ASAdam M. Saunders

Adam M. Saunders

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Adam M. Saunders has a total of 119 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Contrastive learning and advanced bundle metrics outperform fractional anisotropy for Alzheimer’s disease diffusion MRI prediction: findings from the Beyond FA challenge

    by Elyssa McMaster, Nancy Newlin, Trent Schwartz, Adam Saunders, Gaurav Rudravaram, Jiheng Li, André Hucke, Yehyun Suh, Jongyeon Yoon, Eleanor Liard, et al.

    PurposeThe inclusion of diffusion-weighted magnetic resonance imaging (DW-MRI) in major national studies has allowed for quantification and characterization of white matter macrostructure and microstructure in cases of disease and disorder. Diffusion tensor imaging (DTI) yields 3D scalar maps to summarize the DW-MRI signal with quantitative metrics. Fractional anisotropy (FA) is a DTI metric that describes the degree of anisotropy within diffusion and is frequently used to interpret white matter integrity, but it exhibits high sensitivity in pathological interpretation. These properties of FA make it difficult to use as a consistent candidate imaging biomarker, as it makes different ages, cognitive statuses, and pathologies difficult to discern. With the development of other white matter models and metrics, such as structural connectivity network measures and tractography bundle analyses, there are many combinations to analyze for white matter integrity analysis, especially in the context of lower-quality clinical data that FA’s voxel-wise computation can be sensitive to. The widespread collection of DW-MRI data necessitates an analysis of these metrics to interpret white matter integrity beyond what FA can capture.ApproachWe designed a challenge for the medical imaging community to compare crowdsourced DW-MRI candidate imaging biomarkers to a baseline relying on FA in an age, sex, and cognitive status classification task and variables unknown to teams during the challenge. After the challenge submission period ended, we evaluated methods based on sensitivity to environmental variables (humidity and temperature on the day of the scan), and we correlated features with known Alzheimer’s disease (AD) pathology (amyloid beta 42, tau, and ptau cerebrospinal fluid/CSF readings) to evaluate if the methods collected in this study are robust to the limitations of FA. The key innovations are that we crowdsource DW-MRI feature representations for AD, we develop a framework to compare imaging feature representations in this context, we evaluate the sensitivity of DW-MRI imaging candidate biomarkers to environmental variables, and we correlate the novel features to known AD pathological measures.ResultsWe find that bundle-based methods outperform whole-brain methods in age classification tasks, but deep learning, computer vision-based methods show a new avenue for cognitive status prediction tasks.ConclusionThe submissions to our challenge show that there are many approaches to consider when developing new AD feature representations.

  • Charting confidence in white matter brain charts: enabling study planning through stability validation

    by Michael E. Kim, Gaurav Rudravaram, Adam M. Saunders, Chenyu Gao, Karthik Ramadass, Nancy R. Newlin, Praitayini Kanakaraj, Sam Bogdanov, Alexandra Roche, Dario J. Englot, et al.

    PurposeNormative modeling of quantitative brain measurements is being widely adopted as a promising method for charting population-level developmental trajectories and identifying abnormalities in groups and individuals. Generalized additive models for location, scale, and shape (GAMLSS) are a statistical framework that has been identified by the World Health Organization as a robust method for large-scale population modeling. Validating the stability of normative models is essential for appropriate benchmarking and detecting anomalies, especially for patients with neurodegenerative or developmental disorders. As these models continue to expand in the number of features and amount of data, it becomes increasingly important to assess the stability of the models and their predictions.ApproachAn analytic approach is extended from previous work to assess variability of the GAMLSS framework for normative modeling of white matter measurements across the lifespan. The analytic variance and bias are compared with empirical bootstrapping estimates and those of an analogous model that is a nonparametric, data-driven analog.ResultsAcross all models, the analytic approach showed low model variability (<0.5% Coefficient of Variation) for lifespan trajectories that increased with age. Empirical variability of trajectories was slightly elevated by comparison, especially at the beginning of the lifespan. The trajectory variability for the data-driven analog was even larger, showing a ten-fold increase from the empirical GAMLSS variability. For individual centile prediction, the majority of data points for the analytic assessment had a mean absolute difference (MAD) in centile score prediction as z-scores that was less than 0.003 with the standard deviation of these errors below 0.06. The empirical assessment showed an elevated standard deviation (0.15) and MAD of z-score centile predictions (0.06) for the majority of datapoints. By contrast, when compared with the empirical assessment, the nonparametric, data-driven analog showed over a two-fold increase in the MAD of centile z-score predictions (0.14) and a three-fold increase in standard deviation (0.45) for the majority of data points. The divergences in variability across assessments suggest that individual centile predictions are more sensitive to random fluctuations in the data, which the nonparametric data-driven approach exacerbates.ConclusionsResults suggest that the GAMLSS brain charts are a stable method of defining and charting normative white matter development as well as planning future clinical or research studies. Overall, this study provides an assessment of GAMLSS-derived brain chart variance in a large-scale population.

