This study investigated dynamic brain network changes and their genetic correlations in children with idiopathic generalized epilepsy (IGE). We included 26 children with IGE and 35 healthy controls, all participants underwent resting-state functional magnetic resonance imaging and cognitive assessments. Modular variability (MV) in time-varying networks was compared, and correlations with cognition and clinical variables were analyzed, we also explored classification problems using machine learning. Gene sets associated with IGE-related network remodeling were identified using the Allen Human Brain Atlas and gene enrichment analysis tools. The results showed that children with IGE exhibited reduced MV in sensorimotor and frontoparietal networks and increased MV in the default mode network (DMN). MV changes in the left prefrontal and right orbitofrontal cortices correlated with verbal and full-scale IQ scores, respectively. MV changes in the left precuneus/posterior cingulate cortex correlated with performance IQ scores. Transcriptomic analysis revealed 985 genes (FDR < 0.05) whose spatial expression patterns covaried with network alterations, prominently enriched for synaptic signaling and neuroactive ligand-receptor interactions, including GABA receptor subunits (GABRE) and neurodevelopmental regulators (BCL11A). Machine learning confirmed MV as a significant predictor of verbal IQ (permutation P = 0.041), with DMN and frontoparietal regions contributing most to prediction. Dynamic brain network abnormalities in children with IGE were significantly associated with cognitive function and gene expression, providing new insights into the neural mechanisms underlying network dysfunction and cognitive impairment in epilepsy.

Brain tumors are among the most fatal diseases, Often leading to a reduction in life expectancy. Early and accurate diagnosis is essential to guide effective treatment and enhance survival rates. Advances in artificial intelligence, particularly deep learning with Convolutional Neural Networks (CNNs), have revolutionized medical imaging analysis by enabling automated and precise diagnostic tools. This study investigates the effectiveness of deep transfer learning for brain tumor classification using a publicly available 7023 image Brain Tumor MRI Dataset (Figshare, SARTAJ, Br35H), and split into training, validation, and test sets, categorizing four classes : glioma, meningioma, pituitary tumor, and no tumor. The proposed method employs a fine-tuned ResNet-34 model, enhanced with custom classification head, data augmentation techniques, and the Ranger optimizer (combining RAdam and Lookahead for stable convergence). The model achieved an accuracy of 99.66% , surpassing current state-of-the-art approaches.

Pulmonary blood volumes (PBV), currently not assessed by computed tomography pulmonary angiography (CTPA), could provide additional information to routine investigations performed for chronic thromboembolic pulmonary hypertension (CTEPH). We investigated CTPA-based PBV in evaluating hemodynamic outcome from pulmonary endarterectomy (PEA) surgery. A deep learning-based CTPA vascular segmentation model, differentiating arteries and veins, was applied for automated PBV measurements in CTEPH patients who underwent PEA at UK's national CTEPH service. Pulmonary arteries were compartmentalised into "central" (main pulmonary and proximal lobar) and "intrapulmonary". Mean pulmonary arterial pressure >30 mmHg post-PEA defined "clinically relevant" residual PH. Logistic regression models applying CTPA-based PBV to identify residual PH were trained and tested on the discovery and validation cohorts respectively. Paired pre- and postoperative CTPA, in the discovery (n=71) and validation (n=102) cohorts showed that central pulmonary artery volume and total artery to vein volume ratio (A-VR) decreased and pulmonary vein volume increased with hemodynamic improvement post-PEA. Preoperative central pulmonary artery volume and A-VR helped identify patients at risk for clinically relevant residual PH post-PEA (AUROC 0.88 and 0.82 in the discovery and validation cohorts). Postoperative central pulmonary artery volume, A-VR and pulmonary vein volume helped to non-invasively identify patients without clinically relevant residual PH (AUROC 0.91 and 0.88 in the discovery and validation cohorts). Automated quantification of CTPA-based PBV at diagnosis can help stratify risk for residual PH in patients managed with PEA. Utilizing CTPA-derived PBV post-PEA to identify patients without residual PH can potentially reduce the need for routine postoperative right heart catheterization.

Aortic dissection (AD) is a life-threatening condition. We developed an artificial intelligence (AI) algorithm capable of robust, accurate, and automated AD detection and sub-classification. Based on 2010-2023 data from Mannheim University Medical Centre, heterogeneous internal training cases with confirmed AD (n = 70) were manually segmented and, together with non-AD cases (n = 87), used for training of a convolutional neural network (CNN; U-Net architecture) configured using the nnU-Net framework. Internal test dataset was composed of 106 cases. The external test was performed on a public dataset: 100 AD cases from ImageTBAD, Guangdong Provincial People's Hospital, China, and 38 non-AD cases from the AVT dataset (multiple sources). Model performance was evaluated by area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), sensitivity, specificity, precision, and F1-score, and by investigating performance on different subsets of cases. Confidence intervals were determined using DeLong's method and bootstrapping. The best-performing algorithm achieved an AUROC of 98.7% (95% CI: 96.1-100.0%) and an AUPRC of 98.9% (96.0-100.0%) on the internal test dataset, 97.0% (94.7-99.3%) and 99.06% (98.0-99.7%) on the external test datasets, respectively. In the internal test dataset, of 15 unsuspected AD cases, 14 (93.3%) were successfully detected by the algorithm. On the external test dataset, sensitivity, specificity, precision, and F1-score were 92.0%, 100.0%, 100.0%, and 95.8%, respectively. The developed AI pipeline highlighted the capability of optimized CNNs to reliably detect AD across heterogeneous multicenter datasets. The resulting tool will be made publicly available for further scientific evaluation. Artificial Intelligence demonstrated promising potential to detect AD on heterogeneous thoracic CT imaging data. Early detection of aortic dissection (AD) is crucial for timely treatment. A modern convolutional neural network (CNN) achieved 93.5% sensitivity and 100.0% specificity for AD detection on multicenter, heterogeneous CT data. These results demonstrate the potential of streamlined, optimized CNNs for robust AD detection on CT, supporting fast clinical response.

