Quantitative susceptibility mapping (QSM) provides the spatial distribution of magnetic susceptibility within tissues through sequential steps: phase unwrapping and echo combination, mask generation, background field removal, and dipole inversion. Accurate mask generation is crucial, as masks excluding regions outside the brain and without holes are necessary to minimize errors and streaking artifacts during QSM reconstruction. Variations in susceptibility values can arise from different mask generation methods, highlighting the importance of optimizing mask creation. In this study, we propose QSMmask-net, a deep neural network-based method for generating precise QSM masks. QSMmask-net achieved the highest Dice score compared to other mask generation methods. Mean susceptibility values using QSMmask-net masks showed the lowest differences from manual masks (ground truth) in simulations and healthy controls (no significant difference, p > 0.05). Linear regression analysis confirmed a strong correlation with manual masks for hemorrhagic lesions (slope = 0.9814 ± 0.007, intercept = 0.0031 ± 0.001, R² = 0.9992, p < 0.05). We have demonstrated that mask generation methods can affect the susceptibility value estimations. QSMmask-net reduces the labor required for mask generation while providing mask quality comparable to manual methods. The proposed method enables users without specialized expertise to create optimized masks, potentially broadening QSM applicability efficiently.

The intracellular pH (pH_i) is critical for understanding various pathologies, including brain tumors. While conventional pH_i measurement through ³¹P-MRS suffers from low spatial resolution and long scan times, ¹H-based APT-CEST imaging offers higher resolution with shorter scan times. This study aims to directly predict ³¹P-pH_i maps from CEST data by using a fully connected neuronal network. Fifteen tumor patients were scanned on a 3-T Siemens PRISMA scanner and received ¹H-based CEST and T1 measurement, as well as ³¹P-MRS. A neural network was trained voxel-wise on CEST and T1 data to predict ³¹P-pH_i values, using data from 11 patients for training and 4 for testing. The predicted pH_i maps were additionally down-sampled to the original the ³¹P-pH_i resolution, to be able to calculate the RMSE and analyze the correlation, while higher resolved predictions were compared with conventional CEST metrics. The results demonstrated a general correspondence between the predicted deepCEST pH_i maps and the measured ³¹P-pH_i in test patients. However, slight discrepancies were also observed, with a RMSE of 0.04 pH units in tumor regions. High-resolution predictions revealed tumor heterogeneity and features not visible in conventional CEST data, suggesting the model captures unique pH information and is not simply a T1 segmentation. The deepCEST pH_i neural network enables the APT-CEST hidden pH-sensitivity and offers pH_i maps with higher spatial resolution in shorter scan time compared with ³¹P-MRS. Although this approach is constrained by the limitations of the acquired data, it can be extended with additional CEST features for future studies, thereby offering a promising approach for 3D pH imaging in a clinical environment.

To develop and evaluate the performance of fully automated model based on deep learning and multiple logistics regression algorithm for image quality assessment (IQA) of carotid computed tomography angiography (CTA) images. This study retrospectively collected 840 carotid CTA images from four tertiary hospitals. Three radiologists independently assessed the image quality using a 3-point Likert scale, based on the degree of noise, vessel enhancement, arterial vessel contrast, vessel edge sharpness, and overall diagnostic acceptability. An automated assessment model was developed using a training dataset consisting of 600 carotid CTA images. The assessment steps included: (i) selection of objective representative slices; (ii) use of 3D Res U-net approach to extract objective indices from the representative slices and (iii) use of single objective index and multiple indices combinedly to develop logistic regression models for IQA. In the internal and external test datasets (n = 240), the performance of models was evaluated using sensitivity, specificity, precision, F-score, accuracy, the area under the receiver operating characteristic curve (AUC), and the IQA results of models was compared with radiologists' consensus. The representative slices were determined based on the same length model. The performance of multi-index model was excellent in internal and external test datasets with AUCs of 0.98 and 0.97. And the consistency between model and radiologists achieved 91.8% (95% CI: 87.0-96.5) and 92.6% (95 % CI: 86.9-98.4) in internal and external test datasets respectively. The fully automated multi-index model showed equivalent performance to the subjective perceptions of radiologists with greater efficiency for IQA.

