To develop predictive models for fetal growth restriction (FGR) with and without the inclusion of OxNNet-derived first-trimester placental volume (FTPV), thereby evaluating the contribution of FTPV to these models and the extent to which FTPV percentile is associated with subsequent FGR. This study utilized data from the First-trimester Placental Ultrasound (FirstPLUS) study, a longitudinal observational cohort study conducted at King's College Hospital NHS Foundation Trust, London, UK, between March and November 2022. Participants underwent routine ultrasound assessment between 11 + 2 and 14 + 1 weeks' gestation, in addition to three-dimensional placental sonography. The OxNNet toolkit was used for automated placental segmentation and volume calculation. Multivariable logistic regression models were developed to predict FGR, incorporating maternal factors, first-trimester biomarkers (serum pregnancy-associated plasma protein-A, mean arterial blood pressure and uterine artery pulsatility index) and FTPV. The final cohort comprised 3500 pregnancies, of which 250 (7.1%) developed FGR. Low FTPV was found to be a risk factor for FGR, with an odds ratio of 1.736 (95% CI, 1.499-2.015) per unit decrease in FTPV Z-score. Incorporating FTPV into the predictive model based on maternal factors and biomarkers significantly increased the area under the receiver-operating-characteristics curve (AUC) for predicting all cases of FGR, from 0.78 (95% CI, 0.75-0.81) to 0.79 (95% CI, 0.76-0.82) (P = 0.005). Subgroup analysis of normotensive and hypertensive cases demonstrated a statistically significant effect size for the prediction of FGR by FTPV Z-score in both groups. The addition of FTPV to the model based on maternal factors and biomarkers for the prediction of normotensive FGR increased the AUC from 0.77 (95% CI, 0.74-0.80) to 0.78 (95% CI, 0.75-0.81) (P = 0.01). For preterm FGR, the AUC was 0.85 (95% CI, 0.78-0.92) with FTPV and 0.85 (95% CI, 0.79-0.92) without (P = 0.93); the absence of a significant difference may be due to a lack of power. FTPV Z-score is a predictor of FGR. Integrating FTPV into predictive models significantly enhanced the discriminative ability for all cases of FGR, as well as for the subgroup of normotensive FGR. © 2025 The Author(s). Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology.

To address Croatia's high lung cancer mortality and late-stage diagnoses, the Ministry of Health initiated a multidisciplinary effort to design a national lung cancer screening program. Lung cancer remains one of the leading causes of cancer-related mortality both globally and in Croatia. In 2021 alone, Croatia recorded over 3300 new cases of lung cancer and more than 2800 associated deaths, indicating a high mortality burden. In response to this public health concern, the Ministry of Health has established a multidisciplinary Lung Cancer Screening Working Group, tasked with developing a national screening approach. The Program incorporates several innovative elements, including the application of modified International Early Lung Cancer Action Program (I-ELCAP) criteria for nodule management, volumetric analysis assessed by artificial intelligence, complete digitalization, smoking cessation, and nationwide deployment to ensure equitable access. From October 2020 to August 2025, over 50,000 participants were screened, resulting in more than 70,000 LDCT scans performed. The cohort includes 54% male and 46% female participants, with an average age of 62 years. Among these participants, 4.5% had positive results, which required further follow-up. The Croatian National Lung Cancer Screening Program offers unique features as it has been comprehensively incorporated into the existing healthcare infrastructure and is fully reimbursed. A key aspect of the program is the important role assigned to general practitioners (GPs), who are responsible for identifying and referring individuals at high risk for lung cancer. Question No European Union country has implemented a national lung cancer screening program despite evidence from previous trials showing significant mortality reduction. Findings Croatia successfully launched a fully integrated national lung cancer screening program using LDCT, AI-assisted volumetric analysis, modified I-ELCAP criteria, and GP-centered recruitment. Clinical relevance The Croatian model demonstrates the feasibility of national lung cancer screening within a European public healthcare system with full reimbursement, providing a replicable framework for other EU countries implementing lung cancer screening programs.

Pediatric cardiovascular imaging plays an important role in the diagnosis, monitoring, and management of congenital and acquired heart diseases. Although echocardiography remains the most widely used modality in pediatric cardiology, cross-sectional imaging techniques such as cardiac magnetic resonance imaging (MRI) and computed tomography (CT) provide complementary anatomic and functional information. However, time-consuming diagnostic processes and patient-specific characteristics remain major limitations to early and precise diagnosis, as well as optimal clinical outcomes. With technological advancements, artificial intelligence (AI) has been increasingly integrated into cardiovascular magnetic resonance imaging (MRI) and computed tomography (CT) to enhance image acquisition, segmentation, interpretation, and diagnosis, and to facilitate predictive modeling of clinical outcomes. This review summarizes current and emerging AI applications in pediatric cardiovascular MRI and CT, emphasizing workflow optimization, diagnostic automation, and quantitative analysis. Emerging frontiers include multimodal data integration for risk stratification, clinical decision-making support, digital twin models, three-dimensional virtual modeling, and the application of computational fluid dynamics, as well as the potential of AI to improve access to care in low-resource settings.