Global interpretability in machine learning holds great potential for extracting meaningful insights from neuroimaging data to improve our understanding of brain function. Although various approaches exist to identify key contributing features at both local and global levels, the high dimensionality and correlations in neuroimaging data require careful selection of interpretability methods to achieve reliable global insights into brain function using machine learning. In this study, we evaluate multiple interpretability techniques such as SHAP, which relies on feature independence, as well as recent advances that account for feature dependence in the context of global interpretability, and inherently global methods such as SAGE. To demonstrate the practical application, we trained XGBoost models to predict age and fluid intelligence using neuroimaging measures from the UK Biobank dataset. By applying these interpretability methods, we found that mean intensities in subcortical regions are consistently and significantly associated with brain aging, while the prediction of fluid intelligence is driven by contributions of the hippocampus and the cerebellum, alongside established regions such as the frontal and temporal lobes. These results underscore the value of interpretable machine learning methods in understanding brain function through a data-driven approach.

This study aims to understand the supraspinal regulation of balance control in chronic ankle instability (CAI) by characterizing the large-scale communication and interaction via brain functional network topology in CAI and establish the association between topological properties and dynamic balance performance. In this cross-sectional design study, 40 CAI individuals and 39 healthy control (HC) individuals were enrolled. To assess the dynamic balance, the Y-balance test was utilised. To explore the topological structure of brain networks, graph theory was used to analyse resting-state functional MRI data. The CAI group had lower normalized reach distances in the Y-balance test than HC. Compared to HC, CAI exhibited remarkably lower nodal degree centrality (Dc) and higher nodal shortest path length (NLp) within the sensorimotor network (SMN), particularly in the precentral gyrus, temporal cortex, and pre-supplementary motor area of the right hemisphere. CAI showed reduced NLp and increased nodal efficiency in the posterior cingulate cortex of the left hemisphere, a hub region of the default mode subnetwork (DMN). In CAI, high Dc and low NLp in the precentral gyrus of the right hemisphere were substantially correlated to poor performance of the Y-balance test, but not in HC. CAI individuals demonstrated diminished regional processing capability within the SMN and a potential compensatory increase in nodal efficiency within the DMN, which are critical to maintain safe balance in this cohort. These alterations in supraspinal networks could be an effective target for rehabilitation and management in CAI.

Cardiac quantitative MRI (qMRI) is a powerful imaging technique for diagnosing pathologies such as diffuse myocardial fibrosis. One main challenge is cardiac motion, which requires synchronization of data acquisition with the heartbeat, leading to long scan times. We present a novel deep learning-based image registration method for cardiac qMRI that enables non-rigid motion correction of data acquired continuously over multiple cardiac cycles, thereby reducing scan times. Our method is a zero-shot approach that utilizes the physical qMRI signal model for accurate motion estimation. Non-rigid motion of dynamic images is estimated with a U-Net-based architecture. This exploits the intrinsic smoothness of cardiac motion, allowing sharing information between neighboring images. The approach is robust to undersampling artifacts, enabling motion estimation from dynamic images reconstructed from very few k-space data even without advanced image reconstruction methods. We evaluated the method for fast cardiac T1 mapping using a Golden radial sampling scheme on numerical simulations and in-vivo acquisitions. On numerical simulations, our method achieved a 61.64% improvement in T1 accuracy. On in-vivo data, our approach yielded a 45.13% improvement in sharpness of T1 maps, and temporal image alignment of motion-corrected dynamics improved on average by 11.78%. Our method enables accurate non-rigid motion correction of highly undersampled cardiac qMRI data obtained from continuously acquired data. As our method is individually optimized for each scan without the need for training on large datasets, it can easily be adapted to other cardiac qMRI approaches.