Osteopenia is a bone disorder that causes low bone density and affects millions of people worldwide. Diagnosis of this condition is commonly achieved through clinical assessment of bone mineral density (BMD). State of the art machine learning (ML) techniques, such as convolutional neural networks (CNNs) and transformer models, have gained increasing popularity in medicine. In this work, we employ six deep networks for osteopenia vs. healthy bone classification using X-ray imaging from the pediatric wrist dataset GRAZPEDWRI-DX. We apply two explainable AI techniques to analyze and interpret visual explanations for network decisions. Experimental results show that deep networks are able to effectively learn osteopenic and healthy bone features, achieving high classification accuracy rates. Among the six evaluated networks, DenseNet201 with transfer learning yielded the top classification accuracy at 95.2 %. Furthermore, visual explanations of CNN decisions provide valuable insight into the blackbox inner workings and present interpretable results. Our evaluation of deep network classification results highlights their capability to accurately differentiate between osteopenic and healthy bones in pediatric wrist X-rays. The combination of high classification accuracy and interpretable visual explanations underscores the promise of incorporating machine learning techniques into clinical workflows for the early and accurate diagnosis of osteopenia.

Deep complex-valued neural networks (CVNNs) provide a powerful way to leverage complex number operations and representations and have succeeded in several phase-based applications. However, previous networks have not fully explored the impact of complex-valued networks in the frequency domain. Here, we introduce a unified complex-valued deep learning framework - Artificial Fourier Transform Network (AFTNet) - which combines domain-manifold learning and CVNNs. AFTNet can be readily used to solve image inverse problems in domain transformation, especially for accelerated magnetic resonance imaging (MRI) reconstruction and other applications. While conventional methods typically utilize magnitude images or treat the real and imaginary components of k-space data as separate channels, our approach directly processes raw k-space data in the frequency domain, utilizing complex-valued operations. This allows for a mapping between the frequency (k-space) and image domain to be determined through cross-domain learning. We show that AFTNet achieves superior accelerated MRI reconstruction compared to existing approaches. Furthermore, our approach can be applied to various tasks, such as denoised magnetic resonance spectroscopy (MRS) reconstruction and datasets with various contrasts. The AFTNet presented here is a valuable preprocessing component for different preclinical studies and provides an innovative alternative for solving inverse problems in imaging and spectroscopy. The code is available at: https://github.com/yanting-yang/AFT-Net.

A cardiac digital twin is a virtual replica of a patient-specific heart, mimicking its anatomy and physiology. A crucial step of building a cardiac digital twin is anatomical twinning, where the computational mesh of the digital twin is tailored to the patient-specific cardiac anatomy. In a number of studies, the effect of anatomical variation on clinically relevant functional measurements like electrocardiograms (ECGs) is investigated, using computational simulations. While such a simulation environment provides researchers with a carefully controlled ground truth, the impact of anatomical differences on functional measurements in real-world patients remains understudied. In this study, we develop a biventricular statistical shape model and use it to quantify the effect of biventricular anatomy on ECG-derived and demographic features, providing novel insights for the development of digital twins of cardiac electrophysiology. To this end, a dataset comprising high-resolution cardiac CT scans from 271 healthy individuals, including athletes, is utilized. Furthermore, a novel, universal, ventricular coordinate-based method is developed to establish lightweight shape correspondence. The performance of the shape model is rigorously established, focusing on its dimensionality reduction capabilities and the training data requirements. The most important variability in healthy ventricles captured by the model is their size, followed by their elongation. These anatomical factors are found to significantly correlate with ECG-derived and demographic features. Additionally, a comprehensive synthetic cohort is made available, featuring ready-to-use biventricular meshes with fiber structures and anatomical region annotations. These meshes are well-suited for electrophysiological simulations.

To validate a liver lesion detection and classification model using staging computed tomography (CT) scans of colorectal cancer (CRC) patients. A UNet-based deep learning model was trained on 272 public liver tumour CT scans and tested on 220 CRC staging CTs acquired from a single institution (2014-2019). Performance metrics included lesion detection rates by size (<10 mm, 10-20 mm, >20 mm), segmentation accuracy (dice similarity coefficient, DSC), volume measurement agreement (Bland-Altman limits of agreement, LOAs; intraclass correlation coefficient, ICC), and classification accuracy (malignant vs benign) at patient and lesion levels (detected lesions only). The model detected 743 out of 884 lesions (84%), with detection rates of 75%, 91.3%, and 96% for lesions <10 mm, 10-20 mm, and >20 mm, respectively. The median DSC was 0.76 (95% CI: 0.72-0.80) for lesions <10 mm, 0.83 (95% CI: 0.79-0.86) for 10-20 mm, and 0.85 (95% CI: 0.82-0.88) for >20 mm. Bland-Altman analysis showed a mean volume bias of -0.12 cm³ (LOAs: -1.68 to +1.43 cm³), and ICC was 0.81. Lesion-level classification showed 99.5% sensitivity, 65.7% specificity, 53.8% positive predictive value (PPV), 99.7% negative predictive value (NPV), and 75.4% accuracy. Patient-level classification had 100% sensitivity, 27.1% specificity, 59.2% PPV, 100% NPV, and 64.5% accuracy. The model demonstrates strong lesion detection and segmentation performance, particularly for sub-centimetre lesions. Although classification accuracy was moderate, the 100% NPV suggests strong potential as a CRC staging screening tool. Future studies will assess its impact on radiologist performance and efficiency.

