurysms based on ensemble learning algorithms.

Deformable image registration plays an essential role in various medical image tasks. Existing deep learning-based deformable registration frameworks primarily utilize convolutional neural networks (CNNs) or Transformers to learn features to predict the deformations. However, the lack of semantic information in the learned features limits the registration performance. Furthermore, the similarity metric of the loss function is often evaluated only in the pixel space, which ignores the matching of high-level anatomical features and can lead to deformation folding. To address these issues, in this work, we proposed LDM-Morph, an unsupervised deformable registration algorithm for medical image registration. LDM-Morph integrated features extracted from the latent diffusion model (LDM) to enrich the semantic information. Additionally, a latent and global feature-based cross-attention module (LGCA) was designed to enhance the interaction of semantic information from LDM and global information from multi-head self-attention operations. Finally, a hierarchical metric was proposed to evaluate the similarity of image pairs in both the original pixel space and latent-feature space, enhancing topology preservation while improving registration accuracy. Extensive experiments on four public 2D cardiac image datasets, two 3D image datasets, show that the proposed LDM-Morph framework outperformed existing state-of-the-art CNNs-and Transformers-based registration methods regarding accuracy with comparable topology preservation and computational efficiency. Our code is publicly available at: https://github.com/wujiong-hub/LDM-Morph.

Development and Validation of CLARITY (Clinical and Life-impact Assessment of RadiologY): A Patient-Reported Outcome Measure for Medical Imaging - Study Protocol.

A comprehensive wound-related image repository was developed to address critical gaps in existing medical imaging resources, particularly the lack of balanced datasets representing both healthy and pathological lower-limb conditions. The collection comprises 5443 images sourced from two complementary streams: real-world clinical wound cases and controlled acquisition of healthy feet images. The wound component includes 2686 expertly annotated images representing eight clinically significant wound types-diabetic, pressure, trauma, venous, surgical, arterial, cellulitis, and miscellaneous categories. These images were gathered across diverse clinical environments between 2015 and 2019 and meticulously annotated by certified wound specialists, ensuring high-quality segmentation masks including peri‑wound regions. The healthy-foot component consists of 2757 images captured from volunteer participants in naturalistic settings using consumer-grade smartphone cameras. Each participant contributed eight multi-angle images under consistent protocols, enabling robust representation of anatomical variability across sex, skin tone, and foot structure. All images were standardized through controlled resizing procedures, while the wound dataset underwent additional mask generation and augmentation strategies to support downstream segmentation and classification tasks. This unified dataset provides a balanced foundation for developing machine learning models capable of distinguishing between normal and pathological foot conditions while supporting advanced tasks such as wound segmentation, severity assessment, and clinical decision support. By integrating healthy and wound images within a single accessible collection, the dataset mitigates class imbalance issues prevalent in existing resources and enables scalable, generalizable deep learning research in wound detection, monitoring, and medical image analysis.

While pre-trained models, such as large language models, can achieve high performance with minimal fine-tuning, the source datasets used for pre-training often contain irrelevant or blackundant data, which can degrade performance on target tasks. Domain Adaptation Information Gain (DAIG)-based source data selection improves performance by pre-training on source data selected based on rough prior knowledge obtained from target data in advance. However, DAIG’s key component, the transition matrix, lacks flexibility and is limited to handling only single-label classification tasks. To address this limitation, we propose the Generalized DAIG (GDAIG)-guided selection process, a novel framework that extends DAIG to support multi-label classification. GDAIG introduces a soft transition matrix to capture inter-label dependencies and employs binary cross-entropy loss to enable adaptation to multi-label data. By leveraging “rough prior knowledge” from initial training on target data, GDAIG actively selects informative and task-relevant source data for pre-training. Experiments on medical image and general object classification datasets demonstrate that GDAIG consistently outperforms baseline approaches, with particularly significant improvements in scenarios involving label mismatch between source and target domains (partial or no label overlap), where conventional transfer learning methods suffer from noise caused by irrelevant source labels. These results highlight GDAIG’s ability to enhance the effectiveness of pre-trained models through strategic source data selection, thereby optimizing performance for specific target tasks. Our framework goes beyond existing approaches that rely solely on pre-trained models, emphasizing the direct utilization of task-relevant source data. Furthermore, GDAIG provides a practical and effective solution for domains with scarce labeled data, such as medical image analysis. • A GDAIG-guided data selection strategy for multi-label classification is proposed. • GDAIG improves target model performance through task-relevant multi-label data selection. • A probabilistic transition matrix captures inter-label dependencies. • “Rough prior” from target data effectively guides source data pre-training. • GDAIG outperforms conventional baselines across diverse multi-label scenarios.