Image harmonization: A review of statistical and deep learning methods for removing batch effects and evaluation metrics for effective harmonization

Characterization of Portable Ultra-Low Field MRI Scanners for Multi-Center Structural Neuroimaging

The lower infrastructure requirements of portable ultra-low field MRI (ULF-MRI) systems have enabled their use in diverse settings such as intensive care units and remote medical facilities. The UNITY Project is an international neuroimaging network harnessing this technology, deploying portable ULF-MRI systems globally to expand access to MRI for studies into brain development. Given the wide range of environments where ULF-MRI systems may operate, there are external factors that might influence image quality. This work aims to introduce the quality control (QC) framework used by the UNITY Project to investigate how robust the systems are and how QC metrics compare between sites and over time. We present a QC framework using a commercially available phantom, scanned with 64 mT portable MRI systems at 17 sites across 12 countries on four continents. Using automated, open-source analysis tools, we quantify signal-to-noise, image contrast, and geometric distortions. Our results demonstrated that the image quality is robust to the varying operational environment, for example, electromagnetic noise interference and temperature. The Larmor frequency was significantly correlated to room temperature, as was image noise and contrast. Image distortions were less than 2.5 mm, with high robustness over time. Similar to studies at higher field, we found that changes in pulse sequence parameters from software updates had an impact on QC metrics. This study demonstrates that portable ULF-MRI systems can be deployed in a variety of environments for multi-center neuroimaging studies and produce robust results.

Cross-site Validation of AI Segmentation and Harmonization in Breast MRI

This work aims to perform a cross-site validation of automated segmentation for breast cancers in MRI and to compare the performance to radiologists. A three-dimensional (3D) U-Net was trained to segment cancers in dynamic contrast-enhanced axial MRIs using a large dataset from Site 1 (n = 15,266; 449 malignant and 14,817 benign). Performance was validated on site-specific test data from this and two additional sites, and common publicly available testing data. Four radiologists from each of the three clinical sites provided two-dimensional (2D) segmentations as ground truth. Segmentation performance did not differ between the network and radiologists on the test data from Sites 1 and 2 or the common public data (median Dice score Site 1, network 0.86 vs. radiologist 0.85, n = 114; Site 2, 0.91 vs. 0.91, n = 50; common: 0.93 vs. 0.90). For Site 3, an affine input layer was fine-tuned using segmentation labels, resulting in comparable performance between the network and radiologist (0.88 vs. 0.89, n = 42). Radiologist performance differed on the common test data, and the network numerically outperformed 11 of the 12 radiologists (median Dice: 0.85-0.94, n = 20). In conclusion, a deep network with a novel supervised harmonization technique matches radiologists' performance in MRI tumor segmentation across clinical sites. We make code and weights publicly available to promote reproducible AI in radiology.

Harmonizing CT scanner acquisition variability in an anthropomorphic phantom: A comparative study of image-level and feature-level harmonization using GAN, ComBat, and their combination

PURPOSE

Radiomics allows for the quantification of medical images and facilitates precision medicine. Many radiomic features derived from computed tomography (CT) are sensitive to variations across scanners, reconstruction settings, and acquisition protocols. In this phantom study, eight different CT reconstruction parameters were varied to explore image- and feature-level harmonization approaches to improve tissue classification.

METHODS

Varying reconstructions of an anthropomorphic radiopaque phantom containing three lesion categories (metastasis, hemangioma, and benign cyst) and normal liver tissue were used for evaluating two harmonization methods and their combination: (i) generative adversarial networks (GANs) at the image level; (ii) ComBat at the feature level, and (iii) a combination of (i) and (ii). A total of 93 texture and intensity features were extracted from each tissue class before and after image-level harmonization and were also harmonized at the feature level. Reproducibility and stability were assessed via the Concordance Correlation Coefficient (CCC) and pairwise comparisons using paired stability tests. The ability of features to discriminate between tissue classes was assessed by measuring the area under the receiver operating characteristic curve. The global reproducibility and discriminative power were assessed by averaging over the entire dataset and across all tissue types.