Periprocedural aspirin or unfractionated heparin during endovascular treatment in acute ischemic stroke increases symptomatic intracranial hemorrhage (sICH) risk without improving functional outcome. White matter lesions (WMLs) are associated with higher sICH risk and poor functional outcome following stroke. We aimed to assess whether WML volume modifies the effect of aspirin or heparin. In this post-hoc analysis of the MR CLEAN-MED trial, WML volume was automatically determined using deep learning-based segmentation on baseline non-contrast CT scans. Outcomes included good functional outcome (modified Rankin Scale 0-2 at 90 days), any ICH, asymptomatic ICH (aICH), and sICH. Patients received either aspirin or not, and either heparin or not. Multivariable logistic regression evaluated treatment effect and effect modification. Of 628 patients, 614 with baseline CT were included. Median WML volume was 0.59 mL without significant differences between treatment arms. WML volume significantly modified the effect of aspirin on sICH (p = 0.01), but not on functional outcome (p = 0.95), any ICH (p = 0.52), or aICH (p = 0.30). Aspirin was associated with increased sICH risk, which decreased with increasing WML volume (aOR 0.96 [95% CI: 0.93-0.99] per 1 mL). For patients with large WML volumes, aspirin showed no significant effect on sICH risk. The effect of heparin on functional outcome, any ICH, aICH, and sICH was not modified by WML volume (p = 0.53, p = 0.26, p = 0.08, p = 0.63, respectively). WML volume significantly modified the effect of aspirin on sICH risk, with aspirin-associated risk decreasing as WML volume increased. WML volume did not modify the effect of aspirin or heparin on other outcomes. WML volume on non-contrast CT modifies the effect of aspirin during endovascular thrombectomy on sICH risk, yet no WML-based patient subgroup showed save benefits from periprocedural aspirin or heparin treatment. Periprocedural aspirin and unfractionated heparin during endovascular treatment cause a higher hemorrhage risk. WML volume is associated with worse functional outcome and WML volume significantly modifies the effect of aspirin on symptomatic hemorrhage risk, with aspirin-associated risk decreasing with increasing WML volume. No WML-volume-based patient subgroup was identified where aspirin or heparin treatment demonstrated safe clinical benefit.

Body composition has a significant role to predict survival in patients with malignant disease. This study evaluates the importance of body composition for predicting short- and long-term survival after liver transplantation. Additionally, the sex specific differences will be evaluated. Body composition, of all patients who underwent liver transplantation between January 2011 and December 2023 with computed tomography prior liver transplantation, was assessed fully automated with AI based technique. Pre-, intra- and postoperative data were retrospectively reported. Uni- and multivariate regression analyses was performed to identify independent prognostic factors for survival. The statistical analyses was performed separately for male and female with comparison of the both groups. There were 346 patients (60.1%male, 39.9%female) with median age of 52.2 years (SD 10.8) included to the study. The univariate and multivariate cox regression analyses have identified the ratio of the subcutaneous fat volume to muscle volume as well as the ratio of the visceral fat volume to muscle volume as significant prognostic parameter for the overall survival. The separate analyses of the two groups show that these factors predict survival in male and female. However, visceral fat and also the ratio of FVM is significantly higher in male. Based on the results of our study we can conclude that the ratio of visceral fat volume to muscle volume (FVM-ratio) has an essential impact on overall survival after liver transplantation in male and female patients. The fully automated AI based assessment is fast, accurate and investigator independent.

Large-scale radiographic datasets often include errors in labels such as body parts or projection, which can undermine automated image analysis. Therefore, we aimed to develop and externally validate two deep-learning models-one for categorising radiographs by body part, and another for identifying projection and rotation of chest radiographs-using large, diverse datasets. We retrospectively collected radiographs from multiple institutions and public repositories. For the first model (Xp-Bodypart-Checker), we included seven categories (Head, Neck, Chest, Incomplete Chest, Abdomen, Pelvis, Extremities). For the second model (CXp-Projection-Rotation-Checker), we classified chest radiographs by projection (anterior-posterior, posterior-anterior, lateral) and rotation (upright, inverted, left rotation, right rotation). Both models were trained, tuned, and internally tested on separate data, then externally tested on radiographs from different institutions. Model performance was assessed using overall accuracy (micro, macro, and weighted) as well as one-vs.-all area under the receiver operating characteristic curve (AUC). In the Xp-Bodypart-Checker development phase, we included 429,341 radiographs obtained from Institutions A, B, and MURA. In the CXp-Projection-Rotation-Checker development phase, we included 463,728 chest radiographs from CheXpert, PadChest, and Institution A. The Xp-Bodypart-Checker achieved AUC values of 1.00 (99% CI: 1.00-1.00) for all classes other than Incomplete Chest, which had an AUC value of 0.99 (99% CI: 0.98-1.00). The CXp-Projection-Rotation-Checker demonstrated AUC values of 1.00 (99% CI: 1.00-1.00) across all projection and rotation classifications. These models help automatically verify image labels in large radiographic databases, improving quality control across multiple institutions. Question This study examines how deep learning can accurately classify radiograph body parts and detect chest projection/orientation in large, multi-institutional datasets, enhancing metadata consistency for clinical and research workflows. Findings Xp-Bodypart-Checker classified radiographs into seven categories with AUC values of over 0.99 for all classes, while CXp-Projection-Rotation-Checker achieved AUC values of 1.00 across all projection and rotation classifications. Clinical relevance Trained on over 860,000 multi-institutional radiographs, our two deep-learning models classify radiograph body-part and chest radiograph projection/rotation, identifying mislabeled data and enhancing data integrity, thereby improving reliability for both clinical use and deep-learning research.