Brain Informatics

Neurodegenerative diseases exhibit substantial heterogeneity, complicating both diagnosis and prognosis. Identifying clinically meaningful subtypes is crucial for understanding disease mechanisms and can also improve diagnostic precision and prognostic accuracy. Existing subtyping approaches primarily rely on unsupervised learning of patient data for capturing inter-individual variability, often failing to uncover subtypes that are informative for diagnosis or prognosis. To address this limitation, we propose a novel mixture-of-experts (MoE) framework that integrates predictive modeling with subtype identification. Unlike traditional subtyping methods, our approach learns a router to assign individuals to specialized expert networks, each corresponding to a distinct subtype, to improve predictive accuracy. This MoE framework ensures that the discovered subtypes are not only statistically distinct but also clinically informative. We evaluate the framework on a real-world dataset of mild cognitive impairment (MCI) subjects and a semi-simulated dataset, demonstrating superior performance for predicting MCI subjects progression to Alzheimers disease while identifying distinct clinically meaningful MCI subtypes. Code is available at https://github.com/Kateridge/MoESubtyping.

Traumatic brain injury (TBI) triggers widespread biomarker activation, including astrocytic markers such as glial fibrillary acidic protein (GFAP) and microglia markers such as ionized calcium-binding adapter molecule 1 (IBA1). Quantifying and analyzing these biomarkers are critical for understanding injury impact; however, current methods are labor-intensive and time-consuming. In this study, we propose an automated deep learning framework for dual-biomarker segmentation and TBI classification using GFAP and IBA1 immunofluorescent images. Four U-Net variants: Baseline U-Net, U-Net++, MANet, and LinkNet were trained for segmentation. Three classification models, ResNet50, Swin_T, and MaxViT, were trained to distinguish TBI from control images under single- and dual-biomarker conditions. The baseline U-Net achieved the highest segmentation Dice score for GFAP (0.9259), while the U-Net++ achieved the highest Dice score for IBA1 (0.9676). Trained segmentation models demonstrated significantly better performance compared to QuPath alternatives. While GFAP alone supported high classification accuracy, IBA1 alone was less effective. Multimodal fusion of GFAP and IBA1 significantly improved classification performance across all models, with Swin_T achieving the highest overall accuracy (0.9489), and ResNet50 achieving the highest F1-score (0.9499). These findings demonstrate that integrating complementary biomarkers enhances automated TBI classification, and deep learning offers a robust alternative to manual analysis for immunofluorescent brain injury imaging. This framework is scalable to additional biomarkers and injury models, offering a reproducible approach to accelerate biomarker research.

Accurate stroke lesion segmentation is essential for large-scale neuroimaging studies, yet manual delineation remains labor-intensive, and existing automated methods often struggle to generalize across imaging protocols and stages of recovery. We developed MAESTRO, a deep learning framework for automated lesion segmentation across the stroke recovery continuum using T1-weighted (T1) MRI alone. We hypothesized that combining a transformer-based architecture with an image augmentation strategy would improve segmentation accuracy and robustness under heterogeneous imaging conditions. T1 MRI scans and expert-traced lesion masks from 955 stroke participants across 33 international cohorts were used to train and evaluate MAESTRO within the open-source nnU-Net framework. Performance was evaluated on a held-out test set using spatial and volumetric agreement metrics. An exploratory human-in-the-loop (HITL) evaluation compared correction of MAESTRO-generated segmentations with manual tracing from scratch. MAESTRO achieved the strongest performance across several evaluated model configurations, providing the most accurate lesion localization and lesion volume estimates (median Dice = 0.686; Pearson r = 0.861; ICC = 0.792). Segmentation performance was sensitive to lesion size and stroke chronicity but remained robust across diverse imaging conditions. Additionally, using a HITL workflow to correct MAESTRO segmentations reduced annotation time by 47.4% compared to manual tracing while improving accuracy relative to both automated and manual workflows. MAESTRO is publicly available to enable robust, automated stroke lesion segmentation from T1 MRI. When combined with human review and correction, MAESTRO offers a practical approach for generating standardized, high-quality lesion annotations, helping reduce a major practical barrier to large-scale stroke imaging studies.