RESULTS

ComBat improved reproducibility by 31.58% and stability by 5.24%, while GAN increased reproducibility by 8% it reduced stability by 4.33%. Classification analysis revealed that ComBat increased average AUC by 15.19%, whereas GAN decreased AUC by 2.56%.

CONCLUSION

While GAN qualitatively enhances image harmonization, ComBat provides superior statistical improvements in feature stability and classification performance, highlighting the importance of robust feature-level harmonization in radiomics.

Avenues in the Analysis of Enlarged Perivascular Spaces Quantified from Brain Magnetic Resonance Images Acquired at 1.5T and 3T Magnetic Field Strengths

MR imaging-visible perivascular spaces (PVS) have been associated with disease phenotypes, risk factors, sleep measures, and overall brain health. We review avenues in the analysis of PVS quantified from brain MR imaging across dissimilar acquisition protocols, imaging modalities, scanner manufacturers and magnetic field strengths. We conduct a pilot analysis to evaluate different avenues to harmonise PVS assessments from using different parameters using brain MR imaging from 100 adult volunteers, acquired at two different magnetic field strengths with different sequence parameters. The 2024 MICCAI Enlarged Perivascular Spaces Segmentation Challenge provides a representative MRI dataset on which to test other harmonization methods.

Advancing Precision Oncology Through Modeling of Longitudinal and Multimodal Data

Cancer evolves continuously over time through a complex interplay of genetic, epigenetic, microenvironmental, and phenotypic changes. This dynamic behavior drives uncontrolled cell growth, metastasis, immune evasion, and therapy resistance, posing challenges for effective monitoring and treatment. However, today's data-driven research in oncology has primarily focused on cross-sectional analysis using data from a single modality, limiting the ability to fully characterize and interpret the disease's dynamic heterogeneity. Advances in multiscale data collection and computational methods now enable the discovery of longitudinal multimodal biomarkers for precision oncology. Longitudinal data reveal patterns of disease progression and treatment response that are not evident from single-timepoint data, enabling timely abnormality detection and dynamic treatment adaptation. Multimodal data integration offers complementary information from diverse sources for more precise risk assessment and targeting of cancer therapy. In this review, we survey methods of longitudinal and multimodal modeling, highlighting their synergy in providing multifaceted insights for personalized care tailored to the unique characteristics of a patient's cancer. We summarize the current challenges and future directions of longitudinal multimodal analysis in advancing precision oncology.

Flexible nonlinear modeling reveals age-related differences in resting-state functional brain connectivity in autistic males from childhood to mid-adulthood

BACKGROUND

Divergent age-related functional brain connectivity in autism spectrum disorder (ASD) has been observed using resting-state fMRI, although the specific findings are inconsistent across studies. Common statistical regression approaches that fit identical models across functional brain networks may contribute to these inconsistencies. Relationships among functional networks have been reported to follow unique nonlinear developmental trajectories, suggesting the need for flexible modeling. Here we apply generalized additive models (GAMs) to flexibly adapt to distinct network trajectories and simultaneously describe divergent age-related changes from childhood into mid-adulthood in ASD.

METHODS

1107 males, aged 5-40, from the ABIDE I & II cross-sectional datasets were analyzed. Functional connectivity was extracted using a network-based template. Connectivity values were harmonized using COMBAT-GAM. Connectivity-age relationships were assessed with thin-plate spline GAMs. Post-hoc analyses defined the age-ranges of divergent aging in ASD.

RESULTS

Typically developing (TD) and ASD groups shared 15 brain connections that significantly changed with age (FDR-corrected p < 0.05). Network connectivity exhibited diverse nonlinear age-related trajectories across the functional connectome. Comparing ASD and TD groups, default mode to central executive between-network connectivity followed similar nonlinear paths with no group differences. Contrarily, the ASD group had chronic hypoconnectivity throughout default mode-ventral attentional (salience) and default mode-somatomotor aging trajectories. Within-network somatomotor connectivity was similar between groups in childhood but diverged in adolescence with the ASD group showing decreased within-network connectivity. Network connectivity between the somatomotor network and various other functional networks had fully disrupted age-related pathways in ASD compared to TD, displaying significantly different model curvatures and fits.

