A review of deep learning-based Unsupervised Anomaly Detection in brain MRI.

Alzheimer's disease (AD) is a progressive neurodegenerative disorder and the leading cause of dementia globally. Early prediction, prior to the onset of symptoms, is critical for enabling timely interventions. We present a machine learning framework that predicts future AD conversion in cognitively normal (CN) individuals using only structural magnetic resonance imaging (MRI) data. The approach leverages transfer learning with a pre-trained VGG16 model for feature extraction and processes five representative 2D slices per brain MRI scan to generate compact imaging descriptors. These features are classified using an ensemble composed of support vector machines (SVM), random forests (RF), and artificial neural networks (ANN), with outputs combined through soft voting. The model was evaluated using person-wise stratified cross-validation on 1,093 subjects from the OASIS-3 dataset, ensuring no data leakage and providing realistic performance estimates. Across 200 randomized runs, the ensemble achieved a median AUC-ROC of 0.951, accuracy of 0.872, recall of 0.923, and F1 score of 0.811. These results demonstrate that ensemble machine learning can detect preclinical AD signatures from structural MRI, offering a practical, relatively accessible, and cost-effective tool for early risk identification and intervention.

Many organisations collect sensitive data that cannot be freely shared. Hospitals store brain magnetic resonance imaging (MRI) scans on internal servers; banks keep transaction records behind strict firewalls; agricultural services retain crop images in isolated repositories. Federated learning (FL) allows models to be trained without centralising raw data, yet most existing systems address a single domain and offer limited insight into model behaviour and provenance over time. BlockFedX is a cross-domain federated learning system designed to address three simultaneous tasks: credit card fraud detection on tabular data, brain tumour detection on MRI images, and plant disease recognition on leaf images. These three domains were deliberately selected because they represent the principal data modalities in real-world privacy-sensitive deployments-structured tabular records, greyscale medical images, and colour natural images-and because public benchmark datasets exist for all three, enabling reproducible evaluation. The system uses a shared backbone that is updated only where model layers have compatible tensor shapes, while domain-specific output layers remain local at each client. Explanations are computed at the clients using SHAP feature-attribution for tabular data and Grad-CAM visual heatmaps for images; the server receives only compact statistical summaries. The server also applies a distance-based anomaly test on client updates and records model hashes, explanation summaries, and anomaly flags in a hash-chained ledger. Experiments on three public datasets under non-identical client data distributions show that BlockFedX achieves an average fraud-detection F1-score of 0.92, 74.32% mean validation accuracy on BrainMRI, and 77% test accuracy on PlantVillage, while keeping all raw data local. These results are below strong centralised baselines, as expected under compact models and non-IID splits, but the system simultaneously provides three properties rarely combined in prior work: cross-domain federated training via a shape-safe backbone, client-side explanations integrated into the learning loop, and a lightweight tamper-evident record of model evolution across rounds.

To determine the levels of serum biomarkers of astrocytic (glial fibrillary acidic protein, GFAP) and neuronal (ubiquitin C-terminal hydrolase-L1, UCH-L1) injury in children with global developmental delay (GDD) of non-structural etiology and to investigate their associations with developmental profiles and specific brain magnetic resonance imaging (MRI) phenotypes. In this retrospective cross-sectional study, we included 98 children (median age 38.1 months; 72 males) diagnosed with GDD of non-structural etiology. Serum GFAP and UCH-L1 concentrations were measured. Developmental outcomes were assessed across five domains using the Gesell Developmental Scales. Brain MRI scans (n = 70) were systematically reviewed and categorized for specific structural phenotypes. Serum GFAP and UCH-L1 levels exceeding the laboratory-defined cutoffs were detectable in 74.5% and 42.9% of children, respectively. After adjusting for age and sex, higher GFAP level was independently associated with a lower developmental quotient in the language domain (β = -0.314, P = 0.011). GFAP levels were significantly higher in children with widened extracerebral spaces (P = 0.011). Serum GFAP and UCH-L1 levels exceeding the cutoffs are frequently detectable in GDD of non-structural etiology. The specific association of GFAP with language impairment and widened extracerebral spaces provides measurable evidence of astrocytic involvement in functional and structural alterations. These biomarkers may aid in the biological stratification of GDD, though the findings remain exploratory and require validation in independent cohorts.

