DeepComBat: A statistically motivated, hyperparameter-robust, deep learning approach to harmonization of neuroimaging data.

Neuroimaging data acquired using multiple scanners or protocols are increasingly available. However, such data exhibit technical artifacts across batches which introduce confounding and decrease reproducibility. This is especially true when multi-batch data are analyzed using complex downstream models which are more likely to pick up on and implicitly incorporate batch-related information. Previously proposed image harmonization methods have sought to remove these batch effects; however, batch effects remain detectable in the data after applying these methods. We present DeepComBat, a deep learning harmonization method based on a conditional variational autoencoder and the ComBat method. DeepComBat combines the strengths of statistical and deep learning methods in order to account for the multivariate relationships between features while simultaneously relaxing strong assumptions made by previous deep learning harmonization methods. As a result, DeepComBat can perform multivariate harmonization while preserving data structure and avoiding the introduction of synthetic artifacts. We apply this method to cortical thickness measurements from a cognitive-aging cohort and show DeepComBat qualitatively and quantitatively outperforms existing methods in removing batch effects while preserving biological heterogeneity. Additionally, DeepComBat provides a new perspective for statistically motivated deep learning harmonization methods.

ANTsX neuroimaging-derived structural phenotypes of UK Biobank.

UK Biobank is a large-scale epidemiological resource for investigating prospective correlations between various lifestyle, environmental, and genetic factors with health and disease progression. In addition to individual subject information obtained through surveys and physical examinations, a comprehensive neuroimaging battery consisting of multiple modalities provides imaging-derived phenotypes (IDPs) that can serve as biomarkers in neuroscience research. In this study, we augment the existing set of UK Biobank neuroimaging structural IDPs, obtained from well-established software libraries such as FSL and FreeSurfer, with related measurements acquired through the Advanced Normalization Tools Ecosystem. This includes previously established cortical and subcortical measurements defined, in part, based on the Desikan-Killiany-Tourville atlas. Also included are morphological measurements from two recent developments: medial temporal lobe parcellation of hippocampal and extra-hippocampal regions in addition to cerebellum parcellation and thickness based on the Schmahmann anatomical labeling. Through predictive modeling, we assess the clinical utility of these IDP measurements, individually and in combination, using commonly studied phenotypic correlates including age, fluid intelligence, numeric memory, and several other sociodemographic variables. The predictive accuracy of these IDP-based models, in terms of root-mean-squared-error or area-under-the-curve for continuous and categorical variables, respectively, provides comparative insights between software libraries as well as potential clinical interpretability. Results demonstrate varied performance between package-based IDP sets and their combination, emphasizing the need for careful consideration in their selection and utilization.

Pulmonary MRI with hyperpolarized xenon-129 demonstrates novel alterations in gas transfer across the air-blood barrier in asthma.

BACKGROUND: Individuals with asthma can vary widely in clinical presentation, severity, and pathobiology. Hyperpolarized xenon-129 (Xe129) MRI is a novel imaging method to provide 3-D mapping of both ventilation and gas exchange in the human lung. PURPOSE: To evaluate the functional changes in adults with asthma as compared to healthy controls using Xe129 MRI. METHODS: All subjects (20 controls and 20 asthmatics) underwent lung function measurements and Xe129 MRI on the same day. Outcome measures included the pulmonary ventilation defect and transfer of inspired Xe129 into two soluble compartments: tissue and blood. Ten asthmatics underwent Xe129 MRI before and after bronchodilator to test whether gas transfer measures change with bronchodilator effects. RESULTS: Initial analysis of the results revealed striking differences in gas transfer measures based on age, hence we compared outcomes in younger (n = 24, ≤ 35 years) versus older (n = 16, > 45 years) asthmatics and controls. The younger asthmatics exhibited significantly lower Xe129 gas uptake by lung tissue (Asthmatic: 0.98% ± 0.24%, Control: 1.17% ± 0.12%, P = 0.035), and higher Xe129 gas transfer from tissue to the blood (Asthmatic: 0.40 ± 0.10, Control: 0.31% ± 0.03%, P = 0.035) than the younger controls. No significant difference in Xe129 gas transfer was observed in the older group between asthmatics and controls (P > 0.05). No significant change in Xe129 transfer was observed before and after bronchodilator treatment. CONCLUSIONS: By using Xe129 MRI, we discovered heterogeneous alterations of gas transfer that have associations with age. This finding suggests a heretofore unrecognized physiological derangement in the gas/tissue/blood interface in young adults with asthma that deserves further study.

