Magnetic Resonance Imaging Datasets Research Articles

An intelligent magnetic resonance imagining-based multistage Alzheimer's disease classification using swish-convolutional neural networks.

Alzheimer's disease (AD) refers to a neurological disorder that causes damage to brain cells and results in decreasing cognitive abilities and memory. In brain scans, this degeneration can be seen in different ways. The disease can be classified into four stages: Non-demented (ND), moderate demented (MoD), mild demented (MiD), and very mild demented (VMD). To prepare the raw dataset for analysis, the collected magnetic resonance imaging (MRI) images are subjected to several pre-processing techniques in order to improve the performance accuracy of the proposed model. Medical images generally have poor contrast and get affected by noise, which ends up with inaccurate diagnosis. For the different phases of AD to be detected, a clear image is necessary. To address this issue, the influence of the artefacts must be reduced, enhance the contrast, and reduce the loss of information. A novel framework for image enhancement is suggested to increase the accuracy in the detection and identification of AD. In this study, the raw MRI dataset from the Alzheimer's disease neuroimaging initiative (ADNI) database is subjected to skull stripping, contrast enhancement, and image filtering followed by data augmentation to balance the dataset with four types of Alzheimer's classes. The pre-processed data are subjected to five different pre-trained models such as AlexNet, ResNet, VGG 16, EfficientNet, and Inceptionv3 achieving a testing accuracy rate of 91.2%, 88.21%, 92.34%, 93.45%, and 85.12%, respectively. These pre-trained models are compared with the proposed CNN (convolutional neural network) model designed with Adam optimizer and Flatten Swish activation function which reaches the highest accuracy of 96.5% with a learning rate of 0.000001. The five pre-trained CNNmodels along with the proposed swish-based AD-CNN were tested using various performance metrics to evaluate the model efficiency in classifying and identifying the AD classes. From the result analysis, it is evident that the proposed AD-CNN model outperforms all the other models.

Brain tumor diagnosis in MRI scans images using Residual/Shuffle Network optimized by augmented Falcon Finch optimization.

Brain tumor diagnosis is an important task in prognosing and treatment planning of the patients with brain cancer. in the meantime, using the Magnetic Resonance Imaging (MRI) as a commonly used non-invasive imaging technique provide the experts a helpful view for detecting the brain tumors. While deep learning methods have shown significant success in analyzing medical images, they often require careful design of architecture and tuning of hyperparameters to achieve optimal results. This study presents a new approach for diagnosing brain tumors in MRI scans using deep learning, focusing on Residual/Shuffle Networks. The designed network structures offer efficient results when compared to traditional deep learning models. To enhance the proposed network for brain tumor classification, a modified metaheuristic algorithm named Augmented Falcon Finch Optimization (AFFO) is introduced. AFFO utilizes bio-inspired principles to effectively search for the best hyperparameter configurations, thereby enhancing the reliability and accuracy of the deep learning model. The performance of the proposed method is evaluated on a standard brain tumor MRI dataset and compared with existing techniques, including ResNet, AlexNet, VGG-16, Inception V3, and U-Net to illustrate the effectiveness of combining Residual/Shuffle Networks with AFFO for brain tumor diagnosis.

Explainable early detection of Alzheimer's disease using ROIs and an ensemble of 138 3D vision transformers.

Early detection and accurate diagnosis of brain morphological abnormalities are essential for the effective management and treatment of Alzheimer's disease (AD) and mild cognitive impairment (MCI). Structural magnetic resonance imaging (MRI) is a powerful support tool to aid in disease diagnosis and prediction. In this research study, we present an innovative approach to predict Alzheimer's disease (AD) and mild cognitive impairment (MCI) using MRI data, which integrates regional interest (ROI)-based methodology and deep learning within a comprehensible framework. The proposed method involves dividing the brain into 138 predetermined sections based on anatomical information. Next, we apply three-dimensional vision transformers (3D-ViTs) to each ROI individually, harnessing the power of deep learning. To improve prediction accuracy, we employ a deep belief network (DBN) as an ensemble learning model. Evaluating our approach on the baseline structural MRI dataset obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort, and comparing it against five other competing models, we demonstrate its performance across four binary classification tasks and a three-class classification test (AD vs MCI vs CN (Cognitively Normal)). The proposed system outperforms existing models and provides interpretable insights into the brain regions that significantly contribute to solving each classification problem. Our findings align with the existing body of literature and hold promise for guiding future research directions in this domain.