The aim was to determine whether plasma amyloid beta (Aβ) Aβ37, Aβ40, Aβ42, and p-tau217 differ in cerebral amyloid angiopathy (CAA) from controls and associate with specific magnetic resonance imaging (MRI) markers of CAA. Single center hospital-based study (Massachusetts General Hospital) enrolling patients with probable or definite CAA per Boston 2.0 criteria among consecutive individuals presenting with spontaneous intracerebral hemorrhage (ICH) and non-acute ICH related symptoms. Control participants were identified among individuals 55 years or older without history of ICH attending outpatient clinics in the same catchment area and time as the CAA cases. Primary outcome tested whether plasma Aβ37, Aβ40, Aβ42, and p-tau217 differ between patients with CAA and controls. Secondary outcome evaluated the correlation with the severity of selected CAA neuroradiological hallmarks on brain MRI scans obtained within 1 year of sample collection. A total of 186 patients with CAA (mean age, 71.2 ± 8.3 years; males, 107 [58%]) and 401 controls (mean age, 73.9 ± 8.3 years; males, 211 [53%]) were included in the study. CAA cases had lower Aβ40 (β = -0.60, 95% CI = -0.80 to -0.40, P < 0.001), Aβ42 (β = -0.48, 95% CI = -0.70 to -0.27, P < 0.001), and higher p-tau217 (β = 0.73, 95 %CI = 0.54 to 0.91, P < 0.001) concentrations compared to controls. Aβ40 was the only fragment reduced independent of ptau217 levels. Within the CAA group, biomarkers were associated with both hemorrhagic (reduced Aβ42 with cortical superficial siderosis) and non-hemorrhagic (reduced Aβ40 and Aβ42 with enlarged perivascular spaces in centrum semiovale, reduced Aβ37 and increased p-tau217 with white matter hyperintensities) MRI hallmarks of CAA. Plasma Aβ40, Aβ42, and p-tau217 differ between CAA patients and controls and are informative on CAA severity as inferred from brain MRI markers. ANN NEUROL 2026.

BackgroundIndividuals with difficult-to-treat migraine, including resistant migraine (ResM) and refractory migraine (RefM), might experience treatment delays, undergo unnecessary diagnostic tests and receive misdiagnoses, which might influence treatment outcomes. For this reason, we hypothesized that individuals with ResM and RefM might report more diagnostic tests and misdiagnoses in their medical history compared with those with non-resistant/non-refractory migraine (NRNRM).MethodsThis analysis used baseline, cross-sectional data from the REFINE study, a multicenter, prospective observational study conducted in 15 European tertiary headache centers. Adults with episodic or chronic migraine were classified into RefM, ResM, or NRNRM groups. Baseline data were analyzed to assess the frequency of previous diagnostic tests and misdiagnoses.ResultsOverall, 689 participants were included with a median age of 46 years (interquartile range 36-53); 570 participants (82.7%) were female; 355 (51.5%) had NRNRM, 261 (37.9%) ResM, and 73 (10.9%) RefM. Referring to diagnostic tests, 335 participants (48.7%) had one and 237 (34.4%) multiple brain magnetic resonance imaging scans. ResM and RefM participants underwent more diagnostic tests than NRNRM. Overall, 193 participants (28.0%) had at least one prior headache misdiagnosis, most commonly cervical spine disorders and sinusitis; misdiagnoses were more frequent in NRNRM and ResM than in RefM (31.1%, 28.5%, and 15.1%, respectively; p = 0.025). Misdiagnosis rates were not influenced by age, sex, disease duration, or comorbidities.ConclusionsDiagnostic tests use and misdiagnoses are highly prevalent in all the three groups of individuals with ResM, RefM, and NRNRM with some differences across the three groups that may depend on multiple factors. Our findings emphasize a need for better diagnostic accuracy and care pathway across the entire spectrum of migraine, to avoid unnecessary diagnostic tests and misdiagnoses.