Neurological Effects of Repeated Blast Exposure in Special Operations Personnel.

Exposure to blast overpressure has been a pervasive feature of combat-related injuries. Studies exploring the neurological correlates of repeated low-level blast exposure in career "breachers" demonstrated higher levels of tumor necrosis factor alpha (TNFα) and interleukin (IL)-6 and decreases in IL-10 within brain-derived extracellular vesicles (BDEVs). The current pilot study was initiated in partnership with the U.S. Special Operations Command (USSOCOM) to explore whether neuroinflammation is seen within special operators with prior blast exposure. Data were analyzed from 18 service members (SMs), inclusive of 9 blast-exposed special operators with an extensive career history of repeated blast exposures and 9 controls matched by age and duration of service. Neuroinflammation was assessed utilizing positron emission tomography (PET) imaging with [18F]DPA-714. Serum was acquired to assess inflammatory biomarkers within whole serum and BDEVs. The Blast Exposure Threshold Survey (BETS) was acquired to determine blast history. Both self-report and neurocognitive measures were acquired to assess cognition. Similarity-driven Multi-view Linear Reconstruction (SiMLR) was used for joint analysis of acquired data. Analysis of BDEVs indicated significant positive associations with a generalized blast exposure value (GBEV) derived from the BETS. SiMLR-based analyses of neuroimaging demonstrated exposure-related relationships between GBEV, PET-neuroinflammation, cortical thickness, and volume loss within special operators. Affected brain networks included regions associated with memory retrieval and executive functioning, as well as visual and heteromodal processing. Post hoc assessments of cognitive measures failed to demonstrate significant associations with GBEV. This emerging evidence suggests neuroinflammation may be a key feature of the brain response to blast exposure over a career in operational personnel. The common thread of neuroinflammation observed in blast-exposed populations requires further study.

ANTsX neuroimaging-derived structural phenotypes of UK Biobank.

UK Biobank is a large-scale epidemiological resource for investigating prospective correlations between various lifestyle, environmental, and genetic factors with health and disease progression. In addition to individual subject information obtained through surveys and physical examinations, a comprehensive neuroimaging battery consisting of multiple modalities provides imaging-derived phenotypes (IDPs) that can serve as biomarkers in neuroscience research. In this study, we augment the existing set of UK Biobank neuroimaging structural IDPs, obtained from well-established software libraries such as FSL and FreeSurfer, with related measurements acquired through the Advanced Normalization Tools Ecosystem. This includes previously established cortical and subcortical measurements defined, in part, based on the Desikan-Killiany-Tourville atlas. Also included are morphological measurements from two recent developments: medial temporal lobe parcellation of hippocampal and extra-hippocampal regions in addition to cerebellum parcellation and thickness based on the Schmahmann anatomical labeling. Through predictive modeling, we assess the clinical utility of these IDP measurements, individually and in combination, using commonly studied phenotypic correlates including age, fluid intelligence, numeric memory, and several other sociodemographic variables. The predictive accuracy of these IDP-based models, in terms of root-mean-squared-error or area-under-the-curve for continuous and categorical variables, respectively, provides comparative insights between software libraries as well as potential clinical interpretability. Results demonstrate varied performance between package-based IDP sets and their combination, emphasizing the need for careful consideration in their selection and utilization.

