Brain Informatics

Emerging as a significant global health challenge, Alzheimer's Disease (AD) is a progressive neurodegenerative disorder that causes memory loss and cognitive decline. Despite the ever-increasing waiting time for a specialist diagnosis, the need for a cost-effective and fast diagnostic technique is evident. This study explores the development of an explainable deep learning model to diagnose AD using only routine and low-cost clinical data, including demographic information, patient history, and results of neuropsychological tests (limited to those that can be automatically acquired). The analysis was carried out using a dataset provided by the National Alzheimer's Coordinating Center, comprising 167,364 observations and 1,024 features. The findings demonstrate diagnostic performance comparable, and slightly superior, to that of clinicians when evaluated under similar informative constraints. This study introduces two classification models to discriminate whether the presumptive etiological cause of cognitive impairment is Alzheimer's disease. The deep neural network achieved an accuracy of 90\% with an area under the receiver operating characteristic curve (ROC-AUC) of 0.96, whereas the Light Gradient Boosting Machine reached the same accuracy with a ROC-AUC of 0.97.

Purpose: To investigate sex differences in cerebral blood flow through densely parcellated cortical and subcortical regions using explainable artificial intelligence methods and identify neurobiologically interpretable perfusion biomarkers. Methods: High-resolution pseudo-continuous arterial spin labelling (1.875 mm x 1.875 mm x 3 mm) and structural MRI data were curated from 215 healthy young adults (150 females, 95 males; age 18-30 years) from the publicly available I See your Brains (ISYB) dataset. Cerebral blood flow was quantified using atlas-based regional analysis with the Brainnetome Atlas (246 regions) and optimized registration procedures. Sex classification employed diverse machine learning paradigms including linear classifiers, ensemble methods, and kernel-based approaches for regional CBF features, with deep convolutional neural networks (CNN) applied to whole-brain 3D imaging data. Model interpretability was achieved using SHapley Additive exPlanations (SHAP), computed over an ensemble of 500 logistic regression models (100 iterations x 5-fold cross-validation). Regions appearing among the top 20% of discriminative features more than 289 times were considered statistically significant using binomial testing. GradCAM was used to obtain class-specific attribution maps from the CNN model. Results: Perfusion-based features demonstrated superior sex classification performance compared to structural morphometry. Regional CBF analysis using logistic regression achieved 91 +/- 2% balanced accuracy and 0.95 +/- 0.05 ROC-AUC, substantially outperforming morphometric features (85 +/- 8% balanced accuracy, 0.88 +/- 0.06 ROC-AUC). Deep learning classification of 3D CBF maps achieved a performance of 92 +/- 5% balanced accuracy, 0.92 +/- 0.05 ROC-AUC. SHAP analysis identified 30 statistically significant aggregation-agnostic CBF-based biomarker regions using regional CBF, predominantly involving frontoparietal control networks (27%) and default mode networks (17%). Grad-CAM revealed that the 3D CNN model primarily focused on regions within the frontal lobe. Morphometry-based analysis identified 28 discriminative regions with markedly different anatomical distribution (r = 0.21) emphasizing visual (32%) and default mode (14%) networks. Conclusion: Cerebral blood flow patterns provide highly sensitive and biologically interpretable markers of sex differences in young adult brain. The identification of robust perfusion biomarkers through explainable AI demonstrates the clinical potential of ASL imaging for precision medicine applications in neuroscience. We establish a methodological framework for investigating sex-specific brain physiology using non-invasive neuroimaging.