Early identification of individuals at risk of dementia is essential for preventive care and timely enrolment into disease-modifying interventions. However, most existing prediction approaches rely on invasive, costly, or research-only biomarkers that are not scalable within public healthcare systems. Routinely acquired National Health Service (NHS) brain magnetic resonance imaging (MRI) scans, when linked with electronic health records, represent a widely available and privacy-preserving resource for population-level dementia risk stratification. A key challenge for clinical translation is ensuring that machine-learning predictions are reliable, interpretable, and safe to apply, particularly when models are used years before clinical diagnosis. We conducted a retrospective case-control study entirely within a secure NHS Trusted Research Environment using routine T1-weighted brain MRI scans linked to electronic health records from Tayside and Fife, Scotland. The study included 518 participants: 259 individuals who subsequently developed dementia and 259 age- and sex-matched controls. Structural brain features were derived from MRI data and analysed using a support-vector-machine classifier with nested cross-validation to minimise overfitting. Prediction confidence was quantified using distance-from-hyperplane (DFH) calibration, enabling stratification of model outputs by certainty. Primary outcomes were classification accuracy and area under the receiver-operating-characteristic curve (AUC). Secondary analyses examined DFH-stratified performance and the relationship between prediction accuracy and time from scan to first recorded dementia diagnosis. The model predicted future dementia up to five years before first recorded NHS diagnosis with an AUC of 0.71, a performance consistent with real-world clinical imaging rather than research-optimised datasets. Model sensitivity increased for scans acquired closer to diagnosis, indicating stronger predictive signal as disease onset approached. Confidence-based stratification identified a high-confidence subgroup comprising approximately 35% of scans, within which prediction accuracy increased to around 80%. Performance was consistent across heterogeneous routine NHS scanners and imaging protocols, demonstrating robustness and generalisability to real-world clinical data rather than research-optimised acquisitions. Routinely collected NHS brain MRI data can be used to predict future dementia several years before clinical diagnosis. Incorporating confidence calibration transforms a conventional machine-learning classifier into a safety-aware and clinically interpretable framework by enabling selective use of high-certainty predictions. This approach supports scalable early detection, population-level risk stratification, and targeted recruitment into preventive or disease-modifying clinical trials, with clear potential for integration into public health systems.

Associations of brain MRI and perinatal factors with 2-year neurodevelopment in very preterm infants.

Quality control (QC) of magnetic resonance imaging (MRI) data before preprocessing is fundamental, because substandard data are known to introduce additional variability in the form of noise to subsequent analyses. This can result in spurious results of a false effect or the obstruction of a true effect. Consequently, there is a need for a reliable and robust method to identify subpar images, given pre-specified exclusion criteria. Here, we describe how to carry out the visual assessment of T1-weighted, T2-weighted, functional and diffusion MRI scans of the human brain with visual reports generated by MRIQC ( https://mriqc.readthedocs.io/en/stable/ ). We provide guidance and instructions for using the MRIQC software on all the images of the input dataset using typical research settings (i.e., a high-performance computing cluster). This includes installing MRIQC, configuring datasets (30-45 min active, plus 10-15 min of compute time per scan) and executing MRIQC (10-15 min compute time per scan). We then describe how to screen the visual reports generated with MRIQC to identify artifacts and potential quality issues and annotate the latter with the 'rating widget', a utility that enables rapid annotation and minimizes bookkeeping errors (1-5 min per participant). Integrating proper QC checks on the unprocessed data is fundamental to producing reliable statistical results and crucial to identifying faults in the scanning settings, preempting the acquisition of large datasets with persistent artifacts that should have been addressed as they emerged.