Developmental Mouse Brain Common Coordinate Framework.

3D standard reference brains serve as key resources to understand the spatial organization of the brain and promote interoperability across different studies. However, unlike the adult mouse brain, the lack of standard 3D reference atlases for developing mouse brains has hindered advancement of our understanding of brain development. Here, we present a multimodal 3D developmental common coordinate framework (DevCCF) spanning mouse embryonic day (E) 11.5, E13.5, E15.5, E18.5, and postnatal day (P) 4, P14, and P56 with anatomical segmentations defined by a developmental ontology. At each age, the DevCCF features undistorted morphologically averaged atlas templates created from Magnetic Resonance Imaging and co-registered high-resolution templates from light sheet fluorescence microscopy. Expert-curated 3D anatomical segmentations at each age adhere to an updated prosomeric model and can be explored via an interactive 3D web-visualizer. As a use case, we employed the DevCCF to unveil the emergence of GABAergic neurons in embryonic brains. Moreover, we integrated the Allen CCFv3 into the P56 template with stereotaxic coordinates and mapped spatial transcriptome cell-type data with the developmental ontology. In summary, the DevCCF is an openly accessible resource that can be used for large-scale data integration to gain a comprehensive understanding of brain development.

ReMiND: Recovery of Missing Neuroimaging using Diffusion Models with Application to Alzheimer's Disease.

Objective: Missing data is a significant challenge in medical research. In longitudinal studies of Alzheimer's disease (AD) where structural magnetic resonance imaging (MRI) is collected from individuals at multiple time points, participants may miss a study visit or drop out. Additionally, technical issues such as participant motion in the scanner may result in unusable imaging data at designated visits. Such missing data may hinder the development of high-quality imaging-based biomarkers. Furthermore, when imaging data are unavailable in clinical practice, patients may not benefit from effective application of biomarkers for disease diagnosis and monitoring. Methods: To address the problem of missing MRI data in studies of AD, we introduced a novel 3D diffusion model specifically designed for imputing missing structural MRI (Recovery of Missing Neuroimaging using Diffusion models (ReMiND)). The model generates a whole-brain image conditional on a single structural MRI observed at a past visit or conditional on one past and one future observed structural MRI relative to the missing observation. Results: Experimental results show that our method can generate high-quality individual 3D structural MRI with high similarity to ground truth, observed images. Additionally, images generated using ReMiND exhibit relatively lower error rates and more accurately estimated rates of atrophy over time in important anatomical brain regions compared with two alternative imputation approaches: forward filling and image generation using variational autoencoders. Conclusion: Our 3D diffusion model can impute missing structural MRI data at a single designated visit and outperforms alternative methods for imputing whole-brain images that are missing from longitudinal trajectories.

Lung Volume Dependence and Repeatability of Hyperpolarized 129Xe MRI Gas Uptake Metrics in Healthy Volunteers and Participants with COPD.