  • Open Access

    A geometric-to-neural cascade for cerebral microbleed detection in susceptibility-weighted MRI

    by Sam Bogdanov, Gaurav Rudravaram, Adam M. Saunders, Michael E. Kim, James LeFevre, Jack Charles, Stuti Jain, Matthew S. Schrag, Bennett A. Landman

    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 subject’s 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 3×5-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.

  • Phenotype discovery of traumatic brain injury segmentations from heterogeneous multi-site data

    by Adam M. Saunders, Michael E. Kim, Gaurav Rudravaram, Lucas W. Remedios, Chloe Cho, Elyssa M. McMaster, Daniel R. Gillis, Yihao Liu, Lianrui Zuo, Bennett A. Landman, et al.

    Traumatic brain injury (TBI) is intrinsically heterogeneous, and typical clinical outcome measures like the Glasgow Coma Scale complicate this diversity. The large variability in severity and patient outcomes render it difficult to link structural damage to functional deficits. The Federal Interagency Traumatic Brain Injury Research (FITBIR) repository contains large-scale multi-site magnetic resonance imaging data of varying resolutions and acquisition parameters (25 shared studies with 7,693 sessions that have age, sex and TBI status defined – 5,811 TBI and 1,882 controls). To reveal shared pathways of injury of TBI through imaging, we analyzed T1-weighted images from these sessions by first harmonizing to a local dataset and segmenting 132 regions of interest (ROIs) in the brain. After running quality assurance, calculating the volumes of the ROIs, and removing outliers, we calculated the z-scores of volumes for all participants relative to the mean and standard deviation of the controls. We regressed out sex, age, and total brain volume with a multivariate linear regression, and we found significant differences in 37 ROIs between subjects with TBI and controls (p < 0.05 with independent t-tests with false discovery rate correction). We found that differences originated in 1) the brainstem, occipital pole and structures posterior to the orbit, 2) subcortical gray matter and insular cortex, and 3) cerebral and cerebellar white matter using independent component analysis and clustering the component loadings of those with TBI.

  • DeepFixel: crossing white‑matter fiber identification through spherical convolutional neural networks

    by Adam M. Saunders, Lucas W. Remedios, Elyssa M. McMaster, Jongyeon Yoon, Gaurav Rudravaram, Adam Sadriddinov, Praitayini Kanakaraj, Bennett A. Landman, Adam W. Anderson

    Diffusion-weighted magnetic resonance imaging allows for reconstruction of models for structural connectivity in the brain, such as fiber orientation distribution functions (ODFs) that describe the distribution, direction, and volume of white matter fiber bundles in a voxel. Crossing white matter fibers in voxels complicate analysis and can lead to errors in downstream tasks like tractography. We introduce one option for separating fiber ODFs by performing a nonlinear optimization to fit ODFs to the given data and penalizing terms that are not symmetric about the axis of the fiber. However, this optimization is non-convex and computationally infeasible across an entire image (approximately 1.01 × 106 ms per voxel). We introduce DeepFixel, a spherical convolutional neural network approximation for this nonlinear optimization. We model the probability distribution of fibers as a spherical mesh with higher angular resolution than a truncated spherical harmonic representation. To validate DeepFixel, we compare to the nonlinear optimization and a fixel-based separation algorithm of two-fiber and three-fiber ODFs. The median angular correlation coefficient is 1 (interquartile range of 0.00) using the nonlinear optimization algorithm, 0.988 (0.317) using a fiber bundle elements or “fixel”-based separation algorithm, and 0.973 (0.004) using DeepFixel. DeepFixel is more computationally efficient than the non-convex optimization (0.32 ms per voxel). DeepFixel’s spherical mesh representation is successful at disentangling at smaller angular separations and smaller volume fractions than the fixel-based separation algorithm.

