Computed tomography (CT) systems are indispensable for diagnostics but pose risks due to radiation exposure. Low-dose CT (LDCT) mitigates these risks but introduces noise and artifacts that compromise diagnostic accuracy. While deep learning methods, such as convolutional neural networks (CNNs) and generative adversarial networks (GANs), have been applied to LDCT denoising, challenges persist, including difficulties in preserving fine details and risks of model collapse. Recently, the Denoising Diffusion Probabilistic Model (DDPM) has addressed the limitations of traditional methods and demonstrated exceptional performance across various tasks. Despite these advancements, its high computational cost during training and extended sampling time significantly hinder practical clinical applications. Additionally, DDPM's reliance on random Gaussian noise can reduce optimization efficiency and performance in task-specific applications. To overcome these challenges, this study proposes a novel LDCT denoising framework that integrates the Latent Diffusion Model (LDM) with the Cold Diffusion Process. LDM reduces computational costs by conducting the diffusion process in a low-dimensional latent space while preserving critical image features. The Cold Diffusion Process replaces Gaussian noise with a CT denoising task-specific degradation approach, enabling efficient denoising with fewer time steps. Experimental results demonstrate that the proposed method outperforms DDPM in key metrics, including PSNR, SSIM, and RMSE, while achieving up to 2 × faster training and 14 × faster sampling. These advancements highlight the proposed framework's potential as an effective and practical solution for real-world clinical applications.

We introduce a novel framework for learning vector representations of tree-structured geometric data focusing on 3D vascular networks. Our approach employs two sequentially trained Transformer-based autoencoders. In the first stage, the Vessel Autoencoder captures continuous geometric details of individual vessel segments by learning embeddings from sampled points along each curve. In the second stage, the Vessel Tree Autoencoder encodes the topology of the vascular network as a single vector representation, leveraging the segment-level embeddings from the first model. A recursive decoding process ensures that the reconstructed topology is a valid tree structure. Compared to 3D convolutional models, this proposed approach substantially lowers GPU memory requirements, facilitating large-scale training. Experimental results on a 2D synthetic tree dataset and a 3D coronary artery dataset demonstrate superior reconstruction fidelity, accurate topology preservation, and realistic interpolations in latent space. Our scalable framework, named VeTTA, offers precise, flexible, and topologically consistent modeling of anatomical tree structures in medical imaging.

Skull damage caused by craniectomy or trauma necessitates accurate and precise Patient-Specific Implant (PSI) design to restore the cranial cavity. Conventional Computer-Aided Design (CAD)-based methods for PSI design are highly infrastructure-intensive, require specialised skills, and are time-consuming, resulting in prolonged patient wait times. Recent advancements in Artificial Intelligence (AI) provide automated, faster and scalable alternatives. This study introduces the Skull Completion using AI Network (SCAI-Net) framework, a deep-learning-based approach for automated cranial defect reconstruction using Computer Tomography (CT) images. The framework proposes two defect reconstruction variants: SCAI-Net-SDR (Subtraction-based Defect Reconstruction), which first reconstructs the full skull, then performs binary subtraction to obtain the reconstructed defect, and SCAI-Net-DDR (Direct Defect Reconstruction), which generates the reconstructed defect directly without requiring full-skull reconstruction. To enhance model robustness, the SCAI-Net was trained on an augmented dataset of 2760 images, created by combining MUG500+ and SkullFix datasets, featuring artificial defects across multiple cranial regions. Unlike subtraction-based SCAI-Net-SDR, which requires full-skull reconstruction before binary subtraction, and conventional CAD-based methods, which rely on interpolation or mirroring, SCAI-Net-DDR significantly reduces computational overhead. By eliminating the full-skull reconstruction step, DDR reduces training time by 66 % (85 min vs. 250 min for SDR) and achieves a 99.996 % faster defect reconstruction time compared to CAD (0.1s vs. 2400s). Based on the quantitative evaluation conducted on the SkullFix test cases, SCAI-Net-DDR emerged as the leading model among all evaluated approaches. SCAI-Net-DDR achieved the highest Dice Similarity Coefficient (DSC: 0.889), a low Hausdorff Distance (HD: 1.856 mm), and a superior Structural Similarity Index (SSIM: 0.897). Similarly, within the subset of subtraction-based reconstruction approaches evaluated, SCAI-Net-SDR demonstrated competitive performance, achieving the best HD (1.855 mm) and the highest SSIM (0.889), confirming its strong standing among methods using the subtraction paradigm. SCAI-Net generates reconstructed defects, which undergo post-processing to ensure manufacturing readiness. Steps include surface smoothing, thickness validation and edge preparation for secure fixation and seamless digital manufacturing compatibility. End-to-end implant generation time for DDR demonstrated a 96.68 % reduction (93.5 s), while SDR achieved a 96.64 % reduction (94.6 s), significantly outperforming CAD-based methods (2820s). Finite Element Analysis (FEA) confirmed the SCAI-Net-generated implants' robust load-bearing capacity under extreme loading (1780N) conditions, while edge gap analysis validated precise anatomical fit. Clinical validation further confirmed boundary accuracy, curvature alignment, and secure fit within cranial cavity. These results position SCAI-Net as a transformative, time-efficient, and resource-optimized solution for AI-driven cranial defect reconstruction and implant generation.