Purpose: To assess the effect of lung volume on measured values and repeatability of xenon 129 (129Xe) gas uptake metrics in healthy volunteers and participants with chronic obstructive pulmonary disease (COPD). Materials and Methods: This Health Insurance Portability and Accountability Act-compliant prospective study included data (March 2014-December 2015) from 49 participants (19 with COPD [mean age, 67 years ± 9 (SD)]; nine women]; 25 older healthy volunteers [mean age, 59 years ± 10; 20 women]; and five young healthy women [mean age, 23 years ± 3]). Thirty-two participants underwent repeated 129Xe and same-breath-hold proton MRI at residual volume plus one-third forced vital capacity (RV+FVC/3), with 29 also undergoing one examination at total lung capacity (TLC). The remaining 17 participants underwent imaging at TLC, RV+FVC/3, and residual volume (RV). Signal ratios between membrane, red blood cell (RBC), and gas-phase compartments were calculated using hierarchical iterative decomposition of water and fat with echo asymmetry and least-squares estimation (ie, IDEAL). Repeatability was assessed using coefficient of variation and intraclass correlation coefficient, and volume relationships were assessed using Spearman correlation and Wilcoxon rank sum tests. Results: Gas uptake metrics were repeatable at RV+FVC/3 (intraclass correlation coefficient = 0.88 for membrane/gas; 0.71 for RBC/gas, and 0.88 for RBC/membrane). Relative ratio changes were highly correlated with relative volume changes for membrane/gas (r = -0.97) and RBC/gas (r = -0.93). Membrane/gas and RBC/gas measured at RV+FVC/3 were significantly lower in the COPD group than the corresponding healthy group (P ≤ .001). However, these differences lessened upon correction for individual volume differences (P = .23 for membrane/gas; P = .09 for RBC/gas). Conclusion: Dissolved-phase 129Xe MRI-derived gas uptake metrics were repeatable but highly dependent on lung volume during measurement.Keywords: Blood-Air Barrier, MRI, Chronic Obstructive Pulmonary Disease, Pulmonary Gas Exchange, Xenon Supplemental material is available for this article © RSNA, 2023.

An Anatomical Template for the Normalization of Medical Images of Adult Human Hands.

During medical image analysis, it is often useful to align (or 'normalize') a given image of a given body part to a representative standard (or 'template') of that body part. The impact that brain templates have had on the analysis of brain images highlights the importance of templates in general. However, templates for human hands do not exist. Image normalization is especially important for hand images because hands, by design, readily change shape during various tasks. Here we report the construction of an anatomical template for healthy adult human hands. To do this, we used 27 anatomically representative T1-weighted magnetic resonance (MR) images of either hand from 21 demographically representative healthy adult subjects (13 females and 8 males). We used the open-source, cross-platform ANTs (Advanced Normalization Tools) medical image analysis software framework, to preprocess the MR images. The template was constructed using the ANTs standard multivariate template construction workflow. The resulting template image preserved all the essential anatomical features of the hand, including all the individual bones, muscles, tendons, ligaments, as well as the main branches of the median nerve and radial, ulnar, and palmar metacarpal arteries. Furthermore, the image quality of the template was significantly higher than that of the underlying individual hand images as measured by two independent canonical metrics of image quality.

Hyperpolarized Xenon-129: A New Tool to Assess Pulmonary Physiology in Patients with Pulmonary Fibrosis.

PURPOSE: The existing tools to quantify lung function in interstitial lung diseases have significant limitations. Lung MRI imaging using inhaled hyperpolarized xenon-129 gas (129Xe) as a contrast agent is a new technology for measuring regional lung physiology. We sought to assess the utility of the 129Xe MRI in detecting impaired lung physiology in usual interstitial pneumonia (UIP). MATERIALS AND METHODS: After institutional review board approval and informed consent and in compliance with HIPAA regulations, we performed chest CT, pulmonary function tests (PFTs), and 129Xe MRI in 10 UIP subjects and 10 healthy controls. RESULTS: The 129Xe MRI detected highly heterogeneous abnormalities within individual UIP subjects as compared to controls. Subjects with UIP had markedly impaired ventilation (ventilation defect fraction: UIP: 30 ± 9%; healthy: 21 ± 9%; p = 0.026), a greater amount of 129Xe dissolved in the lung interstitium (tissue-to-gas ratio: UIP: 1.45 ± 0.35%; healthy: 1.10 ± 0.17%; p = 0.014), and impaired 129Xe diffusion into the blood (RBC-to-tissue ratio: UIP: 0.20 ± 0.06; healthy: 0.28 ± 0.05; p = 0.004). Most MRI variables had no correlation with the CT and PFT measurements. The elevated level of 129Xe dissolved in the lung interstitium, in particular, was detectable even in subjects with normal or mildly impaired PFTs, suggesting that this measurement may represent a new method for detecting early fibrosis. CONCLUSION: The hyperpolarized 129Xe MRI was highly sensitive to regional functional changes in subjects with UIP and may represent a new tool for understanding the pathophysiology, monitoring the progression, and assessing the effectiveness of treatment in UIP.