The image quality of single-energy CT (SECT) limited the accuracy of automatic segmentation. Dual-energy CT (DECT) may potentially improve automatic segmentation yet the performance and strategy have not been investigated thoroughly. Based on DECT-generated virtual monochromatic images (VMIs), this study proposed a novel deep learning model (MIAU-Net) and evaluated the segmentation performance on the head organs-at-risk (OARs). The VMIs from 40 keV to 190 keV were retrospectively generated at intervals of 10 keV using the DECT of 46 patients. Images with expert delineation were used for training, validation, and testing MIAU-Net for automatic segmentation. Theperformance of MIAU-Net was compared with the existingU-Net, Attention-UNet, nnU-Net and TransFuse methods based on Dice Similarity Coefficient (DSC). Correlationanalysis was performed to evaluate and optimize the impact of different virtual energies on the accuracy of segmentation. Using MIAU-Net, average DSCs across all virtual energy levels were 93.78 %, 81.75 %, 84.46 %, 92.85 %, 94.40 %, and 84.75 % for the brain stem, optic chiasm, lens, mandible, eyes, and optic nerves, respectively, higher than the previous publications using SECT. MIAU-Net achieved the highest average DSC (88.84 %) and the lowest parameters (14.54 M) in all tested models. The results suggested that 60 keV-80 keV is the optimal VMI energy level for soft tissue delineation, while 100 keV is optimal for skeleton segmentation. This work proposed and validated a novel deep learning model for automatic segmentation based on DECT, suggesting potential advantages and OAR-specific optimal energy of using VMIs for automatic delineation.

Artificial intelligence (AI) is transforming multidisciplinary oncology at an unprecedented pace, redefining how clinicians detect, classify, and treat cancer. From earlier and more accurate diagnoses to personalized treatment planning, AI's impact is evident across radiology, pathology, radiation oncology, and medical oncology. By leveraging vast and diverse data-including imaging, genomic, clinical, and real-world evidence-AI algorithms can uncover complex patterns, accelerate drug discovery, and help identify optimal treatment regimens for each patient. However, realizing the full potential of AI also necessitates addressing concerns regarding data quality, algorithmic bias, explainability, privacy, and regulatory oversight-especially in low- and middle-income countries (LMICs), where disparities in cancer care are particularly pronounced. This study provides a comprehensive overview of how AI is reshaping cancer care, reviews its benefits and challenges, and outlines ethical and policy implications in line with ASCO's 2025 theme, <i>Driving Knowledge to Action.</i> We offer concrete calls to action for clinicians, researchers, industry stakeholders, and policymakers to ensure that AI-driven, patient-centric oncology is accessible, equitable, and sustainable worldwide.

Mitral valve (MV) assessment is key to diagnosing valvular disease and to addressing its serious downstream complications. Cardiac magnetic resonance (CMR) has become an essential diagnostic tool in MV disease, offering detailed views of the valve structure and function, and overcoming the limitations of other imaging modalities. Automated detection of the MV leaflets in CMR could enable rapid and precise assessments that enhance diagnostic accuracy. To address this gap, we introduce DeepValve, the first deep learning (DL) pipeline for MV detection using CMR. Within DeepValve, we tested three valve detection models: a keypoint-regression model (UNET-REG), a segmentation model (UNET-SEG) and a hybrid model based on keypoint detection (DSNT-REG). We also propose metrics for evaluating the quality of MV detection, including Procrustes-based metrics (UNET-REG, DSNT-REG) and customized Dice-based metrics (UNET-SEG). We developed and tested our models on a clinical dataset comprising 120 CMR images from patients with confirmed MV disease (mitral valve prolapse and mitral annular disjunction). Our results show that DSNT-REG delivered the best regression performance, accurately locating landmark locations. UNET-SEG achieved satisfactory Dice and customized Dice scores, also accurately predicting valve location and topology. Overall, our work represents a critical first step towards automated MV assessment using DL in CMR and paving the way for improved clinical assessment in MV disease.