Enhanced Magnetic Resonance Imaging-Based Brain Tumor Classification with a Hybrid Swin Transformer and ResNet50V2 Model

Brain tumors can be serious; consequently, rapid and accurate detection is crucial. Nevertheless, a variety of obstacles, such as poor imaging resolution, doubts over the accuracy of data, a lack of diverse tumor classes and stages, and the possibility of misunderstanding, present challenges to achieve an accurate and final diagnosis. Effective brain cancer detection is crucial for patients’ safety and health. Deep learning systems provide the capability to assist radiologists in quickly and accurately detecting diagnoses. This study presents an innovative deep learning approach that utilizes the Swin Transformer. The suggested method entails integrating the Swin Transformer with the pretrained deep learning model Resnet50V2, called (SwT+Resnet50V2). The objective of this modification is to decrease memory utilization, enhance classification accuracy, and reduce training complexity. The self-attention mechanism of the Swin Transformer identifies distant relationships and captures the overall context. Resnet 50V2 improves both accuracy and training speed by extracting adaptive features from the Swin Transformer’s dependencies. We evaluate the proposed framework using two publicly accessible brain magnetic resonance imaging (MRI) datasets, each including two and four distinct classes, respectively. Employing data augmentation and transfer learning techniques enhances model performance, leading to more dependable and cost-effective training. The suggested model achieves an impressive accuracy of 99.9% on the binary-labeled dataset and 96.8% on the four-labeled dataset, outperforming the VGG16, MobileNetV2, Resnet50V2, EfficientNetV2B3, ConvNeXtTiny, and convolutional neural network (CNN) algorithms used for comparison. This demonstrates that the Swin transducer, when combined with Resnet50V2, is capable of accurately diagnosing brain tumors. This method leverages the combination of SwT+Resnet50V2 to create an innovative diagnostic tool. Radiologists have the potential to accelerate and improve the detection of brain tumors, leading to improved patient outcomes and reduced risks.

Evaluation of registration-based vs. manual segmentation of rhesus macaque brain MRIs.

With increasing numbers of magnetic resonance imaging (MRI) datasets becoming publicly available, researchers and clinicians alike have turned to automated methods of segmentation to enable population-level analyses of these data. Although prior research has evaluated the extent to which automated methods recapitulate "gold standard" manual segmentation methods in the human brain, such an evaluation has not yet been carried out for segmentation of MRIs of the macaque brain. Macaques offer the important opportunity to bridge gaps between microanatomical studies using invasive methods like tract tracing, neural recordings, and high-resolution histology and non-invasive macroanatomical studies using methods like MRI. As such, it is important to evaluate whether automated tools derive data of sufficient quality from macaque MRIs to bridge these gaps. We tested the relationship between automated registration-based segmentation using an open source and actively maintained NHP imaging analysis pipeline (AFNI) and gold standard manual segmentation of 4 structures (2 cortical: anterior cingulate cortex and insula; 2 subcortical: amygdala and caudate) across 37 rhesus macaques (Macaca mulatta). We identified some variability in the strength of correlation between automated and manual segmentations across neural regions and differences in relationships with demographic variables like age and sex between the two techniques.

A telescopic independent component analysis on functional magnetic resonance imaging dataset

Abstract Brain function can be modeled as dynamic interactions between functional sources at different spatial scales, and each spatial scale can contain its functional sources with unique information, thus using a single scale may provide an incomplete view of brain function. This paper introduces a novel approach, termed “telescopic independent component analysis (TICA),” designed to construct spatial functional hierarchies and estimate functional sources across multiple spatial scales using fMRI data. The method employs a recursive independent component analysis (ICA) strategy, leveraging information from a larger network to guide the extraction of information about smaller networks. We apply our model to the default mode network (DMN), visual network (VN), and right frontoparietal network (RFPN). We investigate further on the DMN by evaluating the difference between healthy people and individuals with schizophrenia. We show that the TICA approach can detect the spatial hierarchy of the DMN, VN, and RFPN. In addition, the TICA revealed DMN-associated group differences between cohorts that may not be captured if we focus on a single-scale ICA. In sum, our proposed approach represents a promising new tool for studying functional sources.