Non-contrast head CT is the first-line imaging modality for acute neurological emergencies, with demand rising worldwide. However, existing foundation models for head CT interpretation are ill-suited for emergency use because they target general or chronic-disease assessment and optimize reports for lexical overlap rather than the risk-relevant findings central to emergency triage. Here we present CHIEF, a Chinese-language Head CT Interpretation Emergency Foundation model, pretrained on emergency head CT volumes and paired reports with contrastive, generative, and geometry-regularization objectives. Trained and evaluated on 16,563 examinations from seven hospitals, CHIEF achieved an AUROC of 0.9646 for emergency triage and drafted triage-oriented radiology reports, while also supporting image-to-text retrieval for reference-case support and zero-shot abnormality recognition. CHIEF generated reports of substantially higher quality than those from commercial multimodal large language models, which could not be reliably distinguished from human-written ones by radiologists in a blinded Turing test. Overall, CHIEF provides a generalizable foundation for emergency head CT interpretation and radiologist-in-the-loop clinical decision support.

Medical foundation models compress biomedical data into embeddings that support diverse downstream clinical tasks. However, successful model deployment is hampered by performance degradation on external data. It is recognized that embeddings capture acquisition signatures, such as hardware and technical differences, in addition to biology. Effective harmonization must remove the acquisition signature while preserving biological signals, a trade-off that current methods fail to balance adequately. Input-level normalization fails to eliminate acquisition signatures from embeddings, whereas embedding-level methods adjust features in an untargeted manner. We present FEATMAP, a harmonization approach that models acquisition signatures as geometric distortions between manifolds of similarly arranged embeddings. Using paired data that isolate the effect of acquisition signatures, FEATMAP fits a single global affine transformation per foundation model to correct acquisition signatures directly in the embedding space. This targeted, reusable correction aims to preserve biological and demographic variation while harmonizing across acquisition signatures. Across scanner and foundation-model harmonization in digital pathology and field-strength harmonization in brain MRI, FEATMAP improves cross-condition embedding similarity, reduces performance gaps without retraining, and suggests potential for the alignment of disparate embedding spaces.

Large vision-language models (VLMs) have demonstrated remarkable perfor- mance on computational pathology benchmarks, yet their reliability under adversarial or vacuous inputs remains poorly understood. This paper examines the visual grounding behaviour of two Gemini models Gemini 3.0 Flash Pre- view (gemini-flash) and Gemini 3.1 Pro Preview (gemini-pro) on a well known histopathology classification task, and probes for confabulation using a adver- sarial blank-image set. On the real histopathology dataset both models achieve near-perfect accuracy (98.75% - 100%) across three temperatures (0.0, 0.5, 1.0) and three independent runs. On a controlled adversarial set of blank white images labelled as either benign or malignant, however, a stark divergence emerges. Gemini-flash consistently acknowledges the absence of visual content and assigns zero confidence, while Gemini-pro fabricates detailed, clinically plausible histo- logical descriptions and reports high confidence (mean {approx} 0.95) across the same blank inputs, a behaviour we term confident confabulation. The confabulation rate of gemini-pro reaches 77.8% image-responses at temperature 0.0, dropping to 44.4% at temperature 0.5 and rising to 66.7% at temperature 1.0, while gemini- flash records 0% at all temperatures. These findings raise important questions about the safety and trustworthiness of VLMs in clinical decision-support con- texts, and underscore the need for comprehensive evaluation beyond standard accuracy metrics.

Medical imaging pipelines routinely copy single-channel grayscale data into three identical RGB channels before classification, usually without justification. This study tests whether that step affects model predictions. Four coordinated experiments on bit-identical RGB inputs sorted eleven classical machine learning models into three groups: five that were invariant to the copy, two that were nearly invariant, and four whose predictions changed. On the Kaggle Alzheimer MRI Dataset (6,400 images, four classes, five seeds), five models (AdaBoost, HistGradientBoosting, KNN, SVM_Polynomial, and SVM_RBF) produced identical predictions in both conditions for every seed, where KNN is k-nearest neighbors and SVM a support vector machine, with polynomial and radial basis function (RBF) kernels. Two models (GaussianNB and SVM_Linear) differed by at most one of 1,280 samples, a dataset-dependent gap rather than exact invariance. The remaining four (DecisionTree, ExtraTrees, RandomForest, and LogisticRegression) differed substantively. A regularization sweep on Logistic Regression traced its gap to a single cause. As L2 regularization weakened, the color-minus-grayscale macro F1 gap shrank steadily, from +12.07 percentage points at C=0.001 to near zero at C=100 (paired Wilcoxon p=0.0020 under strong regularization), showing the effect scales with feature count rather than image content. A replication on the OASIS dataset, matched in size and class balance, reproduced every grouping, and the Logistic Regression gap reappeared in the same direction at smaller magnitude (+5.30 points macro F1). Two deep networks, ResNet18 and DenseNet121, gave identical predictions across all twenty paired conditions. Channel triplication left most models unchanged while multiplying classical training time 2.3 to 4.0 times without benefit.

