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.

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.

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.

Accurate and timely diagnosis of brain tumors is critical for patient management and treatment planning. Magnetic resonance imaging (MRI) is a widely used modality for brain tumor detection and characterization, often aided by the administration of gadolinium-based contrast agents (GBCAs) to improve tumor visualization. Recently, deep learning models have shown remarkable success in predicting contrast-enhancement in medical images, thereby reducing the need of GBCAs and potentially minimizing patient discomfort and risks. In this paper, we present a study aimed at investigating the generalization capabilities of a neural network trained to predict full contrast in brain tumor images from noncontrast MRI scans. While initial results exhibited promising performance on a specific tumor type at a certain stage using a specific dataset, our attempts to extend this success to other tumor types and diverse patient populations yielded unexpected challenges and limitations. Through a rigorous analysis of the factor contributing to these negative results, we aim to shed light on the complexities associated with generalizing contrast enhancement prediction in medical brain tumor imaging, offering valuable insights for future research and clinical applications.

Two of the most common complaints seen in neurology clinics are Alzheimer's disease (AD) and mild cognitive impairment (MCI), characterized by similar symptoms. The aim of this study was to develop and internally validate the diagnostic value of combined neurological and radiological predictors in differentiating mild AD from MCI as the outcome variable, which helps in preventing AD development. A cross-sectional study of 161 participants was conducted in a general healthcare setting, including 30 controls, 71 mild AD, and 60 MCI. Binary logistic regression was used to identify predictors of interest, with collinearity assessment conducted prior to model development. Model performance was assessed through calibration, shrinkage, and decision-curve analyses. Finally, the combined clinical and radiological model was compared to models utilizing only clinical or radiological predictors. The final model included age, sex, education status, Montreal cognitive assessment, Global Cerebral Atrophy Index, Medial Temporal Atrophy Scale, mean hippocampal volume, and Posterior Parietal Atrophy Index, with the area under the curve of 0.978 (0.934-0.996). Internal validation methods did not show substantial reduction in diagnostic performance. Combined model showed higher diagnostic performance compared to clinical and radiological models alone. Decision curve analysis highlighted the usefulness of this model for differentiation across all probability levels. A combined clinical-radiological model has excellent diagnostic performance in differentiating mild AD from MCI. Notably, the model leveraged straightforward neuroimaging markers, which are relatively simple to measure and interpret, suggesting that they could be integrated into practical, formula-driven diagnostic workflows without requiring computationally intensive deep learning models.

Overuse of radiography studies and underuse of conservative therapies for cranial deformities in neonates is a known inefficiency in pediatric craniofacial healthcare. This study sought to establish whether the introduction of artificial intelligence (AI)-generated craniometrics and craniometric interpretations into craniofacial clinical workflow improved resource utilization patterns in the initial evaluation and management of neonatal cranial deformities. A retrospective chart review of pediatric patients referred for head shape concerns between January 2019 and June 2023 was conducted. Patient demographics, final encounter diagnosis, review of an AI analysis, and provider orders were documented. Patients were divided based on whether an AI cranial deformity analysis was documented as reviewed during the index evaluation, then both groups were propensity matched. Rates of index-encounter radiology studies, physical therapy (PT), orthotic therapy, and craniofacial specialist follow-up evaluations were compared using logistic regression and ANOVA analyses. One thousand patient charts were reviewed (663 conventional encounters, 337 AI-assisted encounters). One-to-one propensity matching was performed between these groups. AI models were significantly more likely to be reviewed during telemedicine encounters and advanced practice provider (APP) visits (54.8% telemedicine vs 11.4% in-person, p < 0.0001; 12.3% physician vs 44.4% APP, p < 0.0001). All AI diagnoses of craniosynostosis versus benign deformities were congruent with final diagnoses. AI model review was associated with a significant increase in the use of orthotic therapies for neonatal cranial deformities (31.5% vs 38.6%, p = 0.0132) but not PT or specialist follow-up evaluations. Radiology ordering rates did not correlate with AI-interpreted data review. As neurosurgeons and pediatricians continue to work to limit neonatal radiation exposure and contain healthcare costs, AI-assisted clinical care could be a cheap and easily scalable diagnostic adjunct for reducing reliance on radiography and encouraging adherence to established clinical guidelines. In practice, however, providers appear to default to preexisting diagnostic biases and underweight AI-generated data and interpretations, ultimately negating any potential advantages offered by AI. AI engineers and specialty leadership should prioritize provider education and user interface optimization to improve future adoption of validated AI diagnostic tools.

There is an urgent need for the precise prediction of cerebral amyloidosis using noninvasive and accessible indicators to facilitate the early diagnosis of individuals with the preclinical stage of Alzheimer's disease (AD). Two hundred and four individuals with subjective cognitive decline (SCD) were enrolled in this study. All subjects completed neuropsychological assessments and underwent 18F-florbetapir PET, structural MRI, and functional MRI. A total of 315 features were extracted from the MRI, demographics, and neuropsychological scales and selected using the least absolute shrinkage and selection operator (LASSO). The logistic regression (LR) model, based on machine learning, was trained to classify SCD as either β-amyloid (Aβ) positive or negative. A nomogram was established using a multivariate LR model to predict the risk of Aβ+. The performance of the prediction model and nomogram was assessed with area under the curve (AUC) and calibration. The final model was based on the right rostral anterior cingulate thickness, the grey matter volume of the right inferior temporal, the ReHo of the left posterior cingulate gyrus and right superior temporal gyrus, as well as MoCA-B and AVLT-R. In the training set, the model achieved a good AUC of 0.78 for predicting Aβ+, with an accuracy of 0.72. The validation of the model also yielded a favorable discriminatory ability with an AUC of 0.88 and an accuracy of 0.83. We have established and validated a model based on cognitive, sMRI, and fMRI data that exhibits adequate discrimination. This model has the potential to predict amyloid status in the SCD group and provide a noninvasive, cost-effective way that might facilitate early screening, clinical diagnosis, and drug clinical trials.