LIMITATIONS

The present analysis includes only male participants and has a restricted age range, limiting analysis of early development and later life aging, years 40 and beyond. Additionally, our analysis is limited to large-scale network cortical functional parcellation. To parse more specificity of brain region connectivity, a fine-grained functional parcellation including subcortical areas may be warranted.

CONCLUSION

Flexible non-linear modeling minimizes statistical assumptions and allows diagnosis-related brain connections to follow independent data-driven age-related pathways. Using GAMs, we describe complex age-related pathways throughout the human connectome and observe distinct periods of divergence in autism.

The MR neuroimaging protocol for the Accelerating Medicines Partnership® Schizophrenia Program

Neuroimaging with MRI has been a frequent component of studies of individuals at clinical high risk (CHR) for developing psychosis, with goals of understanding potential brain regions and systems impacted in the CHR state and identifying prognostic or predictive biomarkers that can enhance our ability to forecast clinical outcomes. To date, most studies involving MRI in CHR are likely not sufficiently powered to generate robust and generalizable neuroimaging results. Here, we describe the prospective, advanced, and modern neuroimaging protocol that was implemented in a complex multi-site, multi-vendor environment, as part of the large-scale Accelerating Medicines Partnership® Schizophrenia Program (AMP® SCZ), including the rationale for various choices. This protocol includes T1- and T2-weighted structural scans, resting-state fMRI, and diffusion-weighted imaging collected at two time points, approximately 2 months apart. We also present preliminary variance component analyses of several measures, such as signal- and contrast-to-noise ratio (SNR/CNR) and spatial smoothness, to provide quantitative data on the relative percentages of participant, site, and platform (i.e., scanner model) variance. Site-related variance is generally small (typically <10%). For the SNR/CNR measures from the structural and fMRI scans, participant variance is the largest component (as desired; 40-76%). However, for SNR/CNR in the diffusion scans, there is substantial platform-related variance (>55%) due to differences in the diffusion imaging hardware capabilities of the different scanners. Also, spatial smoothness generally has a large platform-related variance due to inherent, difficult to control, differences between vendors in their acquisitions and reconstructions. These results illustrate some of the factors that will need to be considered in analyses of the AMP SCZ neuroimaging data, which will be the largest CHR cohort to date.Watch Dr. Harms discuss this article at https://vimeo.com/1059777228?share=copy#t=0 .

Site effects in multisite fetal brain MRI: morphological insights into early brain development

OBJECTIVE

To evaluate multisite effects on fetal brain MRI. Specifically, to identify crucial acquisition factors affecting fetal brain structural measurements and developmental patterns, while assessing the effectiveness of existing harmonization methods in mitigating site effects.

MATERIALS AND METHODS

Between May 2017 and March 2022, T2-weighted fast spin-echo sequences in-utero MRI were performed on healthy fetuses from retrospectively recruited pregnant volunteers on four different scanners at four sites. A generalized additive model (GAM) was used to quantitatively assess site effects, including field strength (FS), manufacturer (M), in-plane resolution (R), and slice thickness (ST), on subcortical volume and cortical morphological measurements, including cortical thickness, curvature, and sulcal depth. Growth models were selected to elucidate the developmental trajectories of these morphological measurements. Welch's test was performed to evaluate the influence of site effects on developmental trajectories. The comBat-GAM harmonization method was applied to mitigate site-related biases.

RESULTS

The final analytic sample consisted of 340 MRI scans from 218 fetuses (mean GA, 30.1 weeks ± 4.4 [range, 21.7-40 weeks]). GAM results showed that lower FS and lower spatial resolution led to overestimations in selected brain regions of subcortical volumes and cortical morphological measurements. Only the peak cortical thickness in developmental trajectories was significantly influenced by the effects of FS and R. Notably, ComBat-GAM harmonization effectively removed site effects while preserving developmental patterns.