DeepComBat: A Statistically Motivated, Hyperparameter-Robust, Deep Learning Approach to Harmonization of Neuroimaging Data.

Neuroimaging data from multiple batches (i.e. acquisition sites, scanner manufacturer, datasets, etc.) are increasingly necessary to gain new insights into the human brain. However, multi-batch data, as well as extracted radiomic features, exhibit pronounced technical artifacts across batches. These batch effects introduce confounding into the data and can obscure biological effects of interest, decreasing the generalizability and reproducibility of findings. This is especially true when multi-batch data is used alongside complex downstream analysis models, such as machine learning methods. Image harmonization methods seeking to remove these batch effects are important for mitigating these issues; however, significant multivariate batch effects remain in the data following harmonization by current state-of-the-art statistical and deep learning methods. We present DeepCombat, a deep learning harmonization method based on a conditional variational autoencoder architecture and the ComBat harmonization model. DeepCombat learns and removes subject-level batch effects by accounting for the multivariate relationships between features. Additionally, DeepComBat relaxes a number of strong assumptions commonly made by previous deep learning harmonization methods and is empirically robust across a wide range of hyperparameter choices. We apply this method to neuroimaging data from a large cognitive-aging cohort and find that DeepCombat outperforms existing methods, as assessed by a battery of machine learning methods, in removing scanner effects from cortical thickness measurements while preserving biological heterogeneity. Additionally, DeepComBat provides a new perspective for statistically-motivated deep learning harmonization methods.

Differential effects of emotional valence on mnemonic performance with greater hippocampal maturity.

The hippocampal formation (HF) facilitates declarative memory, with subfields providing unique contributions to memory performance. Maturational differences across subfields facilitate a shift toward increased memory specificity, with peripuberty sitting at the inflection point. Peripuberty is also a sensitive period in the development of anxiety disorders. We believe HF development during puberty is critical to negative overgeneralization, a common feature of anxiety disorders. To investigate this claim, we examined the relationship between mnemonic generalization and a cross-sectional pubertal maturity index (PMI) derived from partial least squares correlation (PLSC) analyses of subfield volumes and structural connectivity from T1-weighted and diffusion-weighted scans, respectively. Participants aged 9-14 yr, from clinical and community sources, performed a recognition task with emotionally valent (positive, negative, and neutral) images. HF volumetric PMI was positively associated with generalization for negative images. Hippocampal-medial prefrontal cortex connectivity PMI evidenced a behavioral relationship similar to that of the HF volumetric approach. These findings reflect a novel developmentally related balance between generalization behavior supported by the hippocampus and its connections with other regions, with maturational differences in this balance potentially contributing to negative overgeneralization during peripuberty.

Multivariate Residualization in Medical Imaging Analysis.

Nuisance variables in medical imaging research are common, complicating association and prediction studies based on image data. Medical image data are typically high dimensional, often consisting of many highly correlated features. As a result, computationally efficient and robust methods to address nuisance variables are difficult to implement. By-region univariate residualization is commonly used to remove the influence of nuisance variables, as are various extensions. However, these methods neglect multivariate properties and may fail to fully remove influence related to the joint distribution of these regions. Some methods, such as functional regression and others, do consider multivariate properties when controlling for nuisance variables. However, the utility of these methods is limited for data with many image regions due to computational and model complexity. We develop a multivariate residualization method to estimate the association between the image and nuisance variable using a machine learning algorithm and then compute the orthogonal projection of each subject's image data onto this space. We illustrate this method's performance in a set of simulation studies and apply it to data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Automated analysis of low-field brain MRI in cerebral malaria.