INTRODUCTION: The biomarker-based amyloid/ tau/ neurodegeneration (A/T/N) framework has become a popular staging method for Alzheimer's disease (AD) research. Previous studies use the framework either as a rule-based or data-driven approach but typically sacrifice either adaptivity or interpretability. METHODS: We present an interpretable, hybrid method, called Neurosymodal Data Fusion, for predicting incident AD in the ADNI dataset. Specifically, we encode the A/T/N framework as a logic program, where the input biomarker features are extracted by one or more neural networks. RESULTS: Our pipeline predicted four-year incident AD with a sensitivity of up to 0.84. Additionally, our models learned scores for each A/T/N profile, denoting relative importances to model predictions. These scores also indicated that empirically-derived cut-off values for the A and T criteria might be uninformative for the ADNI data. DISCUSSION: Our pipeline provides a novel way to use the A/T/N framework that could potentially improve early AD screening years before clinical manifestations.

Automated interpretation of neuroimaging studies requires simultaneous assessment of multiple imaging evidence variables, each tied to distinct anatomical structures. Vision-language models (VLMs) offer a unified framework for multi-task analysis, but adapting pre-trained VLMs remains challenging. Full fine-tuning is computationally prohibitive, and joint multi-task training requires simultaneous access to all task data, which is often infeasible in clinical settings. Although model merging enables multi-task composition without joint re-training, existing methods focus on post-hoc algorithms with limited extension to VLMs and minimal application to neuroimaging. Here, we present GRadient-guided Adapter Merging (GRAM), a layer-selective low-rank adaptation (LoRA)-based fine-tuning and merging framework for multi-task neuroimaging visual question-answering (VQA). GRAM uses a gradient ratio that contrasts class-specific gradients to identify task-discriminative layers, and applies subspace-constrained projected gradient descent to restrict LoRA updates to directions consistent with the geometry of the pre-trained model. We leveraged a structured VQA benchmark, developed from the National Alzheimer's Coordinating Center (NACC) dataset, that pairs multi-sequence brain MRI studies with question-answer pairs across clinically relevant imaging evidence variables. Experiments on the VQA benchmark showed that GRAM outperformed or matched all-layer LoRA fine-tuning and a standard merging baseline while reducing inter-task interference during merging, and approached or surpassed the performance of joint multi-task training without joint re-training.

Acute ischemic stroke (AIS) is a leading cause of disability and death while effective treatment requires quick and accurate diagnosis. Non-contrast CT (NCCT) is widely used in the initial screening of AIS, but stroke detection is challenging because early changes on NCCT are subtle or indistinguishable. Using hyperacute NCCTs as inputs and diffusion-weighted MRI as ground truth, we trained a deep learning algorithm to classify patients with AIS and segment the stroke lesions. We hypothesized that this approach would accurately detect hyperacute tissue density changes on NCCT. For the classification task, our ResNet50 model delivered the best performance (with 98.5% accuracy, 97.4% precision, and 100% recall on an evaluation set). Classification performance remained strong when restricted to lesions smaller than 5 mL, which constituted the majority of our evaluation cases. For the segmentation task accomplished using a range of U-Net architectures, performance was acceptable for large lesions and declined sharply for smaller lesions. Together, these findings demonstrate the feasibility of deep learning for AIS detection and represent a step towards faster triage and treatment for stroke patients.