CONCLUSION

Our findings pinpointed the key acquisition factors in in-utero fetal brain MRI and underscored the necessity of data harmonization when pooling multisite data for fetal brain morphology investigations.

KEY POINTS

Question How do specific site MRI acquisition factors affect fetal brain imaging? Finding Lower FS and spatial resolution overestimated subcortical volumes and cortical measurements. Cortical thickness in developmental trajectories was influenced by FS and in-plane resolution. Clinical relevance This study provides important guidelines for the fetal MRI community when scanning fetal brains and underscores the necessity of data harmonization of cross-center fetal studies.

Investigation of ComBat Harmonization on Radiomic and Deep Features from Multi-Center Abdominal MRI Data

ComBat harmonization has been developed to remove non-biological variations for data in multi-center research applying artificial intelligence (AI). We investigated the effectiveness of ComBat harmonization on radiomic and deep features extracted from large, multi-center abdominal MRI data. A retrospective study was conducted on T2-weighted (T2W) abdominal MRI data retrieved from individual patients with suspected or known chronic liver disease at three study sites. MRI data were acquired using systems from three manufacturers and two field strengths. Radiomic features and deep features were extracted using the PyRadiomics pipeline and a Swin Transformer. ComBat was used to harmonize radiomic and deep features across different manufacturers and field strengths. Student's t-test, ANOVA test, and Cohen's F score were applied to assess the difference in individual features before and after ComBat harmonization. Between two field strengths, 76.7%, 52.9%, and 26.7% of radiomic features, and 89.0%, 56.5%, and 0.1% of deep features from three manufacturers were significantly different. Among the three manufacturers, 90.1% and 75.0% of radiomic features and 89.3% and 84.1% of deep features from two field strengths were significantly different. After ComBat harmonization, there were no significant differences in radiomic and deep features among manufacturers or field strengths based on t-tests or ANOVA tests. Reduced Cohen's F scores were consistently observed after ComBat harmonization. ComBat harmonization effectively harmonizes radiomic and deep features by removing the non-biological variations due to system manufacturers and/or field strengths in large multi-center clinical abdominal MRI datasets.

Two Axes of White Matter Development

Despite decades of neuroimaging research, how white matter develops along the length of major tracts in humans remains unknown. Here, we identify fundamental patterns of white matter maturation by examining developmental variation along major, long-range cortico-cortical tracts in youth ages 5-23 years using diffusion MRI from three large-scale, cross-sectional datasets (total N = 2,710). Across datasets, we delineate two replicable axes of human white matter development. First, we find a deep-to-superficial axis, in which superficial tract regions near the cortical surface exhibit greater age-related change than deep tract regions. Second, we demonstrate that the development of superficial tract regions aligns with the cortical hierarchy defined by the sensorimotor-association axis, with tract ends adjacent to sensorimotor cortices maturing earlier than those adjacent to association cortices. These results reveal developmental variation along tracts that conventional tract-average analyses have previously obscured, challenging the implicit assumption that white matter tracts mature uniformly along their length. Such developmental variation along tracts may have functional implications, including mitigating ephaptic coupling in densely packed deep tract regions and tuning neural synchrony through hierarchical development in superficial tract regions - ultimately refining neural transmission in youth.