Cancer continues to be a significant international health issue, which demands the invention of new methods for early detection, precise diagnoses, and personalized treatments. Artificial intelligence (AI) has rapidly become a groundbreaking component in the modern era of oncology, offering sophisticated tools across the range of cancer care. In this review, we performed a systematic survey of the current status of AI technologies used for cancer diagnoses and therapeutic approaches. We discuss AI-facilitated imaging diagnostics using a range of modalities such as computed tomography, magnetic resonance imaging, positron emission tomography, ultrasound, and digital pathology, highlighting the growing role of deep learning in detecting early-stage cancers. We also explore applications of AI in genomics and biomarker discovery, liquid biopsies, and non-invasive diagnoses. In therapeutic interventions, AI-based clinical decision support systems, individualized treatment planning, and AI-facilitated drug discovery are transforming precision cancer therapies. The review also evaluates the effects of AI on radiation therapy, robotic surgery, and patient management, including survival predictions, remote monitoring, and AI-facilitated clinical trials. Finally, we discuss important challenges such as data privacy, interpretability, and regulatory issues, and recommend future directions that involve the use of federated learning, synthetic biology, and quantum-boosted AI. This review highlights the groundbreaking potential of AI to revolutionize cancer care by making diagnostics, treatments, and patient management more precise, efficient, and personalized.

The integration of advanced imaging and artificial intelligence technologies in radiation therapy has revolutionized cancer treatment by enhancing precision and adaptability. This study introduces a novel dual-energy computed tomography (DECT) guided intelligent radiation therapy (DEIT) platform designed to streamline and optimize the radiation therapy process. The DEIT system combines DECT, a newly designed dual-layer multileaf collimator, deep learning algorithms for auto-segmentation, and automated planning and quality assurance capabilities. The DEIT system integrates an 80-slice computed tomography (CT) scanner with an 87 cm bore size, a linear accelerator delivering 4 photon and 5 electron energies, and a flat panel imager optimized for megavoltage (MV) cone beam CT acquisition. A comprehensive evaluation of the system's accuracy was conducted using end-to-end tests. Virtual monoenergetic CT images and electron density images of the DECT were generated and compared on both phantom and patient. The system's auto-segmentation algorithms were tested on 5 cases for each of the 99 organs at risk, and the automated optimization and planning capabilities were evaluated on clinical cases. The DEIT system demonstrated systematic errors of less than 1 mm for target localization. DECT reconstruction showed electron density mapping deviations ranging from -0.052 to 0.001, with stable Hounsfield unit consistency across monoenergetic levels above 60 keV, except for high-Z materials at lower energies. Auto-segmentation achieved dice similarity coefficients above 0.9 for most organs with an inference time of less than 2 seconds. Dose-volume histogram comparisons showed improved dose conformity indices and reduced doses to critical structures in auto-plans compared to manual plans across various clinical cases. In addition, high gamma passing rates at 2%/2 mm in both 2-dimensional (above 97%) and 3-dimensional (above 99%) in vivo analyses further validate the accuracy and reliability of treatment plans. The DEIT platform represents a viable solution for radiation treatment. The DEIT system uses artificial intelligence-driven automation, real-time adjustments, and CT imaging to enhance the radiation therapy process, improving efficiency and flexibility.

Cardiac CT plays an important role in diagnosing heart diseases but is conventionally limited by its complex workflow that requires dedicated phase and bolus tracking devices [e.g., electrocardiogram (ECG) gating]. This work reports first progress towards robust and autonomous cardiac CT exams through joint deep learning (DL) and analytical analysis of pulsed-mode projections (PMPs). To this end, cardiac phase and its uncertainty were simultaneously estimated using a novel projection domain cardiac phase estimation network (PhaseNet), which utilizes sliding-window multi-channel feature extraction strategy and a long short-term memory (LSTM) block to extract temporal correlation between time-distributed PMPs. An uncertainty-driven Viterbi (UDV) regularizer was developed to refine the DL estimations at each time point through dynamic programming. Stronger regularization was performed at time points where DL estimations have higher uncertainty. The performance of the proposed phase estimation pipeline was evaluated using accurate physics-based emulated data. PhaseNet achieved improved phase estimation accuracy compared to the competing methods in terms of RMSE (~50% improvement vs. standard CNN-LSTM; ~24% improvement vs. multi-channel residual network). The added UDV regularizer resulted in an additional ~14% improvement in RMSE, achieving accurate phase estimation with <6% RMSE in cardiac phase (phase ranges from 0-100%). To our knowledge, this is the first publication of prospective cardiac phase estimation in the projection domain. Combined with our previous work on PMP-based bolus curve estimation, the proposed method could potentially be used to achieve autonomous cardiac scanning without ECG device and expert-in-the-loop bolus timing.