Cerebral microbleeds (CMBs) are established imaging biomarkers of cerebral small vessel disease and are a defining feature of cerebral amyloid angiopathy (CAA), yet their automated detection in susceptibility-weighted imaging (SWI) remains challenging due to high false-positive rates from vessel cross-sections, iron and calcium deposits, and other hypointense mimics. We present a fully automated, three-stage cascade pipeline that combines subject-adaptive unsupervised candidate generation with two successive lightweight 3D ResNet classifiers, trained with only human-in-the-loop quality-assurance (QA) labels (yes/no per candidate) rather than dense voxel-wise segmentation masks. The candidate generation stage is performed by fitting a Gaussian Mixture Model (GMM) to each subjects SWI intensity histogram to define an adaptive low-intensity threshold, followed by anatomical masking to exclude physiologically irrelevant regions (image edges, ventricles/CSF/choroid plexus, and cerebellum), and filters candidates by size and sphericity. The model was trained and evaluated across a nested 3x5-fold cross-validation on N = 30 subjects from a publicly available labeled microbleed dataset and a CAA cohort (11,424 CMB candidate lesions) with data augmentation during training. Stage A classifies all geometric candidates as CMB or non-CMB and Stage B refines the predicted positives to suppress false positives (cascade AUC = 0.9587, sensitivity = 0.712, specificity = 0.975, PPV = 0.676, F1 = 0.693). The cascade reduces Stage A false positives by 76.8% (888/1,157 false positives eliminated) while retaining competitive sensitivity. Inference was performed on 141 SWI scans, detecting a mean 40.3 CMBs per scan and being preferred for use in 85% of high CMB cases, as evaluated by a blinded neurologist. The inference pipeline outputs binary CMB segmentation NIfTI images and radiologist-ready QA visualizations.

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.

We present FootNet, a 453-image multi-view smartphone foot dataset for binary foot segmentation, with expertannotated masks across six anatomical views (dorsal, medial, and plantar, both left and right). We benchmark four segmentation models under a controlled protocol: U-Net with a MobileNetV2 encoder achieves the best performance (IoU 0.9268, Dice 0.9608, 95 % CI [0.9209, 0.9320]); DeepLabV3 with MobileNetV3-Large scores IoU 0.8984 (Dice 0.9449); UNet++ with MobileNetV2 scores IoU 0.8913 (Dice 0.9391); and SAM ViT-B with oracle boundingbox prompt scores IoU 0.9219 on the matched 191-image subset. Bonferroni-corrected Wilcoxon signed-rank tests (k = 6 comparisons) show U-Net significantly outperforms DeepLab (p < 0.001, r = 0.638) and SAM ViT-B with oracle boundingbox (p = 0.005, r = 0.202); UNet++ does not significantly differ from DeepLab (p = 0.062). Connected-component postprocessing yields negligible benefit (mean {triangleup}IoU = +0.0003, 12 of 453 images improved). The extended dataset is available upon request

Analytical variability in neuroimaging pipelines contributes to concerns about reproducibility in the field. In structural MRI, different segmentation tools produce discrepant morphometric estimates that may influence downstream analyses, such as predictive modeling. We tested whether integrating several segmentation pipelines improves brain age prediction, and characterized the spatial and demographic structure of pipeline differences across several datasets. T1-weighted scans from five open-access datasets were processed with four widely-used structural segmentation pipelines. Brain age models were trained using single-pipeline features and compared with multi-pipeline aggregation strategies. Inter-pipeline variability was assessed across shared subcortical structures and examined in relation to age and sex. Integrating features across distinct segmentation frameworks improved predictive performance relative to individual pipelines, whereas aggregation within closely related software versions provided limited benefit. Variability was spatially structured and volumetric measures were often systematically associated with age and sex. These results suggest that segmentation differences reflect structured, demographically sensitive variation rather than random noise, and that multi-pipeline feature integration can enhance robustness in neuroimaging-based prediction.