Performance of Lung Cancer Prediction Models for Screening-detected, Incidental, and Biopsied Pulmonary Nodules

Purpose To evaluate the performance of eight lung cancer prediction models on patient cohorts with screening-detected, incidentally detected, and bronchoscopically biopsied pulmonary nodules. Materials and Methods This study retrospectively evaluated promising predictive models for lung cancer prediction in three clinical settings: lung cancer screening with low-dose CT, incidentally detected pulmonary nodules, and nodules deemed suspicious enough to warrant a biopsy. The area under the receiver operating characteristic curve of eight validated models, including logistic regressions on clinical variables and radiologist nodule characterizations, artificial intelligence (AI) on chest CT scans, longitudinal imaging AI, and multimodal approaches for prediction of lung cancer risk was assessed in nine cohorts (n = 898, 896, 882, 219, 364, 117, 131, 115, 373) from multiple institutions. Each model was implemented from their published literature, and each cohort was curated from primary data sources collected over periods from 2002 to 2021. Results No single predictive model emerged as the highest-performing model across all cohorts, but certain models performed better in specific clinical contexts. Single-time-point chest CT AI performed well for screening-detected nodules but did not generalize well to other clinical settings. Longitudinal imaging and multimodal models demonstrated comparatively good performance on incidentally detected nodules. When applied to biopsied nodules, all models showed low performance. Conclusion Eight lung cancer prediction models failed to generalize well across clinical settings and sites outside of their training distributions. Keywords: Diagnosis, Classification, Application Domain, Lung Supplemental material is available for this article. © RSNA, 2025 See also commentary by Shao and Niu in this issue.

WMH Contributions to Cognitive Impairment: Rationale and Design of the Diverse VCID Study

As awareness of dementia increases, more individuals with minor cognitive complaints are requesting clinical assessment. Neuroimaging studies frequently identify incidental white matter hyperintensities, raising patient concerns about their brain health and future risk for dementia. Moreover, current US demographics indicate that ≈50% of these individuals will be from diverse backgrounds by 2060. Racial and ethnic minority populations bear a disproportionate burden of vascular risk factors magnifying dementia risk. Despite established associations between white matter hyperintensities and cognitive impairment, including dementia, no study has comprehensively and prospectively examined the impact of individual and combined magnetic resonance imaging measures of white matter injury, their risk factors, and comorbidities on cognitive performance among a diverse, nondemented, stroke-free population with cognitive complaints over an extended period of observation. The Diverse VCID (Diverse Vascular Cognitive Impairment and Dementia) study is designed to fill this knowledge gap through 3 assessments of clinical, behavioral, and risk factors; neurocognitive and magnetic resonance imaging measures; fluid biomarkers of Alzheimer disease, vascular inflammation, angiogenesis, and endothelial dysfunction; and measures of genetic risk collected prospectively over a minimum of 3 years in a cohort of 2250 individuals evenly distributed among Americans of Black/African, Latino/Hispanic, and non-Hispanic White backgrounds. The goal of this study is to investigate the basic mechanisms of small vessel cerebrovascular injury, emphasizing clinically relevant assessment tools and developing a risk score that will accurately identify at-risk individuals for possible treatment or clinical therapeutic trials, particularly individuals of diverse backgrounds where vascular risk factors and disease are more prevalent.

A comparative analysis of image harmonization techniques in mitigating differences in CT acquisition and reconstruction

Objective. The study aims to systematically characterize the effect of CT parameter variations on images and lung radiomic and deep features, and to evaluate the ability of different image harmonization methods to mitigate the observed variations.Approach. A retrospective in-house sinogram dataset of 100 low-dose chest CT scans was reconstructed by varying radiation dose (100%, 25%, 10%) and reconstruction kernels (smooth, medium, sharp). A set of image processing, convolutional neural network (CNNs), and generative adversarial network-based (GANs) methods were trained to harmonize all image conditions to a reference condition (100% dose, medium kernel). Harmonized scans were evaluated for image similarity using peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and learned perceptual image patch similarity (LPIPS), and for the reproducibility of radiomic and deep features using concordance correlation coefficient (CCC).Main Results. CNNs consistently yielded higher image similarity metrics amongst others; for Sharp/10%, which exhibited the poorest visual similarity, PSNR increased from a mean ± CI of 17.763 ± 0.492 to 31.925 ± 0.571, SSIM from 0.219 ± 0.009 to 0.754 ± 0.017, and LPIPS decreased from 0.490 ± 0.005 to 0.275 ± 0.016. Texture-based radiomic features exhibited a greater degree of variability across conditions, i.e. a CCC of 0.500 ± 0.332, compared to intensity-based features (0.972 ± 0.045). GANs achieved the highest CCC (0.969 ± 0.009 for radiomic and 0.841 ± 0.070 for deep features) amongst others. CNNs are suitable if downstream applications necessitate visual interpretation of images, whereas GANs are better alternatives for generating reproducible quantitative image features needed for machine learning applications.Significance. Understanding the efficacy of harmonization in addressing multi-parameter variability is crucial for optimizing diagnostic accuracy and a critical step toward building generalizable models suitable for clinical use.