  • Open Access

    Lifespan Pancreas Morphology for Control Versus Type 2 Diabetes Using AI on Largescale Clinical Imaging

    by Lucas W. Remedios, Chloe Cho, Trent M. Schwartz, Dingjie Su, Gaurav Rudravaram, Chenyu Gao, Aravind R. Krishnan, Adam M. Saunders, Michael E. Kim, Shunxing Bao, et al.

    Understanding how pancreas size and shape change with normal aging is critical for establishing a baseline to detect deviations in type 2 diabetes and other pancreatic disease. We measure pancreas size and shape using morphological measurements from early development through aging (ages 0–90). Our goals are to (1) identify reliable clinical imaging modalities for artificial intelligence (AI) based pancreas measurement, (2) establish normative morphological aging trends, and (3) detect potential deviations in type 2 diabetes. We analyzed a clinically acquired dataset of 2533 patients imaged with abdominal computed tomography (CT) or magnetic resonance imaging (MRI). The patients did not have cancer, pancreas pathology, sepsis, or trauma. We resampled the scans to 3 mm isotropic resolution, segmented the pancreas using automated methods, and extracted 13 morphological pancreas features across the lifespan. First, we assessed pancreas volume trajectories in 1858 control patients across contrast CT, non‐contrast CT, and MRI to determine which modalities provide consistent lifespan trends. Second, we characterized distributions of normative morphological patterns stratified by age group and sex. Third, we used covariate‐adjusted generative additive models for location, scale, and shape (GAMLSS) regression to model pancreas morphology trends in 1350 patients matched for age, sex, and type 2 diabetes status to identify any deviations from normative aging associated with type 2 diabetes. We selected CT for the main analyses of this study, since the MRI appeared to yield different pancreas measurements than CT using our AI‐based method on this dataset of clinically acquired scans. When adjusting for confounders, the aging trends for 10 of 13 morphological features were significantly different between patients with type 2 diabetes and non‐diabetic controls ( p < 0.05 after multiple comparisons corrections). Additionally, we characterized normative morphological aging trends of the pancreas across 13 morphological measurements. We provide lifespan trends demonstrating that the size and shape of the pancreas are altered in type 2 diabetes using 675 control patients and 675 diabetes patients. Moreover, our findings reinforce that the pancreas is smaller in type 2 diabetes. Additionally, we contribute a reference of lifespan pancreas morphology from a large cohort of non‐diabetic control patients in a clinical setting.


  • Open Access

    DeepFixel: Crossing white matter fiber identification through spherical convolutional neural networks

    by Adam M. Saunders, Lucas W. Remedios, Elyssa M. McMaster, Jongyeon Yoon, Gaurav Rudravaram, Adam Sadriddinov, Praitayini Kanakaraj, Bennett A. Landman, Adam W. Anderson

    Diffusion-weighted magnetic resonance imaging allows for reconstruction of models for structural connectivity in the brain, such as fiber orientation distribution functions (ODFs) that describe the distribution, direction, and volume of white matter fiber bundles in a voxel. Crossing white matter fibers in voxels complicate analysis and can lead to errors in downstream tasks like tractography. We introduce one option for separating fiber ODFs by performing a nonlinear optimization to fit ODFs to the given data and penalizing terms that are not symmetric about the axis of the fiber. However, this optimization is non-convex and computationally infeasible across an entire image (approximately 1.01 x 106 ms per voxel). We introduce DeepFixel, a spherical convolutional neural network approximation for this nonlinear optimization. We model the probability distribution of fibers as a spherical mesh with higher angular resolution than a truncated spherical harmonic representation. To validate DeepFixel, we compare to the nonlinear optimization and a fixel-based separation algorithm of two-fiber and three-fiber ODFs. The median angular correlation coefficient is 1 (interquartile range of 0.00) using the nonlinear optimization algorithm, 0.988 (0.317) using a fiber bundle elements or "fixel"-based separation algorithm, and 0.973 (0.004) using DeepFixel. DeepFixel is more computationally efficient than the non-convex optimization (0.32 ms per voxel). DeepFixel's spherical mesh representation is successful at disentangling at smaller angular separations and smaller volume fractions than the fixel-based separation algorithm.