Comparison of Machine Learning Models for Brain Disease Classification Based on CT and MRI Scans

The complexity and severity of brain diseases have led to increased focus on their diagnosis and treatment. Due to the inherent drawbacks of manual medical diagnosis, such as error-prone and costly, and the recent widespread use of Artificial Intelligence (AI) in medicine, it is a worthy topic to explore machine learning in diagnosing brain diseases utilizing Computed Tomography (CT) and Magnetic Resonance Imaging (MRI). In this paper, three brain diseases using CT or MRI datasets from Kaggle were merged and processed, and then three models of decision tree (DT), random forest (RF) and K-Nearest Neighbors (KNN) were used to make classification prediction. Furthermore, their performance was compared through classification reports and confusion matrix to analysis results. The results showed that DT performed worse than the other two models in this task, with an accuracy of 0.91, whereas RF and KNN performed similar overall, each achieving an overall accuracy of 0.96. Notably, RF exhibited less confusion, especially between some similar categories, which indicates the ability of handle complex data more effectively. Additionally, the strengths and weaknesses of the three models are discussed in the paper. These experimental results show that machine learning model, particularly RF and KNN, is a useful tool for diagnosing brain diseases with high accuracy, which could substantially assist clinical practices by providing reliable and efficient diagnostic support.

A multicenter bladder cancer MRI dataset and baseline evaluation of federated learning in clinical application

Bladder cancer (BCa), as the most common malignant tumor of the urinary system, has received significant attention in research on the clinical application of artificial intelligence algorithms. Nevertheless, it has been observed that certain investigations use data from various medical facilities to train models for BCa, which may pose a privacy risk. Given this concern, protecting patient privacy during machine learning algorithm training is a crucial aspect that requires substantial attention. One emerging machine learning paradigm that addresses this concern is federated learning (FL). FL enables multiple entities to collaboratively build machine learning models while preserving data privacy and security. In this study, we present a multicenter BCa magnetic resonance imaging (MRI) dataset. The dataset comprises 275 three-dimensional bladder T2-weighted MRI scans collected from four medical centers, and each scan includes diagnostic pathological labels for muscle invasion and pixel-level annotations of tumor contours. Four FL methods are used to assess the baseline of the dataset for both the task of diagnosing muscle-invasive bladder cancer and automatic bladder tumor lesion segmentation.

MRI Super-Resolution with Partial Diffusion Models.

Diffusion models have achieved impressive performance on various image generation tasks, including image super-resolution. Despite their impressive performance, diffusion models suffer from high computational costs due to the large number of denoising steps. In this paper, we proposed a novel accelerated diffusion model, termed Partial Diffusion Models (PDMs), for magnetic resonance imaging (MRI) super-resolution. We observed that the latents of diffusing a pair of low- and high-resolution images gradually converge and become indistinguishable after a certain noise level. This inspires us to use certain low-resolution latent to approximate corresponding high-resolution latent. With the approximation, we can skip part of the diffusion and denoising steps, reducing the computation in training and inference. To mitigate the approximation error, we further introduced 'latent alignment' that gradually interpolates and approaches the high-resolution latents from the low-resolution latents. Partial diffusion models, in conjunction with latent alignment, essentially establish a new trajectory where the latents, unlike those in original diffusion models, gradually transition from low-resolution to high-resolution images. Experiments on three MRI datasets demonstrate that partial diffusion models achieve competetive super-resolution quality with significantly fewer denoising steps than original diffusion models. In addition, they can be incorporated with recent accelerated diffusion models to further enhance the efficiency.