Fairness audits of clinical AI models rarely make the evidentiary status of subgroup findings explicit: reassuring results may reflect insufficient statistical precision rather than true parity, and audit verdicts can easily reverse under equally defensible analytic choices. We introduce an evidence classification scheme that screens for sample size and precision, and integrates stability across design alternatives directly into the fairness claim. We demonstrate this scheme on the estimation of the brain-age gap (BAG), a potential clinical biomarker, from structural MRI using the Alzheimer's Disease Neuroimaging Initiative (ADNI) data. The male-female and Black-vs-White differences, along with the White-Male and Black-Female intersectional contrasts, are all classified as equivalence supported, stable across regressor choice (ridge vs. gradient-boosted trees) and feature representation (full feature set vs. cortical-thickness-only). The Asian-vs-White and Black-Male comparisons remain classified as insufficient data throughout, as neither meets the pre-specified minimum-sample threshold. The proposed scheme provides a path from raw fairness findings to justified fairness claims via pre-specified thresholds, minimum-information screening, and stability checks across declared design choices.

Background: Net Water Uptake (NWU) is a non-contrast CT (NCCT) biomarker of early cerebral edema in ischemic stroke, calculated from attenuation differences between ischemic and contralateral non-ischemic brain regions. Manual NWU quantification is labor-intensive and prone to inter-operator variability, limiting clinical uptake and research scalability. We developed and internally validated a fully automated NWU evaluation pipeline. Methods: We analyzed 24-hour follow-up NCCT scans from the AcT (Alteplase compared to Tenecteplase) trial. Infarcts were automatically obtained by segmentation framework based on a synchronous image-label diffusion probability model. The images and extracted infarcts were registered to the standard MNI152 space, allowing us to mirror the infarct onto the contralateral hemisphere symmetrically, regardless of size or tilt angle. Subsequently, the mirrored region was inversely transformed to return to its original space. Voxels outside the range of 20-80 Hounsfield Units (HU) were excluded to remove non-parenchymal tissue. Automated NWU was computed as the percentage difference in mean HU between infarct and mirrored contralateral regions. The agreement with manually determined NWU was evaluated using Pearson correlation, mean absolute error (MAE), and Bland-Altman analysis. Results: Of 1,327 patients in the trial, 298 (22.5%) met predefined imaging-quality criteria for the manual validation analysis, including well-aligned raw NCCT scans in the axial plane and clear parenchymal infarct segmentations. Automated 24-hour NWU showed excellent agreement with manual measurements (r = 0.99). Mean absolute error was 0.18% (95% CI: 0.01-0.46). Bland-Altman analysis demonstrated minimal bias (0.09%) and satisfactory limits of agreement (-4.05% to +4.24%). Ninety-nine percent of cases fell within {+/-}5% of the manually determined value. Conclusions: Our automated mirrored segmentation pipeline enables accurate and reproducible NWU quantification from routine 24-hour NCCT scans, matching expert manual measurements with minimal bias.