Background/Objectives: Serum albumin has antioxidant, anti-inflammatory, and antithrombotic properties and reflects nutritional status. Hypoalbuminemia is linked to cognitive decline and frailty. However, the relationship between serum albumin levels and brain structural changes in older adults remains unclear. We aimed to examine the associations between serum albumin levels and total brain, hippocampal, and white matter lesion volumes in cognitively normal, community-dwelling older Japanese adults, accounting for frailty status. Methods: In this cross-sectional study, 7266 participants aged ≥65 years without cognitive decline were included. Serum albumin levels, maximum handgrip strength, and usual gait speed were measured in all participants. Brain magnetic resonance imaging scans were used to evaluate total brain, hippocampal, and white matter lesion volumes. Results: Lower serum albumin levels were significantly associated with smaller total brain and hippocampal volumes after multivariable adjustment (both p for trend < 0.001; partial η2 = 0.005), but not with white matter lesion volumes (p for trend = 0.24; partial η2 = 0.001). In subgroup analyses stratified by frailty status, no significant heterogeneity in the associations between serum albumin levels and each brain volume was observed between groups defined by maximum handgrip strength or usual gait speed. Conclusions: Lower serum albumin levels are associated with smaller total brain and hippocampal volumes in cognitively normal, community-dwelling older Japanese adults, irrespective of frailty status. Serum albumin may serve as a clinically accessible marker of nutritional conditions in relation to these brain structures in older adults.

ADAMTS13 and von Willebrand Factor in relation to vascular cognitive impairment and cerebral small vessel disease: the heart-brain connection study.

Modernly speaking, reviewing large numbers of Magnetic Resonance Imaging (MRI) images and manually discovering a brain tumor by a person is a slow and inaccurate process. It may have effects on the correct medical treatment of the patient. Additionally, it could be a slow and laborious task due to the numerous amounts of image datasets involved. Because brain tumors appear similarly to healthy tissue, tumor region segmentation can be difficult. Therefore, there is a need for a high-quality automatic tumor detection system. CNNs are one type of deep learning technique which are often used for image recognition and image classification tasks currently. CNNs are also commonly used to identify Brain Tumors. In our research, we proposed a CNN model for the purpose of classifying images from MRI scans of brains into two classes (Normal or Tumor). Our proposed model was able to achieve a recall of 97.51%, accuracy of 97.889%, F1-score of 97.84%, precision of 98.18%, specificity of 97.62% and an AUC of 97.57%. Our CNN model will help doctors to find brain tumors in MRI images with great efficiency , therefore, greatly increasing the amount of time saved when treating patients

Structural neuroimaging of the planning network in Inuit adolescents prenatally exposed to lead.

Neuroimaging datasets are increasingly shared in open repositories for research purposes, raising concerns about participant re-identification through facial features visible in brain magnetic resonance imaging (MRI) scans. MRI defacing algorithms address this risk by obscuring identifiable facial structures while preserving brain tissue for analysis. Algorithm performance varies substantially by context. For re-identification prevention, fsl_deface and mri_reface achieve the lowest recognition rates, while afni_refacer and pydeface demonstrate the highest processing success rates. However, all algorithms affect brain volumetric measurements to varying degrees, with some causing failures in automated segmentation pipelines. Performance is notably age-dependent, with specific algorithms underperforming in pediatric or elderly cohorts and in clinical populations with neurological disorders. Optimal algorithm selection depends on research priorities. For preserving brain measurements, mri_reface and SPM-based defacing are preferred; for pediatric studies, FreeSurfer better preserves brain voxels; for electroencephalography (EEG) and magnetoencephalography (MEG) co-registration, AnonyMI provides superior geometrical preservation. This review examines the major defacing algorithms and their validation across diverse datasets, evaluating effectiveness in preventing re-identification, preserving brain measurements, and maintaining compatibility across age groups. A comparative discussion highlights the trade-offs between privacy protection and data utility, emphasizing the need for a study-specific approach when selecting a defacing method.