Adolescent major depressive disorder (MDD) is a serious mental health condition that has been linked to abnormal functional connectivity (FC) patterns within the brain. However, whether FC could be used as a potential biomarker for diagnosis of adolescent MDD is still unclear. The aim of our study was to investigate the potential diagnostic value of whole-brain FC in adolescent MDD. Resting-state functional magnetic resonance imaging data were obtained from 94 adolescents with MDD and 78 healthy adolescents. The whole brain was segmented into 90 regions of interest (ROIs) using the automated anatomical labeling atlas. FC was assessed by calculating the Pearson correlation coefficient of the average time series between each pair of ROIs. A multivariate pattern analysis was employed to classify patients from controls using the whole-brain FC as input features. The linear support vector machine classifier achieved an accuracy of 69.18% using the optimal functional connection features. The consensus functional connections were mainly located within and between large-scale brain networks. The top 10 nodes with the highest weight in the classification model were mainly located in the default mode, salience, auditory, and sensorimotor networks. Our findings highlighted the importance of functional network connectivity in the neurobiology of adolescent MDD, and suggested the possibility of altered FC and high-weight regions as complementary diagnostic markers in adolescents with depression.

Eating disorders (EDs), including anorexia nervosa (AN), bulimia nervosa (BN), and binge eating disorder (BED), are complex psychiatric conditions with high morbidity and mortality. Neuroimaging and machine learning (ML) represent promising approaches to improve diagnosis, understand pathophysiological mechanisms, and predict treatment response. This systematic review aimed to evaluate the application of ML techniques to neuroimaging data in EDs. Following PRISMA guidelines (PROSPERO registration: CRD42024628157), we systematically searched PubMed and APA PsycINFO for studies published between 2014 and 2024. Inclusion criteria encompassed human studies using neuroimaging and ML methods applied to AN, BN, or BED. Data extraction focused on study design, imaging modalities, ML techniques, and performance metrics. Quality was assessed using the GRADE framework and the ROBINS-I tool. Out of 185 records screened, 5 studies met the inclusion criteria. Most applied support vector machines (SVMs) or other supervised ML models to structural MRI or diffusion tensor imaging data. Cortical thickness alterations in AN and diffusion-based metrics effectively distinguished ED subtypes. However, all studies were observational, heterogeneous, and at moderate to serious risk of bias. Sample sizes were small, and external validation was lacking. ML applied to neuroimaging shows potential for improving ED characterization and outcome prediction. Nevertheless, methodological limitations restrict generalizability. Future research should focus on larger, multicenter, and multimodal studies to enhance clinical applicability. Level IV, multiple observational studies with methodological heterogeneity and moderate to serious risk of bias.

Predicting long-term clinical outcomes based on the early DSC PWI MRI scan is valuable for prognostication, resource management, clinical trials, and patient expectations. Current methods require subjective decisions about which imaging features to assess and may require time-consuming postprocessing. This study's goal was to predict multilabel 90-day modified Rankin Scale (mRS) score in acute ischemic stroke patients by combining ensemble models and different configurations of radiomic features generated from Dynamic susceptibility contrast perfusion-weighted imaging. In Follow-up studies, a total of 70 acute ischemic stroke (AIS) patients underwent magnetic resonance imaging within 24 hours poststroke and had a follow-up scan. In the single study, 150 DSC PWI Image scans for AIS patients. The DRF are extracted from DSC-PWI Scans. Then Lasso algorithm is applied for feature selection, then new features are generated from initial and follow-up scans. Then we applied different ensemble models to classify between three classes normal outcome (0, 1 mRS score), moderate outcome (2,3,4 mRS score), and severe outcome (5,6 mRS score). ANOVA and post-hoc Tukey HSD tests confirmed significant differences in model style performance across various studies and classification techniques. Stacking models consistently on average outperformed others, achieving an Accuracy of 0.68 ± 0.15, Precision of 0.68 ± 0.17, Recall of 0.65 ± 0.14, and F1 score of 0.63 ± 0.15 in the follow-up time study. Techniques like Bo_Smote showed significantly higher recall and F1 scores, highlighting their robustness and effectiveness in handling imbalanced data. Ensemble models, particularly Bagging and Stacking, demonstrated superior performance, achieving nearly 0.93 in Accuracy, 0.95 in Precision, 0.94 in Recall, and 0.94 in F1 metrics in follow-up conditions, significantly outperforming single models. Ensemble models based on radiomics generated from combining Initial and follow-up scans can be used to predict multilabel 90-day stroke outcomes with reduced subjectivity and user burden.