Patterns of Brain Maturation in Autism and Their Molecular Associations

In the neurotypical brain, regions develop in coordinated patterns, providing a fundamental scaffold for brain function and behavior. Whether altered patterns contribute to clinical profiles in neurodevelopmental conditions, including autism, remains unclear. To examine if, in autism, brain regions develop differently in relation to each other and how these differences are associated with molecular/genomic mechanisms and symptomatology. This study was an analysis of one the largest deep-phenotyped, case-control, longitudinal (2 assessments separated by approximately 12-24 months) structural magnetic resonance imaging and cognitive-behavioral autism datasets (EU-AIMS Longitudinal European Autism Project [LEAP]; study dates, February 2014-November 2017) and an out-of-sample validation in the Brain Development Imaging Study (BrainMapASD) independent cohort. Analyses were performed during the 2022 to 2023 period. This multicenter study included autistic and neurotypical children, adolescents, and adults. Autistic participants were included if they had an existing autism diagnosis (DSM-IV/International Statistical Classification of Diseases and Related Health Problems, Tenth Revision or DSM-5 criteria). Autistic participants with co-occurring psychiatric conditions (except psychosis/bipolar disorder) and those taking regular medications were included. Neuroanatomy of neurotypical and autistic participants. Intraindividual changes in surface area and cortical thickness over time, analyzed via surface-based morphometrics. A total of 386 individuals in the LEAP cohort (6-31 years at first visit; 214 autistic individuals, mean [SD] age, 17.3 [5.4] years; 154 male [72.0%] and 172 neurotypical individuals, mean [SD] age, 16.35 [5.7] years; 108 male [62.8%]) and 146 individuals in the BrainMapASD cohort (11-18 years at first visit; 49 autistic individuals, mean [SD] age, 14.31 [2.4] years; 42 male [85.7%] and 97 neurotypical individuals, mean [SD] age, 14.10 [2.5] years; 58 male [59.8%]). Maturational between-group differences in cortical thickness and surface area were established that were mostly driven by sensorimotor regions (eg, across features, absolute loadings for early visual cortex ranged from 0.07 to 0.11, whereas absolute loadings for dorsolateral prefrontal cortex ranged from 0.005 to 0.06). Neurodevelopmental differences were transcriptomically enriched for genes expressed in several cell types and during various neurodevelopmental stages, and autism candidate genes (eg, downregulated genes in autism, including those regulating synaptic transmission; enrichment odds ratio =3.7; P =2.6 × -10). A more neurotypical, less autismlike maturational profile was associated with fewer social difficulties and more typical sensory processing (false discovery rate P <.05; Pearson r ≥0.17). Results were replicated in the independently collected BrainMapASD cohort. Results of this case-control study suggest that the coordinated development of brain regions was altered in autism, involved a complex interplay of temporally sensitive molecular mechanisms, and may be associated with both lower-order (eg, sensory) and higher-order (eg, social) clinical features of autism. Thus, examining maturational patterns may provide an analytic framework to study the neurobiological origins of clinical profiles in neurodevelopmental/mental health conditions.

An open-access lumbosacral spine MRI dataset with enhanced spinal nerve root structure resolution

Spinal cord injury (SCI) profoundly affects an individual’s ability to move. Fortunately, recent advancements in neuromodulation, particularly the spatio-temporal epidural electrical stimulation (EES) targeting the spinal nerve roots, promoted rapid rehabilitation of SCI patients. Such neuromodulation techniques require precise anatomical modelling of spinal cord. However, the lack of spine imaging datasets, especially high-quality magnetic resonance imaging (MRI) datasets highlighting nerve roots, hinders the translation of EES into medical practice. To address this problem, we introduce an open-access lumbosacral spine MRI dataset acquired in 14 healthy adults, using constructive interference in steady state (CISS) sequence, double echo steady state (DESS) sequence, and T2-weight turbo spin echo (T2-TSE) sequence, with enhanced nerve root resolution. The dataset also includes the corresponding anatomical annotations of nerve roots and the final reconstructed 3D spinal cord models. The quality of our dataset is assessed using image quality metrics implemented in MRI quality control tool (MRIQC). Our dataset provides a valuable platform to promote a wide range of spinal cord neuromodulation research and collaboration among neurorehabilitation engineers.

Graph theory network analysis reveals widespread white matter damage in brains of patients with classic ALS.