Background: Free-text EEG reports typically lack structure, hindering scalable analysis. We evaluate a large language model (LLM) pipeline to extract structured diagnostic labels and confidence levels from these reports. Methods: We developed a hierarchical annotation schema to classify EEG reports for four specific abnormality types using a four-point confidence scale. To establish ground truth, two certified EEG technicians annotated a diverse dataset of reports authored by neurologists with distinct writing styles. We then implemented a grammar-constrained Mistral-7B pipeline, iteratively prompt-tuned on a development set to mirror these expert annotations. The pipeline's effectiveness was evaluated against the human expert benchmark using core agreement (diagnostic accuracy) and certainty-adjusted agreement (confidence alignment), with classical NLP models serving as a secondary baseline. Results: Mistral-7B significantly outperformed baselines, achieving 96% accuracy for overall abnormality detection, approaching the human benchmark of 98%. Crucially, the model successfully identified rare epileptiform abnormalities where traditional models failed and generalized robustly across distinct reporting styles. While diagnostic accuracy was high, a performance gap persisted in certainty-adjusted agreement, indicating that accurately modeling nuanced clinical confidence remains a challenge. Conclusion: LLMs can effectively automate the extraction of structured diagnostic information from EEG reports with near-human accuracy and strong generalization. While confidence calibration requires further refinement, the combination of accurate classification and explainability makes this pipeline a promising tool for standardizing clinical data at scale. Keywords: Routine Clinical Electroencephalography; Large Language Models; Clinical NLP; Confidence Assessment; Explainable AI; Neurophysiological Evaluation

Introduction Understanding the mechanisms of human neuromaturation constitutes one of the fundamental questions of neuroscience. While it is well described that large-scale brain maturation is initiated within sensorimotor brain regions and progresses to associative cortex, the underlying developmental neurobiology remains to be fully characterized. Animal models have indicated that cortical inhibitory upregulation might be a driver of neurodevelopment. To investigate the hypothesis that cortical inhibitory upregulation plays a similar role in human neuromaturation, we developed a The Virtual Brain (TVB) based computational model (TVB-Child) to explore potential mechanisms of human neurodevelopment. Material and method We created neurodevelopmental dynamic brain network models capturing neurobiological maturation by using the large-scale brain simulator TVB and fitting brain network models to developmental functional MRI (fMRI) from the Human Connectome Project-Development (HCP-D) data set with 640 subjects with an age range of 6-21 years. Age-dependent trajectories in the fMRI data set were first analyzed by combined group-ICA/Dual Regression extracting subject-specific resting-state networks (RSN). Maturational topographical and topological redistribution of these networks were analyzed by linear and non-linear regression of RSN size and degree and strength centrality. Brain network models were fitted to the fMRI functional connectivity obtained from the HCP-D data set. Hypothesizing that cortical inhibition is a driver of neuromaturation, we analyzed spatiotemporal inhibition parameter gradients in the dynamic brain network model for the hypothesized significant correlations with fMRI RSN maturational trajectories. Results While during development frontoparietal (FP) and default mode network (DMN) grew and exhibited an increase in both degree and strength centrality, becoming dominant network hubs, the attention network underwent network pruning with a decrease in size and node degree. The primary sensory network changed little. For the fitted brain network models, we obtained a high degree of reproduction with correlation coefficients between empirical and simulated functional connectivities ranging between 0.80 and 0.95. Values of the feed forward inhibition model parameter wijFFI representing the strength of regional feedforward inhibitory input exhibited the most significant increase with age within the FP and DMN networks. A less pronounced, but significant, age-dependent increase of the inhibitory parameter values were seen in attention networks and no change within primary sensory networks. Conclusion Our study shows that high order (FP, DMN), attention and primary sensory networks exhibit distinct topographical and topological maturation trajectories. Moreover, brain network modeling revealed RSN-specific age-dependent inhibition trajectories, indicating that the model is able to reproduce and thus support candidate mechanisms of neurodevelopment.

