Understanding bone remodeling and disease progression is crucial in preclinical skeletal research, particularly for assessing pharmacological and mechanical interventions in the long bones of murine models. High-resolution micro-computed tomography (micro-CT) imaging enables detailed trabecular bone analysis; however, inconsistent and non-standardized definitions of the volumes of interest (VOIs) across the different trabecular compartments compromise reproducibility and may lead to misleading statistical interpretations. In this study, we introduce a deep learning framework for automated trabecular bone analysis from micro-CT scans (5 µm voxel size) of the epiphyseal-metaphyseal region in the mouse tibia. The epiphyseal-metaphyseal region is classified into four anatomical compartments, epiphyseal bone, growth plate, primary spongiosa, and secondary spongiosa, using a 2D slice-wise classification model combined with a regional probability distribution method to detect the transitional landmarks between these compartments and enable standardized VOI extraction. To validate our method, we trained and tested the model on three micro-CT datasets comprising a total of 40 bone scans, each annotated by three experts to assess inter- and intra-operator variability, and further assessed its generalizability using an additional external dataset. These datasets encompassed diverse experimental conditions, including pharmacological treatments, mechanical loading, and age-related reduced bone density. Our classification model achieved excellent performances (mean F1-score of 0.96 for the epiphyseal bone, 0.95 for the growth plate, 0.92 for the primary spongiosa, and 0.99 for the secondary spongiosa across all datasets; statistical equivalence within 0.05 mm, [Formula: see text]) and demonstrated strong generalizability on the external dataset (mean F1-score of 0.99 for the epiphyseal bone, 0.97 for the growth plate, 0.92 for the primary spongiosa, and 1.0 for the secondary spongiosa; statistical equivalence within 0.05 mm, [Formula: see text]). Following the extraction of the different trabecular compartments, we segmented the trabecular bone within the epiphyseal bone, primary spongiosa, and secondary spongiosa using a deep learning-based segmentation model. We performed a comprehensive morphological and statistical analysis of all trabecular compartments in the mouse tibia, facilitating consistent comparisons across experimental groups and enabling direct comparisons within and between trabecular compartments. This automated method provides a consistent and robust tool for analyzing micro-CT scans of the trabecular bone in the mouse tibia, facilitating advancements in preclinical skeletal research.

The transformative potential of artificial intelligence (AI) in medical Imaging (MI) is well recognized. Yet despite promising reports in research settings, many AI tools fail to achieve clinical adoption in practice. In fact, more generally, there is a documented 17-year average delay between evidence generation and implementation of a technology1. Implementation science (IS) may provide a practical, evidence-based framework to bridge the gap between AI development and real-world clinical imaging use that helps shorten this lag through systematic frameworks, strategies, and hybrid research designs. We outline challenges specific to AI adoption in MI workflows, including infrastructural, educational, and cultural barriers. We highlight the complementary roles of effectiveness research and implementation research, emphasizing hybrid study designs and the role of integrated KT (iKT), stakeholder engagement, and equity-focused co-creation in designing sustainable and generalizable solutions. We discuss integration of Human-Computer Interaction (HCI) frameworks in MI towards usable AI. Adopting IS is not only a methodological advancement; it is a strategic imperative for accelerating translation of innovation into improved patient outcomes.

Recent medical vision-language models have shown promise on tasks such as VQA, report generation, and anomaly detection. However, most are adapted to structured adult imaging and underperform in fetal ultrasound, which poses challenges of multi-view image reasoning, numerous diseases, and image diversity. To bridge this gap, we introduce FetalMind, a medical AI system tailored to fetal ultrasound for both report generation and diagnosis. Guided by clinical workflow, we propose Salient Epistemic Disentanglement (SED), which injects an expert-curated bipartite graph into the model to decouple view-disease associations and to steer preference selection along clinically faithful steps via reinforcement learning. This design mitigates variability across diseases and heterogeneity across views, reducing learning bottlenecks while aligning the model's inference with obstetric practice. To train FetalMind at scale, we curate FetalSigma-1M dataset, the first large-scale fetal ultrasound report corpus, comprising 20K reports from twelve medical centers, addressing the scarcity of domain data. Extensive experiments show that FetalMind outperforms open- and closed-source baselines across all gestational stages, achieving +14% average gains and +61.2% higher accuracy on critical conditions while remaining efficient, stable, and scalable. Project Page: https://hexiao0275.github.io/FetalMind.