Objective: Amyotrophic lateral sclerosis (ALS) exhibits several different presentations and clinical phenotypes. Of these, classic ALS (ALS-Cl), which is the most common phenotype, presents with relatively equal amounts of upper motor neuron and lower motor neuron signs. Magnetic resonance imaging (MRI) provides a noninvasive way to assess central nervous system damage in these patients. To our knowledge no study is available where exploratory whole brain grey matter (GM) and white matter (WM) network analysis is performed considering only the ALS-Cl subgroup of ALS patients. Methods: GM voxel-based morphometry analysis and WM network analysis using graph theory was performed in the MRI dataset of 14 neurologic controls and 25 ALS-Cl patients. Results and Conclusions: No significant GM differences were observed between ALS-Cl and neurologic controls. WM network revealed significant (p < 0.05) reduction and increase in degree measure in several extramotor brain regions of ALS-Cl patients. Both global and local graph metrics revealed significant abnormal values in ALS-Cl patients when compared to neurologic controls. Significant WM changes in ALS-Cl patients with no significant GM changes suggest that neurodegeneration may onset as an "axonopathy" in this ALS subtype.

Enhanced MRI brain tumor detection and classification via topological data analysis and low-rank tensor decomposition

The advent of artificial intelligence in medical imaging has paved the way for significant advancements in the diagnosis of brain tumors. This study presents a novel ensemble approach that uses magnetic resonance imaging (MRI) to identify and categorize common brain cancers, such as pituitary, meningioma, and glioma. The proposed workflow is composed of a two-fold approach: firstly, it employs non-trivial image enhancement techniques in data preprocessing, low-rank Tucker decomposition for dimensionality reduction, and machine learning (ML) classifiers to detect and predict the type of brain tumor. Secondly, persistent homology (PH), a topological data analysis (TDA) technique, is exploited to extract potential critical areas in MRI scans. When paired with the ML classifier output, this additional information can help domain experts to identify areas of interest that might contain tumor signatures, improving the interpretability of ML predictions. When compared to automated diagnoses, this transparency adds another level of confidence and is essential for clinical acceptance. The performance of the system was quantitatively evaluated on a well-known MRI dataset, with an overall classification accuracy of 97.28% using an extremely randomized trees model. The promising results show that the integration of TDA, ML, and low-rank approximation methods is a successful approach for brain tumor identification and categorization, providing a solid foundation for further study and clinical application.

Intersubject Dynamic Conditional Correlation: A Novel Method to Track the Framewise Network Implication during Naturalistic Stimuli.

Background: Naturalistic stimuli have become increasingly popular in modern cognitive neuroscience. These stimuli have high ecological validity due to their rich and multilayered features. However, their complexity also presents methodological challenges for uncovering neural network reconfiguration. Dynamic functional connectivity using the sliding-window technique is commonly used but has several limitations. In this study, we introduce a new method called intersubject dynamic conditional correlation (ISDCC). Method: ISDCC uses intersubject analysis to remove intrinsic and non-neuronal signals, retaining only intersubject-consistent stimuli-induced signals. It then applies dynamic conditional correlation (DCC) based on the generalized autoregressive conditional heteroskedasticity to calculate the framewise functional connectivity. To validate ISDCC, we analyzed simulation data with known network reconfiguration patterns and two publicly available narrative functional Magnetic Resonance Imaging (fMRI) datasets. Results: (1) ISDCC accurately unveiled the underlying network reconfiguration patterns in simulation data, demonstrating greater sensitivity than DCC; (2) ISDCC identified synchronized network reconfiguration patterns across listeners; (3) ISDCC effectively differentiated between stimulus types with varying temporal coherence; and (4) network reconfigurations unveiled by ISDCC were significantly correlated with listener engagement during narrative comprehension. Conclusion: ISDCC is a precise and dynamic method for tracking network implications in response to naturalistic stimuli.