We propose a study protocol for routine clinical electroencephalograms (EEGs) from public hospitals, which represents a vast resource for neuroscience research. These non-invasive measures of brain function, paired with rich clinical annotations from large and diverse patient populations, are critical for developing robust artificial intelligence (AI) models and conducting population-level studies. This protocol presents a scalable methodology for curating and harmonizing extensive clinical EEG datasets, encompassing over 40,000 individual studies, to facilitate research applications. Key steps include: (i) integration of raw EEG recordings with corresponding clinical records, including neurological reports, diagnostic codes, and potentially medication data; and (ii) spatial standardization of EEG signals by mapping them to a common brain space defined by functional and anatomical landmarks. The resulting harmonized datasets enable the development of large-scale EEG foundation models, the discovery of novel EEG waveform representations, and the creation of normative "brain charts" for electrophysiological assessment across the lifespan. By enabling standardised, large-scale analyses of real-world clinical EEG data, this protocol supports data-intensive solutions for EEG applications and addresses the challenge of generalising AI models. Our approach promotes the translation of AI tools from research to diverse patient populations, advancing population neuroscience.

While Blood Oxygenation Level-Dependent (BOLD) fMRI remains the gold standard for mapping functional brain networks with MRI, its vascular origins inherently conflate haemodynamic effects with neural activity, limiting its sensitivity in white matter (WM) or its interpretation in neurovascular diseases. Apparent Diffusion Coefficient (ADC) fMRI offers an alternative, diffusion-based contrast that is theoretically more sensitive to neuromorphological coupling and therefore more specific to neuronal activation, though investigated primarily during task-based conditions. This study aimed to comprehensively evaluate the efficacy of isotropic ADC-fMRI in detecting established resting-state networks (RSNs) and to extend this methodology to the investigation of grey-to-white matter (GM-WM) functional connectivity. Our analyses revealed a gradient of ADC detectability shaped by the degree of static functional cohesion and structural tethering of each network. The visual and somatomotor networks, being both highly segregated and strongly anchored to underlying structural pathways, yielded the most robust detection. The default mode network (DMN) and dorsal attention network (DAN) reached group-level significance but with lower effect sizes, and their detection proved fragile across analytical approaches. The frontoparietal network (FPN) and salience network (SAN), whose functional identity is defined by dynamic cross-network reconfiguration, did not reach significance. This gradient partially mirrors the established hierarchy of network segregation observed in BOLD, while further suggesting that ADC sensitivity depends on the structural grounding of each network. Furthermore, ADC demonstrated superior sensitivity to GM-WM functional coupling compared to BOLD. GM-WM functional connectivity profiles derived from ADC were significantly more aligned with underlying structural WM architecture across subjects. Taken together, these findings position isotropic ADC-fMRI as a viable complementary modality to BOLD, offering a more direct window into the neural and structural foundations of brain connectivity.

Human brain magnetic resonance imaging (MRI) revolutionized our ability to non-invasively probe individual differences in neuroanatomy. These anatomical scans, in turn, also allow us to accurately localize functional MRI (fMRI) activity. However, extracting anatomical labels and structural characteristics, such as cortical surface area or thickness, is a computationally demanding task, taking on the order of hours per brain volume. This is an intrinsically multi-scale problem given that local image structure defines fine boundaries, whereas accurate assignments depend on broader anatomical context. Here, we introduce ScaleSurfer, a three-dimensional convolutional vision transformer model based on multi-scale learning. Convolution blocks capture local anatomical detail and a transformer bottleneck integrates the distributed spatial context. This approach provides rapid, whole-brain morphometric feature estimation, including volume, cortical thickness, surface area, and curvature. Importantly, ScaleSurfer accomplishes this nearly five orders of magnitude faster than current pipelines, taking 150-500 ms instead of ~5 hours. We validated ScaleSurfer on multiple datasets, showing stable learning across heterogeneous MRI collections, and demonstrate feasibility by training an interpretable Alzheimer's disease classifier that identifies reductions in primarily medial temporal lobe subregions compared to healthy controls. ScaleSurfer positions multi-scale representation learning as a practical route toward faster, anatomically faithful structural MRI processing, whose speed paves the way for nearly real-time anatomical quality control during scanning.

