Multimodal medical imaging integrates diverse data types, such as structural and functional neuroimaging, to provide complementary insights that enhance deep learning predictions and improve outcomes. This study focuses on a neuroimaging prediction framework based on both structural and functional neuroimaging data. We propose a next-generation prediction model, \textbf{MultiViT2}, which combines a pretrained representative learning base model with a vision transformer backbone for prediction output. Additionally, we developed a data augmentation module based on the latent diffusion model that enriches input data by generating augmented neuroimaging samples, thereby enhancing predictive performance through reduced overfitting and improved generalizability. We show that MultiViT2 significantly outperforms the first-generation model in schizophrenia classification accuracy and demonstrates strong scalability and portability.

Foundation models (FMs), large neural networks pretrained on extensive and diverse datasets, have revolutionized artificial intelligence and shown significant promise in medical imaging by enabling robust performance with limited labeled data. Although numerous surveys have reviewed the application of FM in healthcare care, brain imaging remains underrepresented, despite its critical role in the diagnosis and treatment of neurological diseases using modalities such as MRI, CT, and PET. Existing reviews either marginalize brain imaging or lack depth on the unique challenges and requirements of FM in this domain, such as multimodal data integration, support for diverse clinical tasks, and handling of heterogeneous, fragmented datasets. To address this gap, we present the first comprehensive and curated review of FMs for brain imaging. We systematically analyze 161 brain imaging datasets and 86 FM architectures, providing information on key design choices, training paradigms, and optimizations driving recent advances. Our review highlights the leading models for various brain imaging tasks, summarizes their innovations, and critically examines current limitations and blind spots in the literature. We conclude by outlining future research directions to advance FM applications in brain imaging, with the aim of fostering progress in both clinical and research settings.

Computer-aided surgical simulation is a critical component of orthognathic surgical planning, where accurately simulating face-bone shape transformations is significant. The traditional biomechanical simulation methods are limited by their computational time consumption levels, labor-intensive data processing strategies and low accuracy. Recently, deep learning-based simulation methods have been proposed to view this problem as a point-to-point transformation between skeletal and facial point clouds. However, these approaches cannot process large-scale points, have limited receptive fields that lead to noisy points, and employ complex preprocessing and postprocessing operations based on registration. These shortcomings limit the performance and widespread applicability of such methods. Therefore, we propose a Transformer-based coarse-to-fine point movement network (TCFNet) to learn unique, complicated correspondences at the patch and point levels for dense face-bone point cloud transformations. This end-to-end framework adopts a Transformer-based network and a local information aggregation network (LIA-Net) in the first and second stages, respectively, which reinforce each other to generate precise point movement paths. LIA-Net can effectively compensate for the neighborhood precision loss of the Transformer-based network by modeling local geometric structures (edges, orientations and relative position features). The previous global features are employed to guide the local displacement using a gated recurrent unit. Inspired by deformable medical image registration, we propose an auxiliary loss that can utilize expert knowledge for reconstructing critical organs. Our framework is an unsupervised algorithm, and this loss is optional. Compared with the existing state-of-the-art (SOTA) methods on gathered datasets, TCFNet achieves outstanding evaluation metrics and visualization results. The code is available at https://github.com/Runshi-Zhang/TCFNet.

Early screening for gastrointestinal (GI) diseases is critical for preventing cancer development. With the rapid advancement of deep learning technology, artificial intelligence (AI) has become increasingly prominent in the early detection of GI diseases. Capsule endoscopy is a non-invasive medical imaging technique used to examine the gastrointestinal tract. In our previous work, we developed a near-infrared fluorescence capsule endoscope (NIRF-CE) capable of exciting and capturing near-infrared (NIR) fluorescence images to specifically identify subtle mucosal microlesions and submucosal abnormalities while simultaneously capturing conventional white-light images to detect lesions with significant morphological changes. However, limitations such as low camera resolution and poor lighting within the gastrointestinal tract may lead to misdiagnosis and other medical errors. Manually reviewing and interpreting large volumes of capsule endoscopy images is time-consuming and prone to errors. Deep learning models have shown potential in automatically detecting abnormalities in NIRF-CE images. This study focuses on an improved deep learning model called Retinex-Attention-YOLO (RAY), which is based on single-modality image data and built on the YOLO series of object detection models. RAY enhances the accuracy and efficiency of anomaly detection, especially under low-light conditions. To further improve detection performance, we also propose a multimodal deep learning model, Multimodal-Retinex-Attention-YOLO (MRAY), which combines both white-light and fluorescence image data. The dataset used in this study consists of images of pig stomachs captured by our NIRF-CE system, simulating the human GI tract. In conjunction with a targeted fluorescent probe, which accumulates at lesion sites and releases fluorescent signals for imaging when abnormalities are present, a bright spot indicates a lesion. The MRAY model achieved an impressive precision of 96.3%, outperforming similar object detection models. To further validate the model's performance, ablation experiments were conducted, and comparisons were made with publicly available datasets. MRAY shows great promise for the automated detection of GI cancers, ulcers, inflammations, and other medical conditions in clinical practice.