A central challenge of medical imaging studies is to extract biomarkers that characterize disease pathology or outcomes. Modern automated approaches have found tremendous success in high-resolution, high-quality magnetic resonance images. These methods, however, may not translate to low-resolution images acquired on magnetic resonance imaging (MRI) scanners with lower magnetic field strength. In low-resource settings where low-field scanners are more common and there is a shortage of radiologists to manually interpret MRI scans, it is critical to develop automated methods that can augment or replace manual interpretation, while accommodating reduced image quality. We present a fully automated framework for translating radiological diagnostic criteria into image-based biomarkers, inspired by a project in which children with cerebral malaria (CM) were imaged using low-field 0.35 Tesla MRI. We integrate multiatlas label fusion, which leverages high-resolution images from another sample as prior spatial information, with parametric Gaussian hidden Markov models based on image intensities, to create a robust method for determining ventricular cerebrospinal fluid volume. We also propose normalized image intensity and texture measurements to determine the loss of gray-to-white matter tissue differentiation and sulcal effacement. These integrated biomarkers have excellent classification performance for determining severe brain swelling due to CM.

Posterior white matter hyperintensities are associated with reduced medial temporal lobe subregional integrity and long-term memory in older adults.

White matter hyperintensities are a marker of small vessel cerebrovascular disease that are strongly related to cognition in older adults. Similarly, medial temporal lobe atrophy is well-documented in aging and Alzheimer's disease and is associated with memory decline. Here, we assessed the relationship between lobar white matter hyperintensities, medial temporal lobe subregional volumes, and hippocampal memory in older adults. We collected MRI scans in a sample of 139 older adults without dementia (88 females, mean age (SD) = 76.95 (10.61)). Participants were administered the Rey Auditory Verbal Learning Test (RAVLT). Regression analyses tested for associations among medial temporal lobe subregional volumes, regional white matter hyperintensities and memory, while adjusting for age, sex, and education and correcting for multiple comparisons. Increased occipital white matter hyperintensities were related to worse RAVLT delayed recall performance, and to reduced CA1, dentate gyrus, perirhinal cortex (Brodmann area 36), and parahippocampal cortex volumes. These medial temporal lobe subregional volumes were related to delayed recall performance. The association of occipital white matter hyperintensities with delayed recall performance was fully mediated statistically only by perirhinal cortex volume. These results suggest that white matter hyperintensities may be associated with memory decline through their impact on medial temporal lobe atrophy. These findings provide new insights into the role of vascular pathologies in memory loss in older adults and suggest that future studies should further examine the neural mechanisms of these relationships in longitudinal samples.

Deformation-based morphometry identifies deep brain structures protected by ocrelizumab.

BACKGROUND: Despite advancements in treatments for multiple sclerosis, insidious disease progression remains an area of unmet medical need, for which atrophy-based biomarkers may help better characterize the progressive biology. METHODS: We developed and applied a method of longitudinal deformation-based morphometry to provide voxel-level assessments of brain volume changes and identified brain regions that were significantly impacted by disease-modifying therapy. RESULTS: Using brain MRI data from two identically designed pivotal trials of relapsing multiple sclerosis (total N = 1483), we identified multiple deep brain regions, including the thalamus and brainstem, where volume loss over time was reduced by ocrelizumab (p < 0.05), a humanized anti-CD20 + monoclonal antibody approved for the treatment of multiple sclerosis. Additionally, identified brainstem shrinkage, as well as brain ventricle expansion, was associated with a greater risk for confirmed disability progression (p < 0.05). CONCLUSIONS: The identification of deep brain structures has a strong implication for developing new biomarkers of brain atrophy reduction to advance drug development for multiple sclerosis, which has an increasing focus on targeting the progressive biology.

Negative overgeneralization is associated with anxiety and mechanisms of pattern completion in peripubertal youth.