  • Open Access

    Phenotype discovery of traumatic brain injury segmentations from heterogeneous multi-site data

    by Adam M. Saunders, Michael E. Kim, Gaurav Rudravaram, Lucas W. Remedios, Chloe Cho, Elyssa M. McMaster, Daniel R. Gillis, Yihao Liu, Lianrui Zuo, Bennett A. Landman, et al.

    Traumatic brain injury (TBI) is intrinsically heterogeneous, and typical clinical outcome measures like the Glasgow Coma Scale complicate this diversity. The large variability in severity and patient outcomes render it difficult to link structural damage to functional deficits. The Federal Interagency Traumatic Brain Injury Research (FITBIR) repository contains large-scale multi-site magnetic resonance imaging data of varying resolutions and acquisition parameters (25 shared studies with 7,693 sessions that have age, sex and TBI status defined - 5,811 TBI and 1,882 controls). To reveal shared pathways of injury of TBI through imaging, we analyzed T1-weighted images from these sessions by first harmonizing to a local dataset and segmenting 132 regions of interest (ROIs) in the brain. After running quality assurance, calculating the volumes of the ROIs, and removing outliers, we calculated the z-scores of volumes for all participants relative to the mean and standard deviation of the controls. We regressed out sex, age, and total brain volume with a multivariate linear regression, and we found significant differences in 37 ROIs between subjects with TBI and controls (p < 0.05 with independent t-tests with false discovery rate correction). We found that differences originated in 1) the brainstem, occipital pole and structures posterior to the orbit, 2) subcortical gray matter and insular cortex, and 3) cerebral and cerebellar white matter using independent component analysis and clustering the component loadings of those with TBI.

  • Open Access

    Data-driven abdominal phenotypes of type 2 diabetes in lean, overweight, and obese cohorts from computed tomography

    by Lucas W. Remedios, Chloe Cho, Trent M. Schwartz, Dingjie Su, Gaurav Rudravaram, Chenyu Gao, Aravind R. Krishnan, Adam M. Saunders, Michael E. Kim, Shunxing Bao, et al.

    PurposeAlthough elevated body mass index (BMI) is a well-known risk factor for type 2 diabetes, the disease’s presence in some lean adults and absence in others with obesity suggests that more detailed measurements of body composition may uncover abdominal phenotypes of type 2 diabetes. With artificial intelligence (AI) and computed tomography (CT), we can now leverage robust image segmentation to extract detailed measurements of size, shape, and tissue composition from abdominal organs, abdominal muscle, and abdominal fat depots in 3D clinical imaging at scale. This creates an opportunity to empirically define body composition signatures linked to type 2 diabetes risk and protection using large-scale clinical data.ApproachWe studied imaging records of 1728 de-identified patients from Vanderbilt University Medical Center with BMI collected from the electronic health record. To uncover BMI-specific diabetic abdominal patterns from clinical CT, we applied our design four times: once on the full cohort (n=1728) and once on lean (n=497), overweight (n=611), and obese (n=620) subgroups separately. Briefly, our experimental design transforms abdominal scans into collections of explainable measurements, identifies which measurements most strongly predict type 2 diabetes and how they contribute to risk or protection, groups scans by shared model decision patterns, and links those decision patterns back to interpretable abdominal phenotypes in the original explainable measurement space of the abdomen using the following steps. (1) To capture abdominal composition: we represented each scan as a collection of 88 automatically extracted measurements of the size, shape, and fat content of abdominal structures using TotalSegmentator. (2) To learn key predictors: we trained a 10-fold cross-validated random forest classifier with SHapley Additive exPlanations (SHAP) analysis to rank features and estimate their risk-versus-protective effects for type 2 diabetes. (3) To validate individual effects: for the 20 highest-ranked features, we ran univariate logistic regressions to quantify their independent associations with type 2 diabetes. (4) To identify decision-making patterns: we embedded the top-20 SHAP profiles with uniform manifold approximation and projection and applied silhouette-guided K-means to cluster the random forest’s decision space. (5) To link decisions to abdominal phenotypes: we fit one-versus-rest classifiers on the original anatomical measurements from each decision cluster and applied a second SHAP analysis to explore whether the random forest’s logic had identified abdominal phenotypes.ResultsAcross the full, lean, overweight, and obese cohorts, the random forest classifier achieved a mean area under the receiver operating characteristic curve (AUC) of 0.72 to 0.74. SHAP highlighted shared type 2 diabetes signatures in each group—fatty skeletal muscle, older age, greater visceral and subcutaneous fat, and a smaller or fat-laden pancreas. Univariate logistic regression confirmed the direction of 14 to 18 of the top 20 predictors within each subgroup (p<0.05). Clustering the model’s decision space further revealed type 2 diabetes-enriched abdominal phenotypes within the lean, overweight, and obese subgroups.ConclusionsWe found similar abdominal signatures of type 2 diabetes across the separate lean, overweight, and obese groups, which suggests that the abdominal drivers of type 2 diabetes may be consistent across weight classes. Although our model had a modest AUC, the explainable components allowed for a clear interpretation of feature importance. In addition, in both lean and obese subgroups, the most important feature for identifying type 2 diabetes was fatty skeletal muscle.