The utility of ¹⁸F-fluorodeoxyglucose positron emission tomography with low-dose computed tomography (¹⁸F-FDG PET/CT) to diagnose patients with atypical features but clinically confirmed PMR is unknown. This study explored ¹⁸F-FDG PET/CT findings among atypical PMR cases and contrasted results with a typical PMR population. The performance of an AI model for PMR diagnosis was also tested. Natural language processing searched the electronic data warehouse at our institution for ¹⁸F-FDG PET/CT reports with PMR findings. Medical record review confirmed the clinical diagnosis for potential cases. Required entry Criteria of the 2012 EULAR/ACR Classification for PMR (age ≥50 years, bilateral shoulder aching and abnormal CRP and/or ESR) stratified patients as "typical" or"atypical". Qualitative and semi-quantitative (SUV_max) scoring of ¹⁸F-FDG uptake at characteristic musculoskeletal sites was undertaken for steroid-naïve, atypical PMR patients and compared with results for steroid-naïve, typical cases from an earlier study. A ResNet50 AI model was tested on the atypical cohort's images, with GradCam maps generated and cross-checked by an experienced clinician. 225 ¹⁸F-FDG PET/CT reports were retrieved from >38,000 scans; 121 patients had a clinical PMR diagnosis. Seventeen steroid-naïve, atypical cases were compared with 35 steroid-naïve, typical patients. The frequency of abnormalities at key musculoskeletal sites was similar between groups, however mean SUV_maxwas significantly higher in the typical cohort. The AI model identified a PMR diagnosis in 16/17 (94.1%) atypical patients. Characteristic findings of PMR are appreciated on ¹⁸F-FDG PET/CT among atypical patients, but avidity is reduced. AI technology can reliably detect this imaging pattern. Words: 249 _{(max. 250)}.

This study explores the feasibility and effectiveness of an interpretable machine learning model for assessing the pathological grading of pancreatic ductal adenocarcinoma (PDAC) using radiomics and topological features derived from contrast-enhanced CT habitat subregions. A retrospective study was conducted on a total of 306 patients with PDAC from two hospitals: a training cohort (n = 176), a validation cohort (n = 76), and a test cohort (n = 54). K-means clustering analysis was first used to segment portal venous phase CT images into three habitat regions. Radiomics features of the whole-tumour region, along with radiomics and topological features of each habitat region, were extracted respectively. LASSO regression was applied for feature dimensionality reduction to construct the radiomics score (Rad-score) for the whole-tumour region and the habitat score (H-score) for each habitat region. Meanwhile, logistic regression was used to identify statistically significant predictors from clinical and semantic features. Five machine learning algorithms were used to construct Habitat-TDA models, with interpretability analysis performed via SHAP analysis. Total volume, diabetes, and M staging were identified as independent risk factors for predicting the pathological grading of PDAC, and were used to construct the Clinical model. 6 radiomics features with non-zero coefficients were selected to calculate the Rad-score, which was further used to construct the WholeRad model. In the three habitat regions, 6, 5, and 6 topological and radiomics features were included to generate the H-score. The logistic regression algorithm performed best in the validation and test cohorts and was ultimately selected as the classifier for constructing the Habitat-TDA model. SHAP analysis showed that H-score1, derived from Habitat Region 1 (the habitat region with the lowest average CT value), has the most significant average impact on the model output intensity. The AUC values of the Habitat-TDA model in the training, validation, and test cohorts were 0.894, 0.872, and 0.829, all outperforming the clinical model (0.784, 0.765, 0.731) and WholeRad model (0.817, 0.810, 0.773). The Habitat-TDA model improves the accuracy and interpretability of preoperative predictions of PDAC grading, providing a promising tool for personalised management.

High Ki-67 expression in clear cell renal cell carcinoma (ccRCC) predicts poor prognosis but requires postoperative assessment. In a multicenter retrospective study of 627 ccRCC patients, we developed and validated a multi-modal model, integrating multi-scale radiomics and deep learning (DL) features, for non-invasive, preoperative Ki-67 prediction. Using ensemble machine learning algorithms, unimodal models were constructed from preoperative CT-derived multi-scale radiomics (intratumoral, habitat, peritumoral), 2D/3D DL, and clinical features. A stacking strategy was used to fuse the best-performing unimodal models. The fusion model demonstrated superior performance, achieving an Area Under the Curve (AUC) of 0.756 (95% CI 0.692-0.821) in the external test set. The model demonstrated excellent calibration and the highest clinical net benefit, with habitat radiomics identified as the dominant predictive component via SHAP analysis. Our validated multi-modal model significantly improves the preoperative prediction of Ki-67 expression compared to unimodal approaches, offering a promising tool to guide individualized surgical and surveillance strategies.