Lecanemab is an anti-amyloid monoclonal antibody, recently approved in the UK as a treatment for mild cognitive impairment (MCI) and mild dementia due to Alzheimer's disease (AD) in adults who are apolipoprotein E ε4 gene (APOE4) heterozygotes or non-carriers.A group of UK neurologists, old age psychiatrists and geriatricians with expertise in AD convened to agree appropriate use recommendations for lecanemab in UK clinical practice. The primary focus of these recommendations is safety.Eligibility criteria for lecanemab in the UK include (a) a clinical diagnosis of MCI or mild dementia due to AD, (b) the presence of amyloid-β pathology, confirmed using approved methods (ie, an amyloid positron emission tomography scan or cerebrospinal fluid assay) and (c) APOE4 heterozygous or non-carrier status. Eligibility screening should be conducted in secondary care and those identified as being potentially eligible for lecanemab should be referred to a specialist centre for confirmation of the likely pathological diagnosis, APOE4 counselling and testing and a multidisciplinary consensus decision regarding treatment eligibility. Lecanemab is administered as an intravenous infusion every 2 weeks, and those eligible for treatment should have brain magnetic resonance imaging (MRI) scans prior to the 1st, 5th, 7th and 14th infusions. Specific guidance is provided for safety monitoring and management of potential adverse reactions, including amyloid-related imaging abnormalities and infusion-related reactions.The introduction of lecanemab into UK clinical practice provides an important opportunity to improve services for all people living with dementia, not just those eligible for lecanemab treatment.

BackgroundTraumatic brain injury (TBI) and other acute intracranial pathologies disrupt the glymphatic system, a recently described waste-clearance network that facilitates the removal of metabolic byproducts from the brain. Dysfunction of this system after injury may contribute to impaired clearance of toxic metabolites, cerebral edema, and elevated intracranial pressure. This study aimed to evaluate glymphatic dynamics using intrathecal (IT) administration of gadoterate meglumine via an external ventricular drain (EVD) to better understand the impact of intracranial injury on glymphatic flow.MethodsThis single-center retrospective study, conducted from July 2025 to November 2025, enrolled six patients who were admitted for an intracranial pathology that required placement of an EVD to study the glymphatic system utilizing IT administration of contrast agent gadoterate meglumine. Serial magnetic resonance imaging (MRI) brain scans were performed pre-contrast and at least one additional time, either four, 12, or 36 hours after contrast administration. Intensity measurements were then taken on the images and compared at the following brain parenchymal locations: bilateral frontal white and grey matter, bilateral temporal white and grey matter, bilateral parietal white and grey matter, central pons located ventral to the cerebral aqueduct, central medulla located ventral to the fourth ventricle, and bilateral cerebellar white matter. ResultsNo immediate procedural complications were observed following IT contrast administration. Patients demonstrated variable degrees of MRI signal change that correlated with presumed glymphatic function, with increased enhancement observed in regions of preserved flow and reduced enhancement in areas affected by intracranial pathology. In some cases, transient decreases in signal intensity were observed following contrast administration, which may reflect localized high-concentration contrast effects with susceptibility-related signal loss (“first-pass” phenomenon), altered glymphatic transport, or impaired clearance. Reduced glymphatic tracer propagation and diminished enhancement were observed in patients with neurological decline, whereas enhanced glymphatic transport was noted following EVD placement. One patient experienced neurological deterioration following IT contrast administration; however, given the presence of multiple confounding clinical factors, a causal relationship could not be established.ConclusionsThe glymphatic system plays a critical role in intracranial homeostasis and appears to be disrupted following acute brain injury. IT contrast-enhanced MRI enables visualization of glymphatic dynamics and demonstrates regional variation in tracer movement corresponding to underlying pathology. While low-dose IT gadolinium administration has demonstrated tolerability in prior human studies, the present findings highlight that definitive conclusions regarding safety and causality cannot be drawn from this small cohort. These results support the feasibility of IT contrast-enhanced MRI for evaluating glymphatic function, but larger prospective studies are needed to better define safety, pathophysiological mechanisms, and clinical implications.

Brain tumor classification using hybrid spinal-EfficientNet using MRI images.