Explainability methods applied to deep learning models for Alzheimer's disease neuroimaging produce attribution maps that vary substantially across methods and architectures, yet no validated quantitative framework exists for determining which method most faithfully localises attribution signal within established AD biomarker anatomy at the individual subject level. Existing validation approaches rely on group-level comparisons or qualitative visual inspection, leaving individual-level biomarker alignment uncharacterised. We introduce the Biomarker Fidelity Score (BFS), a quantitative tool measuring spatial overlap between individual-level 3D explainability attention maps and atlas-registered AD-relevant neuroimaging ROIs across thirteen anatomically defined structures including hippocampus, entorhinal cortex, amygdala, and parahippocampal gyrus. Five explainability methods (GradCAM++, Integrated Gradients, DeepSHAP, LRP, ScoreCAM) were benchmarked across three volumetric architectures (3D ResNet-18, DenseNet-121, Swin-UNETR) on 327 balanced ADNI-3 subjects. Integrated Gradients achieved the highest BFS across all architectures while GradCAM++ consistently showed the lowest biomarker alignment (all p<0.001, Friedman test). The complete BFS pipeline replicated these rankings without retraining on 207 independent OASIS-3 subjects, with maximum absolute difference of 0.0005 across all fifteen method-architecture combinations and Spearman rank correlation of 0.964 between cohort rankings. By offering an externally validated, individual-level, biomarker-grounded quantitative standard, BFS equips clinicians and AI developers with practical guidance for selecting trustworthy explainability methods in AD neuroimaging.

Longitudinal dementia progression prediction is essential for clinical decision-making. However, models often degrade on external cohorts due to systemic missingness -- where certain biomarkers available during training are completely absent at test time -- compounded by distribution shifts and patient-specific variability. Here, we propose Progression-aware Feature Fusion with Test-Time Adaptation (ProFuse-TTA), a two-stage hierarchical Transformer for longitudinal dementia prediction. Stage 1 learns per-biomarker temporal representations from irregular observations without imputation. Stage 2 fuses them via cross-feature attention, with simulated modality dropout during training for robustness to systemic missingness. At inference, a lightweight test-time adaptation module performs per-individual calibration. We trained on ADNI and evaluated on three external cohorts comprising 2,316 participants and 13,205 timepoints, with controlled modality ablation experiments isolating the effect of systemic missingness. We compared against six baselines, four from a recent benchmark study and two new baselines including one built on a tabular foundation model. ProFuse-TTA achieved the best cross-dataset performance in 8 of 9 settings across clinical diagnosis, MMSE, and hippocampal volume prediction, and ranked first in 14 of 15 ablation scenarios. The model maintained superior performance across varying input lengths and prediction horizons up to 6 years. Pretrained ADNI models are available at XXX.

Mapping hierarchical brain networks within traditional Euclidean space causes significant structural distortion, undermining neuroimaging diagnostic frameworks. While hyperbolic models like the Poincare ball preserve these nested topologies, they demand heavy computational overhead due to intricate Mobius operations and curved geodesics. This paper introduces a highly efficient non-Euclidean framework for analyzing neurocognitive decline utilizing the Beltrami-Klein ball model. By projecting hyperbolic geodesics as Euclidean straight lines, this approach converts complex distance calculations into simple dot products, radically reducing processing demands. We validated our methodology against state-of-the-art Poincare and Lorentz baselines using datasets for Schizophrenia, Parkinsons Disease, and Alzheimers Disease. The Klein-based framework demonstrates superior performance, delivering both higher diagnostic precision and accelerated processing velocities across all three neurocognitive disorders.

Understanding brain-behavior relationships requires models capturing the distributed, interactive, and multiscale nature of neural systems. Traditional univariate approaches and single-biomarker models are inherently limited in this context, as they fail to represent dependencies across regions and the hierarchical organization of brain networks. In this study, we propose a graph-based multivariable framework for brain imaging analysis that integrates key organizational principles of brain function-including segregation, integration, modularity, and temporal dynamics-within a unified graph neural network architecture. The framework represents brain data as hierarchical graphs, where node features encode regional activation and temporal variability, and graph structure captures interactions within and between functional modules. The proposed approach is evaluated using functional near-infrared spectroscopy (fNIRS) data as a case study, where subject-specific brain graphs are constructed from task-based recordings acquired shortly after cochlear implant activation to predict speech understanding outcomes one year later. Under leave-one-subject-out validation, the model demonstrates strong predictive performance (R = 0.73, p < 0.001), outperforming previously reported single-biomarker approaches. Perturbation-based analyses further show that predictions are driven by distributed patterns of activity and interaction across regions and modalities, rather than isolated features. These results illustrate the capability of the proposed framework to capture complex brain organization and highlight its potential as a generalizable platform for multivariable analysis and prediction in neuroimaging applications beyond the specific clinical use case considered here.