Arterial stiffness is a key predictor of cardiovascular mortality. This study utilizes a physics-informed neural network (PINN) model to estimate spatially varying arterial stiffness by leveraging ultrasound-based Pulse Wave Imaging (PWI) and Vector Flow Imaging (VFI). The PWI and VFI frameworks provide high-frame-rate wall displacement and blood flow velocity data, which are incorporated into a PINN constrained by linearized 1D differential equations modeling pulse wave propagation in heterogeneous vessels. The model was validated using in-silico simulations and physical plaque phantoms. The proposed PINN model effectively captured variations in localized compliance, which is indicative of arterial wall stiffness, in both in-silico and phantom experiments, and remained robust under varying inlet and outlet boundary conditions. Notably, incorporating flow velocity data enhanced reconstruction accuracy by 13.84% over displacement-only methods. The approach yielded low bias in homogeneous cases (1.41%), with higher biases for stiffer (3.95%) and softer (8.10%) plaque scenarios, mainly due to limitations in the 1D modeling and lack of explicit boundary condition integration. The findings presented herein indicate the PINN framework has strong potential for non-invasive assessment of focal arterial stiffness, such as in atherosclerotic plaques. Future work aims to include nonlinear vascular dynamics and extend the model to 2D or 3D to better capture complex blood flow behavior seen in stenotic arteries.

Radioembolization is a liver cancer treatment delivering radioactive microspheres (20-60 μm) to tumors via a catheter in the hepatic arterial tree. Treatment response depends on multiple factors including the complex hepatic artery anatomy, variable blood flow, and microsphere transport. Patient-specific digital twins powered by computational fluid dynamics (CFD) and physics-informed artificial intelligence (AI) methods offer a promising solution to optimize planning. This review discusses core principles of CFD and generative AI applied to radioembolization, emphasizing physics-informed networks and their role in translating digital twins into clinical practice for enhanced personalization and precision in treatment delivery.

This study aimed to develop an ultrasound (US)-based deep learning (DL) model to evaluate the presence of crescents in patients with immunoglobulin A nephropathy (IgAN). We created a training set consisting of 2,682 US images obtained from 931 patients with IgAN at the First Affiliated Hospital of Anhui Medical University. The external testing set included 198 patients from Nanchong Central Hospital based on the same criteria. Five DL models were trained in the training set and tested in the testing set. The performance of each model was evaluated for the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value and negative predictive value. The DenseNet121 model achieved an accuracy of 0.773 in the external testing set for predicting the presence of crescents, with a sensitivity of 57.6% and specificity of 87.9%. Using renal ultrasound imaging data, DL may be able to predict crescent status in IgAN, providing clinicians with a potentially non-invasive means to better understand crescent status in patients with IgAN. Not applicable. The online version contains supplementary material available at 10.1186/s12880-025-01958-w.

Breast cancer is the most diagnosed cancer in women, with HER2 status critically guiding treatment decisions. Noninvasive prediction of HER2 status from dynamic contrast-enhanced MRI (DCE-MRI) could streamline diagnostics and reduce reliance on biopsy. However, preprocessing high-dynamic-range DCE-MRI into standardized 8-bit RGB format for pretrained neural networks is nontrivial, and normalization strategy significantly affects model performance. We benchmarked intensity normalization strategies using a Triple-Head Dual-Attention ResNet that processes RGB-fused temporal sequences from three DCE phases. Trained on a multicenter cohort (n=1,149) from the I-SPY trials and externally validated on BreastDCEDL_AMBL (n=43 lesions), our model outperformed transformer-based architectures, achieving 0.75 accuracy and 0.74 AUC on I-SPY test data. N4 bias field correction slightly degraded performance. Without fine-tuning, external validation yielded 0.66 AUC, demonstrating cross-institutional generalizability. These findings highlight the effectiveness of dual-attention mechanisms in capturing transferable spatiotemporal features for HER2 stratification, advancing reproducible deep learning biomarkers in breast cancer imaging.