We examined whether long-term exposure to visceral-adipose-tissue (VAT) influences brain atrophy and cognitive performance years after lifestyle intervention. In the Follow-Interventions-Trials (FIT) project, 533 adults (age=61.4 y, 86% men) from four prior 18-24-month lifestyle randomized-clinical-trials underwent abdominal/brain magnetic-resonance-imaging (MRI)s and Montreal-Cognitive-Assessment (MoCA) testing 5-16 y after interventions. Lower VAT exposure, calculated by area-under-the-curve, from baseline, post-intervention, and follow-up, independently resulted in higher MoCA scores. VAT loss during intervention predicted higher brain volumes at follow-up, independent of weight loss. Among participants with three brain and VAT MRI scans, lower long-term VAT was associated with a slower rate of brain atrophy. These patterns were not observed for deep/superficial subcutaneous-adipose-tissues. Improved glycemic control parameters, rather than lipid or inflammatory markers, were mostly related to the favorable longitudinal brain outcomes. This long-term, large-scale intervention and follow-up MRI study suggests that sustained visceral fat loss, rather than weight loss, is linked to better cognition and attenuation of brain atrophy years later, mainly via improved glycemic control. Trial registration: DIRECT (Clinical-trials-identifier: NCT00160108); CASCADE (Clinical-trials-identifier: NCT00784433); CENTRAL (Clinical-trials-identifier: NCT01530724); DIRECT-PLUS (Clinical-trials-identifier: NCT03020186).

Magnetic Resonance Imaging (MRI) exposes patients to radiofrequency energy measured by the specific absorption rate (SAR), a key safety metric. This study aimed to compare SAR values in brain MRI scans performed at 1.5 Tesla (T) and 3T to inform safer imaging practices. A retrospective analysis of 200 adult brain MRI scans (100 at 1.5T and 100 at 3T) from King Khalid University Hospital was conducted. Data included SAR, demographics, scan parameters, and contrast use. Statistical tests assessed differences (P<.05). Brain SAR was significantly higher in 1.5T scans (mean=3.01W/kg) than in 3T (mean=1.37W/kg). Higher SAR values were noted in females and younger patients. Factors like image type, sequence, weight, flip angle, and contrast use significantly impacted SAR. SAR is more influenced by imaging parameters and patient characteristics than MRI magnetic field strength. Personalized MRI protocols and SAR monitoring are essential for patient safety.

Brain tumors remain among the most dangerous life-threatening neurological disorders, and their accurate identification, segmentation, and classification are critical for improving treatment outcomes. Traditional cancer detection methods are often labor-intensive and susceptible to human error, whereas an automated solution is proposed to foster early diagnosis and reduce inaccuracies. This study presents FCDS-DeepVision, a lightweight convolutional neural network built from scratch to classify brain disorders using magnetic resonance imaging scans. The preprocessing stage applies a combination of intensity normalization, noise suppression, and contrast enhancement to improve the visibility of diagnostically relevant structures within the magnetic resonance imaging scans. To further reduce background interference, a segmentation step is introduced to localize anatomically meaningful brain regions prior to classification. This localized representation supports a more reliable separation of normal tissue from pathological patterns associated with glioma, meningioma, and pituitary tumors. The proposed framework is evaluated using a publicly available dataset of 7,020 brain magnetic resonance imaging scans collected from Kaggle. Experimental results indicate that the model maintains fast inference while producing stable and reproducible predictions across multiple evaluation metrics, including precision, recall, F1-score, Area Under the Curve, and overall accuracy. Rather than relying on a single performance indicator, the analysis emphasizes consistency across metrics under controlled experimental settings. The results suggest that such an automated pipeline can assist diagnostic workflows by reducing manual effort and limiting variability introduced through subjective interpretation. While the current findings are encouraging, further investigation is required to assess robustness under broader data distributions, larger cohorts, and real clinical acquisition conditions. Extending the dataset and validating the framework in practical diagnostic environments remain important directions for future work.