Fairness concerns stemming from known and unknown biases in healthcare practices have raised questions about the trustworthiness of Artificial Intelligence (AI)-driven Clinical Decision Support Systems (CDSS). Studies have shown unforeseen performance disparities in subpopulations when applied to clinical settings different from training. Existing unfairness mitigation strategies often struggle with scalability and accessibility, while their pursuit of group-level prediction performance parity does not effectively translate into fairness at the point of care. This study introduces FairICP, a flexible and cost-effective post-implementation framework based on Inductive Conformal Prediction (ICP), to provide users with actionable knowledge of model uncertainty due to subpopulation level biases at the point of care. FairICP applies ICP to identify the model's scope of competence through group specific calibration, ensuring equitable prediction reliability by filtering predictions that fall within the trusted competence boundaries. We evaluated FairICP against four benchmarks on three medical imaging modalities: (1) Cardiac Magnetic Resonance Imaging (MRI), (2) Chest X-ray and (3) Dermatology Imaging, acquired from both private and large public datasets. Frameworks are assessed on prediction performance enhancement and unfairness mitigation capabilities. Compared to the baseline, FairICP improved prediction accuracy by 7.2% and reduced the accuracy gap between the privileged and unprivileged subpopulations by 2.2% on average across all three datasets. Our work provides a robust solution to promote trust and transparency in AI-CDSS, fostering equality and equity in healthcare for diverse patient populations. Such post-process methods are critical to enabling a robust framework for AI-CDSS implementation and monitoring for healthcare settings.

As one of the leading causes of death worldwide, early detection of lung disease is a very important step to improve the effectiveness of treatment. By using medical image data, such as X-ray or CT-scan, classification of lung disease can be done. Deep learning methods have been widely used to recognize complex patterns in medical images, but this approach has the constraints of requiring large data variations and high computing resources. In overcoming these constraints, the lightweight architecture in deep learning can provide a more efficient solution based on the number of parameters and computing time. This method can be applied to devices with low processor specifications on portable devices such as mobile phones. This article presents a comprehensive review of 23 research studies published between 2020 and 2025, focusing on various lightweight architectures and optimization techniques aimed at improving the accuracy of lung disease detection. The results show that these models are able to significantly reduce parameter sizes, resulting in faster computation times while maintaining competitive accuracy compared to traditional deep learning architectures. From the research that has been done, it can be seen that SqueezeNet applied on public COVID-19 datasets is the best basic architecture with high accuracy, and the number of parameters is 570 thousand, which is very low. On the other hand, UNet requires 31.07 million parameters, and SegNet requires 29.45 million parameters trained on CT scan images from Italian Society of Medical and Interventional Radiology and Radiopedia, so it is less efficient. For the combination method, EfficientNetV2 and Extreme Learning Machine (ELM) are able to achieve the highest accuracy of 98.20 % and can significantly reduce parameters. The worst performance is shown by VGG and UNet with a decrease in accuracy from 91.05 % to 87 % and an increase in the number of parameters. It can be concluded that the lightweight architecture can be applied to medical image classification in the diagnosis of lung disease quickly and efficiently on devices with limited specifications.