Focal cortical dysplasia (FCD) is a principal cause of pharmacoresistant focal epilepsy, yet its structural MRI signature, subtle cortical thickening, blurring of the gray-white matter junction, is frequently undetected even by experienced neuroradiologists, delaying or precluding referral for curative surgical resection. Here we develop a machine learning pipeline for FCD detection that prioritizes mechanistic interpretability over model complexity. In a subsample of 50 subjects (25 FCD, 25 age-matched controls) drawn from a public structural MRI cohort, we register all scans to a common stereotactic template and derive hemispheric asymmetry features across 48 cortical regions, exploiting the characteristic unilaterality of FCD pathology. Among four classifiers evaluated under leave-one-out cross-validation, an L1-regularized logistic regression achieves the highest accuracy (78\%, permutation p=0.02), substantially outperforming tree-based ensembles, which perform at or below chance in this feature-to-sample regime. The fitted model selects a sparse subset of 21 of 96 features, with the largest-magnitude contributions localized to inferior and middle frontal gyri and temporal pole and superior temporal gyrus, regions consistent with the known anatomical distribution of FCD. These findings indicate that hemispheric asymmetry, combined with a sufficiently regularized, interpretable classifier, captures a modest but statistically robust and anatomically grounded signal for FCD detection, offering a transparent complement to existing deep learning approaches for presurgical evaluation.

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.

Understanding vascular contributions to disease is a major unmet need. White matter lesions (WML) are an accepted imaging marker of cerebral small vessel disease, giving insights into its related pathologies. A unified approach for WML analyses in large multi-site data is lacking despite the need for pooling of data to overcome the limitations of often small heterogenous MRI studies which make subtyping and identifying patterns within disease groups difficult. Our ENIGMA-PD-WML pipeline is an open-source containerized pipeline containing all the code and packages required for pre-processing, processing and post-processing of T1-weighted and FLAIR data, outputting accurate and reproducible binary WML maps using a UNet approach. The pipeline provides a standardized image analysis approach for WML and outputs data in both native and MNI space to allow for sharing and pooling of data from multiple sites for large-data analysis. In addition to a reliable standardized approach for WML segmentation, key priorities when developing the pipeline included: usability, i.e., requiring minimal manual input and technical expertise to use, and suitability to run on various MRI scanners and acquisition parameters as is common in multi-site data. This paper describes the pipeline in detail, with rationale for each step, providing transparency and facilitating its usage to overcome reproducibility issues in large-scale WML analyses.