Abnormal brain entropy dynamics in ADHD

BackgroundBrain entropy (BEN) is a novel measure for irregularity and complexity of brain activities, which has been used to characterize abnormal brain activities in many brain disorders including attention-deficit/hyperactivity disorder (ADHD). While most research assumes BEN is stationary during scan sessions, the brain in resting state is also a highly dynamic system. The BEN dynamics in ADHD has not been explored. MethodsWe used a sliding window approach to derive the dynamical brain entropy (dBEN) from resting-state functional magnetic resonance imaging (rfMRI) dataset that includes 98 ADHD patients and 111 healthy controls (HCs). We identified 3 reoccurring BEN states. We tested whether the BEN dynamics differ between ADHD and HC, and whether they are associated with ADHD symptom severity. ResultsOne BEN states, characterized by low overall BEN and low within-state BEN located in SMN (sensorimotor network) and VN (visual network), its FW (fractional window) and MDT (mean dwell time) were increased in ADHD and positively correlated with ADHD severity; another state characterized by high overall BEN and low within-state BEN located in DMN (default mode network) and ECN (executive control network), its FW and MDT were decreased in ADHD and negatively correlated with ADHD severity. LimitationsThe window length of dBEN analysis can be further optimized to suit more datasets. The co-variation between dBEN and other dynamical brain metrics was not explored. ConclusionOur findings revealed abnormal BEN dynamics in ADHD, providing new insights into clinical diagnosis and neuropathology of ADHD.

A Telescopic Independent Component Analysis on Functional Magnetic Resonance Imaging Data Set.

Our study introduces "telescopic independent component analysis (TICA)", a novel approach using a recursive ICA strategy for exploring brain function across multiple spatial scales using fMRI data. TICA constructs spatial hierarchies and identifies functional sources such as default mode network (DMN), visual network (VN), and right frontoparietal network (RFPN). By applying TICA, we uncovered hierarchical structures within these networks and revealed differences in the DMN between healthy individuals and those with schizophrenia. This approach offers a promising new tool for studying function dynamics comprehensively.

Exploiting K-Space in Magnetic Resonance Imaging Diagnosis: Dual-Path Attention Fusion for K-Space Global and Image Local Features.

Magnetic resonance imaging (MRI) diagnosis, enhanced by deep learning methods, plays a crucial role in medical image processing, facilitating precise clinical diagnosis and optimal treatment planning. Current methodologies predominantly focus on feature extraction from the image domain, which often results in the loss of global features during down-sampling processes. However, the unique global representational capacity of MRI K-space is often overlooked. In this paper, we present a novel MRI K-space-based global feature extraction and dual-path attention fusion network. Our proposed method extracts global features from MRI K-space data and fuses them with local features from the image domain using a dual-path attention mechanism, thereby achieving accurate MRI segmentation for diagnosis. Specifically, our method consists of four main components: an image-domain feature extraction module, a K-space domain feature extraction module, a dual-path attention feature fusion module, and a decoder. We conducted ablation studies and comprehensive comparisons on the Brain Tumor Segmentation (BraTS) MRI dataset to validate the effectiveness of each module. The results demonstrate that our method exhibits superior performance in segmentation diagnostics, outperforming state-of-the-art methods with improvements of up to 63.82% in the HD95 distance evaluation metric. Furthermore, we performed generalization testing and complexity analysis on the Automated Cardiac Diagnosis Challenge (ACDC) MRI cardiac segmentation dataset. The findings indicate robust performance across different datasets, highlighting strong generalizability and favorable algorithmic complexity. Collectively, these results suggest that our proposed method holds significant potential for practical clinical applications.

Decoding pain: uncovering the factors that affect the performance of neuroimaging-based pain models.

Neuroimaging-based pain biomarkers, when combined with machine learning techniques, have demonstrated potential in decoding pain intensity and diagnosing clinical pain conditions. However, a systematic evaluation of how different modeling options affect model performance remains unexplored. This study presents the results from a comprehensive literature survey and benchmark analysis. We conducted a survey of 57 previously published articles that included neuroimaging-based predictive modeling of pain, comparing classification and prediction performance based on the following modeling variables-the levels of data, spatial scales, idiographic vs population models, and sample sizes. The findings revealed a preference for population-level modeling with brain-wide features, aligning with the goal of clinical translation of neuroimaging biomarkers. However, a systematic evaluation of the influence of different modeling options was hindered by a limited number of independent test results. This prompted us to conduct benchmark analyses using a locally collected functional magnetic resonance imaging dataset (N = 124) involving an experimental thermal pain task. The results demonstrated that data levels, spatial scales, and sample sizes significantly impact model performance. Specifically, incorporating more pain-related brain regions, increasing sample sizes, and averaging less data during training and more data during testing improved performance. These findings offer useful guidance for developing neuroimaging-based biomarkers, underscoring the importance of strategic selection of modeling approaches to build better-performing neuroimaging pain biomarkers. However, the generalizability of these findings to clinical pain requires further investigation.