Moderate-to-severe traumatic brain injury (msTBI) is recognized as a chronic and evolving neurological condition characterized by progressive structural brain changes and persistent cognitive impairment. While prior studies have demonstrated widespread atrophy following msTBI, less is known regarding the longitudinal trajectory of gray matter (GM) changes during recovery and post-rehabilitation. The current study used longitudinal voxel-based morphometry (VBM) to characterize GM volume changes over a period of 9 months, in individuals with msTBI relative to healthy controls (HC). Associations between regional GM volume and neuropsychological functioning were examined. Twenty-eight participants (14 msTBI, 14 HC) completed MRI and neuropsychological assessments across three timepoints spanning outpatient rehabilitation and follow-up. Longitudinal VBM analyses revealed significant group and time interactions within subcortical and limbic regions. Relative to HC, individuals with msTBI showed lower GM volume in these regions at baseline, with trajectories that converged toward HC values (right hippocampus) or increased relative to HC over the rehabilitation period (bilateral pulvinar), whereas the right amygdala and inferior cerebellar vermis remained persistently reduced. Significant longitudinal improvements in memory and psychomotor speed during the rehabilitation period were demonstrated in msTBI. Greater (preserved) GM volume within the right hippocampus, thalamus, and bilateral pulvinar was associated with better performance across measures of verbal memory, processing speed, executive functioning, and cognitive flexibility. These findings suggest that msTBI is associated with dynamic structural brain changes involving subcortical, limbic, and cerebellar networks, and that the rehabilitation period was accompanied by relative volumetric stabilization in these regions and by meaningful cognitive improvement.

Background: Emerging artificial intelligence and machine learning (AI/ML) tools can help generate robust knowledge to support precision rehabilitation approaches for varied patient populations. There is a large amount of research-generated and clinical rehabilitation data available for this purpose; however, a pronounced lack of interoperability prevents large-scale data aggregation. Common data models (CDMs) such as Observational Medical Outcomes Partnership (OMOP) have improved data interoperability across healthcare settings, and more recently, for clinical rehabilitation data, specifically. However, the application of these CDMs to research-generated data has not yet been explored. Therefore, as a foundational step, our study evaluated the breadth and depth of OMOP CDM coverage for data in a multi-site repository of harmonized rehabilitation research data: the Enhancing NeuroImaging Genetics through Meta-Analysis Stroke Recovery (ENIGMA-SR) database. Methods: Two raters independently mapped data elements representing 46 demographics and medical history (DMH) ENIGMA-SR variables and 95 distinct ENIGMA-SR rehabilitation assessments to OMOP standard concepts. Initial rater agreement was assessed for data element inclusion in OMOP and for specific OMOP concepts used (primary metric: Gwet's agreement coefficient [AC]). Mapping differences were reconciled, and final mappings were descriptively analyzed to examine (1) overall OMOP inclusion, (2) inclusion of more granular levels (subscales, items) of complex assessments, and (3) mapped OMOP concept characteristics. Results: Initial rater agreement was good/very good for overall OMOP inclusion of DMH and assessment data elements and for OMOP concepts mapped across almost all assessment data elements (Gwet's AC: 0.79-0.89). Initial OMOP concept agreement was more variable for DMH data elements; however, all mapping differences were successfully reconciled to 100%. Overall, DMH data elements had higher OMOP inclusion than rehabilitation assessments: 84.8% (39/46) vs. 58.9% (56/95). OMOP coverage was particularly limited for complex assessment subscale- and item-level data elements (9.4% [3/32]; 19.2% [14/73]) and did not match the granularity level represented in ENIGMA-SR data for 56.2% (41/73) of complex assessments. DMH and top-level assessment data elements were frequently mapped to multiple OMOP concepts (median: 6, 2; range: 1-23, 1-8), and for > 50% of these data elements the concepts spanned 2-3 different OMOP domains. Conclusion: For ENIGMA-SR, the OMOP CDM has good coverage of DMH data, moderate top-level coverage of rehabilitation assessments, and very limited coverage of assessment subscales and items. This uneven coverage, combined with variability in OMOP concepts and domains mapped to equivalent data points, presents challenges for aggregating clinical and research-generated rehabilitation data into AI/ML-ready datasets. Moreover, software tools currently available to facilitate the mapping process do not effectively accommodate content- and structure-related features inherent to research-generated data. Going forward, the utility of the OMOP CDM to aggregate multi-source rehabilitation data may be improved by expanding the catalogue of OMOP rehabilitation-related concepts, building cross-walks to research-oriented data standards, and adapting emerging computational tools to streamline the mapping process.