Deep learning (DL) has been proposed for magnetic resonance imaging (MRI) prostate segmentation for various clinical tasks, including radiotherapy treatment planning. In other applications, DL models have exhibited performance bias by protected attributes such as race. To investigate possible race bias in prostate MRI segmentation, DL models were trained on five clinical T2-weighted MRI datasets with varying White/Black race imbalance, plus one public dataset with unknown races, and evaluated on 32 White/Black matched clinical subjects. For the models trained with differing levels of race imbalance, the best performance for both races was when the training set was race-balanced. A linear mixed-effects model analysis showed that Dice Similarity Coefficient (DSC) differences between Black and White subjects depended on race representation in the training data, with a slight reduction in White-Black performance gap as Black representation increased (p < 0.05). The model trained on public data showed no difference in performance between races for DSC. The findings reveal the potential for race bias in DL prostate MRI segmentation performance when training sets are highly imbalanced. We argue for transparency in race reporting in DL prostate segmentation training data and reporting of test performance across demographic groups, with appropriate ethical/legal safeguards.

Training and refining both custom and pre-trained convolutional neural network (CNN) models for calculation of intracochlear positional index (ICPI) is as effective as manual calculation. The ICPI is a position factor that is known to influence cochlear implant performance, however manual calculation on computed tomography (CT) imaging is labour-intensive and prone to calculation errors. Automation of this process with machine learning via a custom built CNN model aims to reduce the difficulty in obtaining this position factor. Increasing the number of training epochs will improve accuracy. Our study aims to develop a validated CNN for ICPI calculation, which may improve surgical electrode positioning. Custom built CNN model and pre-trained ResNet 50 model trained and validated on 34 images, and tested on eight CT images of temporal bones with cochlear implants. The ground truth was manually established by calculating the distance from modiolus to electrode (DE) and lateral wall (DL), and applied to derive the ICPI. The pre-trained ResNet-50 model outperformed the custom-built CNN, with improvement statistically significant on evaluation metrics. The ResNet-50 model has lower mean absolute error and root mean squared error (RMSE). In both models, increasing the number of training epochs from ten to 100 improves accuracy of the ICPI calculation. Our machine learning models successfully achieved automation of ICPI calculation, with increasing accuracy increasing training epochs to 100 iterations. Future studies should explore optimizing these models and validating them on broader datasets to enhance their applicability in real-world scenarios by comparison to speech and audiometric outcomes.

One of the leading killers globally now is lung cancer. It ranks high among the dangerous malignant tumors that people may have. It is the leading cause of cancer-related fatalities in both men and women globally, and its mortality rate is higher than that of any other malignant tumor. Medical image segmentation has seen the successful use of many deep learning framework-based techniques in recent years. The use of computer tomography (CT) has increased in the detection of lung cancer, a significant malignancy. Patients with lung cancer have a better chance of surviving if the disease is detected early. Early diagnosis allows professionals to deliver appropriate therapy within a specific period, which in turn reduces the fatality rate. The healthcare industry benefits greatly from the advanced services deep learning models offer. In this study, we present a deep neural network architecture named as lung swarm net that combines DenseNet201 with PSO to classify lung cancer from CT scans of the lung. We used ResNet50 for the segmentation procedure and DenseNet-201 with Particle Swarm Optimization (PSO) for the classification in order to identify lung cancer. According to the experimental data, the suggested model outperforms other current models in terms of accuracy.

Neoadjuvant therapy (NAC) is a standard treatment for breast cancer, yet only some patients gain significant benefit. Identifying those most likely to benefit from NAC is crucial. Single-modality data often overlook patient heterogeneity, so we developed an interpretable, attention-based multimodal full information feature fusion transformer, MuFi, to predict NAC responses by integrating whole slide images (WSI) and magnetic resonance imaging (MRI). Data from 567 biopsy-confirmed breast cancer patients from two institutions were retrospectively analyzed, with a training cohort (n = 290), validation cohort (n = 73), and external test cohort (n = 204). Multimodal data included pre-treatment pathology slides, MRI scans, and clinical information. A memory-efficient multimodal model was used to fuse WSIs and MRI, with a transformer capturing interactions between histological patches and MRI features. MuFi achieved AUCs of 81.9% and 78.5% in discovery and validation cohorts and 79.3% in external testing, outperforming clinical, single-modality and late-fusion-based models. Integrating clinical data (cT and molecular subtype) with MuFi and Feature Re-calibration based Multiple Instance Learning (FRMIL) models further increased AUCs to 90.2%, 81.8%, and 81.6% across the cohorts, indicating enhanced predictive accuracy and generalizability, especially in external testing. By fusing pathology and radiology features, MuFi improves decision reliability and identifies critical multimodal predictors. This integration framework better captures patient heterogeneity, supporting personalized NAC decision-making through improved accuracy and generalizability.