  • Open Access

    Influence of early through late fusion on pancreas segmentation from imperfectly registered multimodal magnetic resonance imaging

    by Lucas W. Remedios, Han Liu, Samuel W. Remedios, Lianrui Zuo, Adam M. Saunders, Shunxing Bao, Yuankai Huo, Alvin C. Powers, John Virostko, Bennett A. Landman

    PurposeCombining different types of medical imaging data, through multimodal fusion, promises better segmentation of anatomical structures, such as the pancreas. Strategic implementation of multimodal fusion could improve our ability to study diseases such as diabetes. However, where to perform fusion in deep learning models is still an open question. It is unclear if there is a single best location to fuse information when analyzing pairs of imperfectly aligned images or if the optimal fusion location depends on the specific model being used. Two main challenges when using multiple imaging modalities to study the pancreas are (1) the pancreas and surrounding abdominal anatomy have a deformable structure, making it difficult to consistently align the images and (2) breathing by the individual during image collection further complicates the alignment between multimodal images. Even after using state-of-the-art deformable image registration techniques, specifically designed to align abdominal images, multimodal images of the abdomen are often not perfectly aligned. We examine how the choice of different fusion points, ranging from early in the image processing pipeline to later stages, impacts the segmentation of the pancreas on imperfectly registered multimodal magnetic resonance (MR) images.ApproachOur dataset consists of 353 pairs of T2-weighted (T2w) and T1-weighted (T1w) abdominal MR images from 163 subjects with accompanying pancreas segmentation labels drawn mainly based on the T2w images. Because the T2w images were acquired in an interleaved manner across two breath holds and the T1w images on one breath hold, there were three different breath holds impacting the alignment of each pair of images. We used deeds, a state-of-the-art deformable abdominal image registration method to align the image pairs. Then, we trained a collection of basic UNets with different fusion points, spanning from early to late layers in the model, to assess how early through late fusion influenced segmentation performance on imperfectly aligned images. To investigate whether performance differences on key fusion points are generalized to other architectures, we expanded our experiments to nnUNet.ResultsThe single-modality T2w baseline using a basic UNet model had a median Dice score of 0.766, whereas the same baseline on the nnUNet model achieved 0.824. For each fusion approach, we analyzed the differences in performance with Dice residuals, by subtracting the baseline score from the fusion score for each datapoint. For the basic UNet, the best fusion approach was from early/mid fusion and occurred in the middle of the encoder with a median Dice residual of +0.012 compared with the baseline. For the nnUNet, the best fusion approach was early fusion through naïve image concatenation before the model, with a median Dice residual of +0.004 compared with the baseline. After Bonferroni correction, the distributions of the Dice scores for these best fusion approaches were found to be statistically significant (p<0.05) via the paired Wilcoxon signed-rank test against the baseline.ConclusionsFusion in specific blocks can improve performance, but the best blocks for fusion are model-specific, and the gains are small. In imperfectly registered datasets, fusion is a nuanced problem, with the art of design remaining vital for uncovering potential insights. Future innovation is needed to better address fusion in cases of imperfect alignment of abdominal image pairs. The code associated with this project is available here https://github.com/MASILab/influence_of_fusion_on_pancreas_segmentation.