A long-standing debate surrounding the neural bases of social concepts concerns the role of anterior temporal lobe (ATL). One perspective suggests ATL subregions are dedicated specifically to social knowledge; another suggests the ATLs constitute a domain-general hub for conceptual knowledge, but with graded functional specialisation depending on connectivity to modality specific spokes. The positions have been difficult to adjudicate due to many confounding factors in tests of social and non-social knowledge. We address these challenges via three innovations in assessment of knowledge in frontotemporal dementia (FTD). First, we introduce a new task that controls for several potential confounds. Second, we apply mixed linear models to behavioural data analysis, allowing further control over confounding factors. Third, we extend the mixed-model approach to lesion-symptom mapping, identifying cortical regions where structural pathology yields a disproportionate impairment on social versus non-social knowledge when other factors are controlled. We used these techniques to probe social and non-social knowledge in FTD subtypes: semantic dementia (SD), associated with asymmetric-bilateral ATL atrophy (n=21), and behavioural-variant (bvFTD), characterised by frontoinsular atrophy (n=24). When confounding factors were controlled, people with SD showed an equal impairment for social and non-social concepts, whereas those with bvFTD were disproportionately impaired on social concepts. The differential impairment of social concepts was associated with atrophy in the insula, orbitofrontal and ventromedial prefrontal cortex and other regions implicated in social knowledge generally. The results suggest that the bilateral ATLs constitute a domain-general semantic hub, whereas ventral prefrontal and insula cortex contribute preferentially to knowledge about people.

Deep learning reads 12 lead electrocardiograms at close to expert level on public benchmarks, yet most reports give one accuracy figure for the whole test set and stop there. We trained three architectures that are standard in this field, a 1D ResNet, a convolutional network with a bidirectional LSTM, and a convolutional network with a bidirectional LSTM followed by a transformer encoder, on the PTB-XL dataset to classify the five diagnostic superclasses, and then looked at how each one performed across sex and age. On the held out fold all three reached a macro AUC near 0.92, in line with the strongest published results on this benchmark, and the simplest model, the 1D ResNet, was marginally the best at 0.9241. The averages hid a steady pattern. Every model scored lower for female patients than for male patients, and every model scored lowest for patients aged 80 and over, where the 1D ResNet fell to 0.8878 and the transformer to 0.8693. Adding complexity did not close either gap and slightly widened the gap by age. Overall accuracy on PTB-XL is close to solved for these model families, but the benefit is not shared evenly, and a single headline number hides the patients a model serves worst. We release the full stratified evaluation to support fairness aware reporting.

Transformer-based deep learning models have shown great potential for decoding visual EEG signals. However, their internal attention mechanisms are often evaluated primarily on optimization objectives, leaving their alignment with biological brain connectivity an open question. This study empirically evaluates how variations in EEG preprocessing strategies affect these attention representations using the Adaptive Thinking Mapper (ATM) model as a framework. We compared a baseline pipeline (MVNN only) against a comprehensive cleaning pipeline integrating ICA and notch filtering. The models were evaluated through cross-generalization, noise robustness, and spectral-temporal ablation analyses. Furthermore, we investigated the structural correspondence between the model's data-driven attention weights and neurophysiological reference networks (GPDC, PDC, and DTF) using Node Strength Correlation and Representational Similarity Analysis (RSA). The results show that the comprehensive preprocessing successfully suppresses non-neural artifacts, such as frontal noise and electrical interference, while maintaining comparable decoding accuracy and baseline robustness. Alignment analyses revealed that the broad spatial organization of the learned attention patterns remains highly stable across pipelines, capturing key directed connectivity dynamics with subtle, metric-dependent variations in global representational geometry. This work provides an empirical exploration into bridging data-driven attention weights with neurophysiological consistency, offering insights toward more transparent brain-computer interfaces.