The test-time adaptation (TTA) of deep-learning-based semantic segmentation models, specific to individual patient data, was addressed in this study. The existing TTA methods in medical imaging are often unconstrained, require anatomical prior information or additional neural networks built during training phase, making them less practical, and prone to performance deterioration. In this study, a novel framework based on information geometric principles was proposed to achieve generic, off-the-shelf, regularized patient-specific adaptation of models during test-time. By considering the pre-trained model and the adapted models as part of statistical neuromanifolds, test-time adaptation was treated as constrained functional regularization using information geometric measures, leading to improved generalization and patient optimality. The efficacy of the proposed approach was shown on three challenging problems: 1) improving generalization of state-of-the-art models for segmenting COVID-19 anomalies in Computed Tomography (CT) images 2) cross-institutional brain tumor segmentation from magnetic resonance (MR) images, 3) segmentation of retinal layers in Optical Coherence Tomography (OCT) images. Further, it was demonstrated that robust patient-specific adaptation can be achieved without adding significant computational burden, making it first of its kind based on information geometric principles.

A diagnostic medical foundation model (MedFM) is an artificial intelligence (AI) system engineered to accurately determine diagnoses across various medical imaging modalities and specialties. To train MedFM, we created the PubMed Central Medical Images Dataset (PMCMID), the largest annotated medical image dataset to date, comprising 16,126,659 images from 3,021,780 medical publications. Using AI- and ontology-based methods, we identified 4,482,237 medical images (e.g., clinical photos, X-rays, ultrasounds) and generated comprehensive annotations. To optimize MedFMs performance and assess biases, 13,266 images were manually annotated to establish a multimodal benchmark. MedFM achieved physician-level performance in diagnosis tasks spanning radiology, dermatology, and infectious diseases without requiring specific training. Additionally, we developed the Image2Paper app, allowing clinicians to upload medical images and retrieve relevant literature. The correct diagnoses appeared within the top ten results in 88.4% and at least one relevant differential diagnosis in 93.0%. MedFM and PMCMID were made publicly available. FundingResearch reported here was partially supported by the National Cancer Institute (NCI) (R01 CA260271), the Saudi Company for Artificial Intelligence (SCAI) Authority, and the German Federal Ministry for Economic Affairs and Climate Action (BMWK) under the project DAKI-FWS (01MK21009E). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Medical image reconstruction from measurement data is a vital but challenging inverse problem. Deep learning approaches have achieved promising results, but often requires paired measurement and high-quality images, which is typically simulated through a forward model, i.e., retrospective reconstruction. However, training on simulated pairs commonly leads to performance degradation on real prospective data due to the retrospective-to-prospective gap caused by incomplete imaging knowledge in simulation. To address this challenge, this paper introduces imaging Knowledge-Informed Dynamic Optimal Transport (KIDOT), a novel dynamic optimal transport framework with optimality in the sense of preserving consistency with imaging physics in transport, that conceptualizes reconstruction as finding a dynamic transport path. KIDOT learns from unpaired data by modeling reconstruction as a continuous evolution path from measurements to images, guided by an imaging knowledge-informed cost function and transport equation. This dynamic and knowledge-aware approach enhances robustness and better leverages unpaired data while respecting acquisition physics. Theoretically, we demonstrate that KIDOT naturally generalizes dynamic optimal transport, ensuring its mathematical rationale and solution existence. Extensive experiments on MRI and CT reconstruction demonstrate KIDOT's superior performance.

Recently, Ultra-High Field MRI (UHF-MRI) has become more available and one of the best tools to study the brain. One common step in quantitative neuroimaging is the brain segmentation. However, the differences between UHF-MRI and 1.5-3T images are such that the automatic segmentation techniques optimized at these field strengths usually produce unsatisfactory segmentation results for UHF images. It has been particularly challenging to perform quantitative analyses as typically done with 1.5-3T data, considerably limiting the potential of UHF-MRI. Hence, we propose a novel Deep Learning (DL)-based segmentation technique called GOUHFI: Generalized and Optimized segmentation tool for Ultra-High Field Images, designed to segment UHF images of various contrasts and resolutions. For training, we used a total of 206 label maps from four datasets acquired at 3T, 7T and 9.4T. In contrast to most DL strategies, we used a previously proposed domain randomization approach, where synthetic images generated from the label maps were used for training a 3D U-Net. GOUHFI was tested on seven different datasets and compared to techniques like FastSurferVINN and CEREBRUM-7T. GOUHFI was able to the segment six contrasts and seven resolutions tested at 3T, 7T and 9.4T. Average Dice-Sorensen Similarity Coefficient (DSC) scores of 0.87, 0.84, 0.91 were computed against the ground truth segmentations at 3T, 7T and 9.4T. Moreover, GOUHFI demonstrated impressive resistance to the typical inhomogeneities observed at UHF-MRI, making it a new powerful segmentation tool that allows to apply the usual quantitative analysis pipelines also at UHF. Ultimately, GOUHFI is a promising new segmentation tool, being the first of its kind proposing a contrast- and resolution-agnostic alternative for UHF-MRI, making it the forthcoming alternative for neuroscientists working with UHF-MRI or even lower field strengths.