Radiomic detection of abnormal brain regions in tuberous sclerosis complex.

Radiomics refers to the extraction of quantitative information from medical images and is most commonly utilized in oncology to provide ancillary information for solid tumor diagnosis, prognosis, and treatment response. The traditional radiomic pipeline involves segmentation of volumes of interest with comparison to normal brain. In other neurologic disorders, such as epilepsy, lesion delineation may be difficult or impossible due to poor anatomic definition, small size, and multifocal or diffuse distribution. Tuberous sclerosis complex (TSC) is a rare genetic disease in which brain magnetic resonance imaging (MRI) demonstrates multifocal abnormalities with variable imaging and epileptogenic features. The purpose of this study was to develop a radiomic workflow for identification of abnormal brain regions in TSC, using a whole-brain atlas-based approach with generation of heatmaps based on signal deviation from normal controls. This was a retrospective pilot study utilizing high-resolution whole-brain 3D FLAIR MRI datasets from retrospective enrollment of tuberous sclerosis complex (TSC) patients and normal controls. Subjects underwent MRI including high-resolution 3D FLAIR sequences. Preprocessing included skull stripping, coregistration, and intensity normalization. Using the Brainnetome and Harvard-Oxford atlases, brain regions were parcellated into 318 discrete regions. Expert neuroradiologists spatially labeled all tubers in TSC patients using ITK-SNAP. The pyradiomics toolbox was used to extract 88 radiomic features based on IBSI guidelines, comparing tuber-affected and non-tuber-affected parenchyma in TSC patients, as well as normal brain tissue in control patients. For model training and validation, regions with tubers from 20 TSC patients and 30 normal control subjects were randomly divided into two training sets (80%) and two validation sets (20%). Additional model testing was performed on a separate group of 20 healthy controls. LASSO (least absolute shrinkage and selection operator) was used to perform variable selection and regularization to identify regions containing tubers. Relevant radiomic features selected by LASSO were combined to produce a radiomic score ω, defined as the sum of squared differences from average control group values. Region-specific ω scores were converted to heat maps and spatially coregistered with brain MRI to reflect overall radiomic deviation from normal. The proposed radiomic workflow allows for quantification of deviation from normal in 318 regions of the brain with the use of a summative radiomic score ω. This score can be used to generate spatially registered heatmaps to identify brain regions with radiomic abnormalities. The pilot study of TSC showed radiomic scores ω that were statistically different in regions containing tubers from regions without tubers/normal brain (p<0.0001). Our model exhibits an AUC of 0.81 (95% confidence interval: 0.78-0.84) on the testing set, and the best threshold obtained on the training set, when applied to the testing set, allows us to identify regions with tubers with a specificity of 0.91 and a sensitivity of 0.60. We describe a whole-brain atlas-based radiomic approach to identify abnormal brain regions in TSC patients. This approach may be helpful for identifying specific regions of interest based on relatively greater signal deviation, particularly in clinical scenarios with numerous or poorly defined anatomic lesions.

Fourier Convolution Block with global receptive field for MRI reconstruction

Reconstructing images from under-sampled Magnetic Resonance Imaging (MRI) signals significantly reduces scan time and improves clinical practice. However, Convolutional Neural Network (CNN)-based methods, while demonstrating great performance in MRI reconstruction, may face limitations due to their restricted receptive field (RF), hindering the capture of global features. This is particularly crucial for reconstruction, as aliasing artifacts are distributed globally. Recent advancements in Vision Transformers have further emphasized the significance of a large RF. In this study, we proposed a novel global Fourier Convolution Block (FCB) with whole image RF and low computational complexity by transforming the regular spatial domain convolutions into frequency domain. Visualizations of the effective RF and trained kernels demonstrated that FCB improves the RF of reconstruction models in practice. The proposed FCB was evaluated on four popular CNN architectures using brain and knee MRI datasets. Models with FCB achieved superior PSNR and SSIM than baseline models and exhibited more details and texture recovery. The code is publicly available at https://github.com/Haozhoong/FCB.