Clinical informatics pipelines increasingly compute validated clinical endpoints from upstream NLP outputs. Even when the endpoint is defined by an established rubric, translating that rubric across representations - natural language instructions, program logic, and reference implementations - can introduce specification drift, where ostensibly equivalent calculators yield meaningfully different scores. We study this phenomenon for the Expanded Disability Status Scale (EDSS), a standard measure of disability in multiple sclerosis. Holding constant a shared set of functional system (FS) subscores extracted by a large language model (LLM), we compare EDSS values computed across three representations of the same scoring rubric: prompt-executed natural language, LLM-generated code, and a canonical reference implementation. We characterize disagreement structure, distributional shifts, and clinically salient boundary flips, and we propose an audit workflow that treats endpoint computation as a first-class verification target in clinical NLP systems.

Objective: Deep cervical lymphatic-venous anastomosis (LVA) has shown promise in treating Alzheimer's disease (AD), yet no preoperative tool exists to identify potential responders. We developed and evaluated a novel A {beta} PET based scoring system that quantifies regional amyloid burden according to anatomical proximity to the meningeal lymphatic vessels (MLVs) to predict treatment response. Methods: We retrospectively enrolled 58 AD patients who had undergone upper cervical LVA. Eleven regions of interest (ROIs) adjacent to the superior sagittal and straight sinuses were scored based on anatomical proximity to MLVs (higher = closer) and functional relevance to AD (functional score = 1 for AD-related ROIs), yielding a regional assigned score (RAS). Standardized uptake value ratios (SUVRs) were obtained for each ROI. The total SUVR (Stotal) was calculated as {sum}(SUVR x RAS) over all ROIs, and S4+5 was defined as the same sum restricted to ROIs with RAS 4 or 5. These scores, along with baseline demographic characteristics, were evaluated for their ability to predict treatment response using LASSO-logistic regression and receiver operating characteristic (ROC) curve analysis. Results: Forty-one patients (70.7%) were responders. At baseline, responders had significantly higher SUVR of the associative visual cortex (SAVC) (1.68{+/-}0.26 vs. 1.53{+/-}0.12, P=0.0394) and higher S4+5 (32.69{+/-}4.45 vs. 30.14{+/-}3.07, P=0.0358) than non-responders. In univariate analysis, S4+5 was the only significant predictor (OR=1.183, 95% CI: 1.005-1.391, P=0.0433); SAVC was borderline significant (OR=16.654, 95% CI: 0.999-277.63, P=0.0501), while SUVR of the posterior cingulate cortex (SPCC) and Mini-Mental State Examination (MMSE) showed only weak trends (P=0.0714 and P=0.0889, respectively). In the multivariable model, MMSE was independently associated with treatment response (adjusted OR = 1.43, 95% CI: 1.06-1.93, P = 0.022); with SPCC and SUVR of the superior parietal cortex (SsPL) reaching marginal significance (P=0.055 and P=0.051, respectively). The apparent AUC was 0.920, decreasing to a Bootstrap-corrected AUC of 0.780 (95% CI: 0.708-0.884) after optimism correction (optimism = 0.139). The Brier score was 0.097. The covariates-only model yielded a corrected AUC of only 0.574, confirming the incremental value of PET DOI data. Conclusion: This exploratory study introduces a novel A{beta} PET scoring system grounded in MLV anatomy that, combined with baseline MMSE, demonstrates modest predictive potential for LVA response in AD. The findings warrant validation in larger, multicenter cohorts.