Three-dimensional visualization is increasingly used in dentistry for diagnostics, education, and treatment design. The accurate replication of geometry and color is crucial for these applications. Image-based rendering, which uses 2-dimensional photos to generate photo-realistic 3-dimensional representations, provides an affordable and practical option, aiding both regular and remote health care. This study explores an advanced novel view synthesis (NVS) method called Gaussian splatting (GS), a differentiable image-based rendering approach, to assess its feasibility for dental scene capturing. The rendering quality and resource usage were compared with representative NVS methods. In addition, the linear measurement trueness of extracted craniofacial meshes was evaluated against a commercial facial scanner and 3 smartphone facial scanning apps, while teeth meshes were assessed against 2 intraoral scanners and a desktop scanner. GS-based representation demonstrated superior rendering quality, achieving the highest visual quality, fastest rendering speed, and lowest resource usage. The craniofacial measurements showed similar trueness to commercial facial scanners. The dental measurements had larger deviations than intraoral and desktop scanners did, although all deviations remained within clinically acceptable limits. The GS-based representation shows great potential for developing a convenient and cost-effective method of capturing dental scenes, offering a balance between color fidelity and trueness suitable for clinical applications.

Automatic segmentation of anatomical structures is critical in medical image analysis, aiding diagnostics and treatment planning. Skin segmentation plays a key role in registering and visualising multimodal imaging data. 3D skin segmentation enables applications in personalised medicine, surgical planning, and remote monitoring, offering realistic patient models for treatment simulation, procedural visualisation, and continuous condition tracking. This paper analyses and compares algorithmic and AI-driven skin segmentation approaches, emphasising key factors to consider when selecting a strategy based on data availability and application requirements. We evaluate an iterative region-growing algorithm and the TotalSegmentator, a deep learning-based approach, across different imaging modalities and anatomical regions. Our tests show that AI segmentation excels in automation but struggles with MRI due to its CT-based training, while the graphics-based method performs better for MRIs but introduces more noise. AI-driven segmentation also automates patient bed removal in CT, whereas the graphics-based method requires manual intervention.

Incomplete multi-modal medical image segmentation faces critical challenges from modality imbalance, including imbalanced modality missing rates and heterogeneous modality contributions. Due to their reliance on idealized assumptions of complete modality availability, existing methods fail to dynamically balance contributions and neglect the structural relationships between modalities, resulting in suboptimal performance in real-world clinical scenarios. To address these limitations, we propose a novel model, named Dynamic Modality-Aware Fusion Network (DMAF-Net). The DMAF-Net adopts three key ideas. First, it introduces a Dynamic Modality-Aware Fusion (DMAF) module to suppress missing-modality interference by combining transformer attention with adaptive masking and weight modality contributions dynamically through attention maps. Second, it designs a synergistic Relation Distillation and Prototype Distillation framework to enforce global-local feature alignment via covariance consistency and masked graph attention, while ensuring semantic consistency through cross-modal class-specific prototype alignment. Third, it presents a Dynamic Training Monitoring (DTM) strategy to stabilize optimization under imbalanced missing rates by tracking distillation gaps in real-time, and to balance convergence speeds across modalities by adaptively reweighting losses and scaling gradients. Extensive experiments on BraTS2020 and MyoPS2020 demonstrate that DMAF-Net outperforms existing methods for incomplete multi-modal medical image segmentation. Extensive experiments on BraTS2020 and MyoPS2020 demonstrate that DMAF-Net outperforms existing methods for incomplete multi-modal medical image segmentation. Our code is available at https://github.com/violet-42/DMAF-Net.

AI requires extensive datasets, while medical data is subject to high data protection. Anonymization is essential, but poses a challenge for some regions, such as the head, as identifying structures overlap with regions of clinical interest. Synthetic data offers a potential solution, but studies often lack rigorous evaluation of realism and utility. Therefore, we investigate to what extent synthetic data can replace real data in segmentation tasks. We employed head and neck cancer CT scans and brain glioma MRI scans from two large datasets. Synthetic data were generated using generative adversarial networks and diffusion models. We evaluated the quality of the synthetic data using MAE, MS-SSIM, Radiomics and a Visual Turing Test (VTT) performed by 5 radiologists and their usefulness in segmentation tasks using DSC. Radiomics indicates high fidelity of synthetic MRIs, but fall short in producing highly realistic CT tissue, with correlation coefficient of 0.8784 and 0.5461 for MRI and CT tumors, respectively. DSC results indicate limited utility of synthetic data: tumor segmentation achieved DSC=0.064 on CT and 0.834 on MRI, while bone segmentation a mean DSC=0.841. Relation between DSC and correlation is observed, but is limited by the complexity of the task. VTT results show synthetic CTs' utility, but with limited educational applications. Synthetic data can be used independently for the segmentation task, although limited by the complexity of the structures to segment. Advancing generative models to better tolerate heterogeneous inputs and learn subtle details is essential for enhancing their realism and expanding their application potential.