This study examines neural mechanisms of negative overgeneralization, the increased likelihood of generalizing negative information, in peri-puberty. Theories suggest that weak pattern separation [overlapping representations are made distinct, indexed by dentate gyrus/ cornu ammonis (CA)3 hippocampal subfield activation] underlies negative overgeneralization. We alternatively propose that neuro-maturational changes that favor pattern completion (cues reinstate stored representations, indexed by CA1 activation) are modulated by circuitry involved in emotional responding [amygdala, medial prefrontal cortices (mPFC)] to drive negative overgeneralization. Youth (n = 34, 9-14 years) recruited from community and clinic settings participated in an emotional mnemonic similarity task while undergoing magnetic resonance imaging. At study, participants indicated the valence of images; at test, participants made recognition memory judgments. Critical lure stimuli, which were similar to images at study, were presented at test, and errors ('false alarms') to negative relative to neutral stimuli reflected negative overgeneralization. Negative overgeneralization was related to greater and more similar patterns of activation in CA1 and both dorsal mPFC (dmPFC)and ventral mPFC (vmPFC) for negative relative to neutral stimuli. At study, amygdala exhibited greater functional coupling with CA1 and dmPFC during negative items that were later generalized. Negative overgeneralization is rooted in amygdala and mPFC modulation at encoding and pattern completion at retrieval.

Mitigating site effects in covariance for machine learning in neuroimaging data.

To acquire larger samples for answering complex questions in neuroscience, researchers have increasingly turned to multi-site neuroimaging studies. However, these studies are hindered by differences in images acquired across multiple sites. These effects have been shown to bias comparison between sites, mask biologically meaningful associations, and even introduce spurious associations. To address this, the field has focused on harmonizing data by removing site-related effects in the mean and variance of measurements. Contemporaneously with the increase in popularity of multi-center imaging, the use of machine learning (ML) in neuroimaging has also become commonplace. These approaches have been shown to provide improved sensitivity, specificity, and power due to their modeling the joint relationship across measurements in the brain. In this work, we demonstrate that methods for removing site effects in mean and variance may not be sufficient for ML. This stems from the fact that such methods fail to address how correlations between measurements can vary across sites. Data from the Alzheimer's Disease Neuroimaging Initiative is used to show that considerable differences in covariance exist across sites and that popular harmonization techniques do not address this issue. We then propose a novel harmonization method called Correcting Covariance Batch Effects (CovBat) that removes site effects in mean, variance, and covariance. We apply CovBat and show that within-site correlation matrices are successfully harmonized. Furthermore, we find that ML methods are unable to distinguish scanner manufacturer after our proposed harmonization is applied, and that the CovBat-harmonized data retain accurate prediction of disease group.

Image- versus histogram-based considerations in semantic segmentation of pulmonary hyperpolarized gas images.

PURPOSE: To characterize the differences between histogram-based and image-based algorithms for segmentation of hyperpolarized gas lung images. METHODS: Four previously published histogram-based segmentation algorithms (ie, linear binning, hierarchical k-means, fuzzy spatial c-means, and a Gaussian mixture model with a Markov random field prior) and an image-based convolutional neural network were used to segment 2 simulated data sets derived from a public (n = 29 subjects) and a retrospective collection (n = 51 subjects) of hyperpolarized 129Xe gas lung images transformed by common MRI artifacts (noise and nonlinear intensity distortion). The resulting ventilation-based segmentations were used to assess algorithmic performance and characterize optimization domain differences in terms of measurement bias and precision. RESULTS: Although facilitating computational processing and providing discriminating clinically relevant measures of interest, histogram-based segmentation methods discard important contextual spatial information and are consequently less robust in terms of measurement precision in the presence of common MRI artifacts relative to the image-based convolutional neural network. CONCLUSIONS: Direct optimization within the image domain using convolutional neural networks leverages spatial information, which mitigates problematic issues associated with histogram-based approaches and suggests a preferred future research direction. Further, the entire processing and evaluation framework, including the newly reported deep learning functionality, is available as open source through the well-known Advanced Normalization Tools ecosystem.