Objective: Accurate volumetric analysis of the brain and cerebrospinal fluid (CSF) is essential for monitoring hydrocephalus, a significant pediatric neurological condition. While computed tomography (CT) provides high-quality volumetric assessment, its associated ionizing radiation poses risks, especially for children. Low-field magnetic resonance imaging (LF-MRI) offers a safer and more accessible alternative, particularly in resource-constrained settings. However, its lower resolution and increased susceptibility to structural distortions make accurate segmentation challenging. This study aims to demonstrate that reliable volumetric measurements can be obtained from LF-MRI that are comparable to CT, enabling safer and more frequent monitoring of infants with hydrocephalus. Approach: We propose EnSegNet-Cross, a cross-modality, enhancement-aware segmentation network for brain volume analysis using LF-MRI. The framework leverages high-fidelity CT data during training but requires only LF-MRI during inference. At the core of the framework is a novel cross-modal topological penalty designed to minimize discrepancies between predicted LF-MRI and CT structures. A central contribution is the integration of a three-dimensional topological loss based on persistent homology, which penalizes topological discrepancies in CSF regions, specifically CSF holes formed by enclosed brain parenchyma, between CT and LF-MRI segmentations. Incorporating these structural priors facilitates generalization across heterogeneous clinical cases while eliminating the need for CT data during inference, resulting in more anatomically coherent and topologically faithful segmentations. Main Results: On a curated cohort of infants with hydrocephalus who had paired LF-MRI and CT scans, including cases with infectious and non-infectious causes, EnSegNet-Cross consistently outperformed state-of-the-art machine learning alternatives. It achieved the highest Dice score of 0.8532 plus/minus 0.03 and Volume Score of 0.9318 plus/minus 0.03. The method also demonstrated robust performance in challenging cases with confounding factors, achieving a Dice score of 0.8340 plus/minus 0.03 and a Volume Score of 0.9111 plus/minus 0.05. By leveraging CT-derived topological priors, EnSegNet-Cross successfully handled anatomically complex scenarios in which conventional models failed. Significance: EnSegNet-Cross provides a reliable and interpretable solution for brain and CSF segmentation, particularly in complex cases of hydrocephalus. This study demonstrates that high-fidelity volumetric estimates can be achieved using only LF-MRI, facilitating frequent, radiation-free monitoring. By bridging the fidelity gap between low-quality LF-MRI and high-resolution CT through clinically grounded enhancement and topological supervision, EnSegNet-Cross offers a robust clinical tool for brain volumetric analysis in infants with hydrocephalus using LF-MRI.

Early detection of schizophrenia (SZ) remains challenging due to the subtlety of early-stage brain alterations and reliance on subjective clinical assessment. We propose a frequency-aware 3D convolutional neural network (CNN) pipeline that integrates NeuroMark-HiFi high-pass spatial filtering with a modified VGGNet3D architecture featuring 3D Laplacian kernel initialization and dilated convolutions. Using the FBIRN dataset (N=311; 150 healthy controls, 161 SZ) with all 53 intrinsic connectivity networks (ICNs) per subject, we evaluate four experimental conditions across two hyperparameter configurations to isolate the contributions of enhanced input representations and frequency-aware model design. Under the optimized configuration, Condition 3 (HiFi + Laplacian initialization) achieved the best mean test accuracy of 75.54% with a peak single-fold accuracy of 87.10%, representing a 5.44% absolute gain over the optimized baseline. These results demonstrate that high-frequency spatial features are more discriminative for SZ classification than raw intensities, and that aligning Laplacian-initialized kernels with HiFi-filtered input creates a beneficial inductive bias--even with a compact model of approximately 1.4M parameters.

Stroke is a leading cause of death and long-term disability worldwide, affecting approximately 15 million individuals annually. Prompt and accurate subtype differentiation between ischemic and hemorrhagic stroke is clinically critical, as the two conditions demand diametrically opposite interventions - thrombolytic therapy versus surgical decompression. Yet the majority of existing deep learning approaches reduce this problem to binary detection, and virtually none address the opacity of their decision-making in a clinically actionable manner. We present CerebAI, an explainable, deployment-oriented three-class CT stroke classification system built on a fine-tuned ConvNeXt-Base backbone with Integrated Gradients (IG) attribution. Trained on 6,774 non-contrast CT scans stratified across No Stroke, Ischemic Stroke, and Hemorrhagic Stroke, CerebAI achieves a weighted F1-score of 0.9746 (95% CI: [0.9625, 0.9851]), accuracy of 97.47%, macro-averaged AUC of 0.9921, mean Intersection-over-Union (mIoU) of 0.9276, Expected Calibration Error (ECE) of 0.0115, mean Brier Score of 0.0150, and Cohen's {kappa} of 0.9483 - surpassing ResNet-50, EfficientNet-B4, and Vision Transformer (ViT-B/16) baselines across all reported metrics. Integrated Gradients produce pixel-precise saliency maps that localize pathological regions with greater anatomical fidelity than Gradient-weighted Class Activation Mapping (Grad-CAM), a finding we support with side-by-side qualitative comparison. CerebAI additionally incorporates a native DICOM processing pipeline to facilitate future clinical translation. Code and model weights are publicly available to support reproducibility and further research.