DSC-MRI perfusion is a medical imaging technique for diagnosing and prognosing brain tumors and strokes. Its analysis relies on mathematical deconvolution, but noise or motion artifacts in a clinical environment can disrupt this process, leading to incorrect estimate of perfusion parameters. Although deep learning approaches have shown promising results, their calibration typically rely on third-party deconvolution algorithms to generate reference outputs and are bound to reproduce their limitations. To adress this problem, we propose a physics-informed autoencoder that leverages an analytical model to decode the perfusion parameters and guide the learning of the encoding network. This autoencoder is trained in a self-supervised fashion without any third-party software and its performance is evaluated on a database with glioma patients. Our method shows reliable results for glioma grading in accordance with other well-known deconvolution algorithms despite a lower computation time. It also achieved competitive performance even in the presence of high noise which is critical in a medical environment.

Positron Emission Tomography (PET) myocardial perfusion imaging (MPI) is a powerful tool for predicting coronary artery disease (CAD). Coronary artery calcium (CAC) provides incremental risk stratification to PET-MPI and enhances diagnostic accuracy. We assessed additive value of CAC score, derived from PET/CT attenuation maps to stress TPD results using the novel 18F-flurpiridaz tracer in detecting significant CAD. Patients from 18F-flurpiridaz phase III clinical trial who underwent PET/CT MPI with 18F-flurpiridaz tracer, had available CT attenuation correction (CTAC) scans for CAC scoring, and underwent invasive coronary angiography (ICA) within a 6-month period between 2011 and 2013, were included. Total perfusion deficit (TPD) was quantified automatically, and CAC scores from CTAC scans were assessed using artificial intelligence (AI)-derived segmentation and manual scoring. Obstructive CAD was defined as ≥50% stenosis in Left Main (LM) artery, or 70% or more stenosis in any of the other major epicardial vessels. Prediction performance for CAD was assessed by comparing the area under receiver operating characteristic curve (AUC) for stress TPD alone and in combination with CAC score. Among 498 patients (72% males, median age 63 years) 30.1% had CAD. Incorporating CAC score resulted in a greater AUC: manual scoring (AUC=0.87, 95% Confidence Interval [CI] 0.34-0.90; p=0.015) and AI-based scoring (AUC=0.88, 95%CI 0.85-0.90; p=0.002) compared to stress TPD alone (AUC 0.84, 95% CI 0.80-0.92). Combining automatically derived TPD and CAC score enhances 18F-flurpiridaz PET MPI accuracy in detecting significant CAD, offering a method that can be routinely used with PET/CT scanners without additional scanning or technologist time.

Low muscle mass is a risk factor for chronic kidney disease. In this study, we examined the relationship between muscle mass and mortality, as well as end-stage kidney disease (ESKD), in patients with autosomal dominant polycystic kidney disease (ADPKD). Retrospective cohort study. 1,443 patients with ADPKD from eight tertiary-care hospitals in Korea between 2006 and 2020. Computed tomography images were obtained at the third lumbar vertebra to measure the skeletal muscle area (SMA) using an artificial intelligence system. SMA indexed for w a height² s classified as low-attenuation muscle area (LAMA) or normal-attenuation muscle area (NAMA) based on muscle quality. All-cause mortality and ESKD. Cox proportional hazards regression, adjusted for sex, age, creatinine, glucose, and height-adjusted total kidney volume, was used to investigate the associations of muscle indices with all-cause mortality and ESKD. Subgroup analyses were conducted based on body mass index categories: low or normal (<25 kg/m²) and overweight or obese (≥25 kg/m²). The study population included more than half female patients, and the mean estimated glomerular filtration rate was 68.4 ml/min/1.73m². Mean follow-up was 5.14 years. Greater SMA/height² and NAMA/height² were associated with a lower risk of mortality (HRs 0.58 (95% CI 0.39-0.88) and 0.55 (95% CI, 0.39-0.79), respectively). Greater NAMA/height² was associated with a 26% lower ESKD incidence (0.74 (0.59,0.92), but a greater LAMA/height² was associated with a lower ESKD incidence (HR 1.18, 95% CI 1.01-1.37). A higher NAMA/LAMA ratio was associated with a lower ESKD incidence (HR 0.74, 95% CI 0.60-0.92). Greater muscle mass was associated with a lower risk of mortality among overweight individuals and a lower risk of ESKD in underweight individuals. Lack of details about muscle strength and performance. Among individuals with ADPKD, greater and higher-quality muscle mass were associated with lower risk of mortality and progression of CKD to ESKD.