Reproducible Brain Charts: An open data resource for mapping brain development and its associations with mental health

Major mental disorders are increasingly understood as disorders of brain development. Large and heterogeneous samples are required to define generalizable links between brain development and psychopathology. To this end, we introduce the Reproducible Brain Charts (RBC), an open data resource that integrates data from 5 large studies of brain development in youth from three continents (N=6,346; 45% Female). Confirmatory bifactor models were used to create harmonized psychiatric phenotypes that capture major dimensions of psychopathology. Following rigorous quality assurance, neuroimaging data were carefully curated and processed using consistent pipelines in a reproducible manner with DataLad, the Configurable Pipeline for the Analysis of Connectomes (C-PAC), and FreeSurfer. Initial analyses of RBC data emphasize the benefit of careful quality assurance and data harmonization in delineating developmental effects and associations with psychopathology. Critically, all RBC data - including harmonized psychiatric phenotypes, unprocessed images, and fully processed imaging derivatives - are openly shared without a data use agreement via the International Neuroimaging Data-sharing Initiative. Together, RBC facilitates large-scale, reproducible, and generalizable research in developmental and psychiatric neuroscience.

Grey-Matter Structure Markers of Alzheimer's Disease, Alzheimer's Conversion, Functioning and Cognition: A Meta-Analysis Across 11 Cohorts

Alzheimer's disease (AD) brain markers are needed to select people with early-stage AD for clinical trials and as quantitative endpoint measures in trials. Using 10 clinical cohorts (N = 9140) and the community volunteer UK Biobank (N = 37,664) we performed region of interest (ROI) and vertex-wise analyses of grey-matter structure (thickness, surface area and volume). We identified 94 trait-ROI significant associations, and 307 distinct cluster of vertex-associations, which partly overlap the ROI associations. For AD versus controls, smaller hippocampus, amygdala and of the medial temporal lobe (fusiform and parahippocampal gyri) was confirmed and the vertex-wise results provided unprecedented localisation of some of the associated region. We replicated AD associated differences in several subcortical (putamen, accumbens) and cortical regions (inferior parietal, postcentral, middle temporal, transverse temporal, inferior temporal, paracentral, superior frontal). These grey-matter regions and their relative effect sizes can help refine our understanding of the brain regions that may drive or precede the widespread brain atrophy observed in AD. An AD grey-matter score evaluated in independent cohorts was significantly associated with cognition, MCI status, AD conversion (progression from cognitively normal or MCI to AD), genetic risk, and tau concentration in individuals with none or mild cognitive impairments (AUC in 0.54-0.70, p-value < 5e-4). In addition, some of the grey-matter regions associated with cognitive impairment, progression to AD ('conversion'), and cognition/functional scores were also associated with AD, which sheds light on the grey-matter markers of disease stages, and their relationship with cognitive or functional impairment. Our multi-cohort approach provides robust and fine-grained maps the grey-matter structures associated with AD, symptoms, and progression, and calls for even larger initiatives to unveil the full complexity of grey-matter structure in AD.

Altered Neural Responses to Punishment Learning in Conduct Disorder

BACKGROUND

Conduct disorder (CD) is associated with deficits in the use of punishment for reinforcement learning (RL) and subsequent decision making, contributing to reckless, antisocial, and aggressive behaviors. Here, we used functional magnetic resonance imaging (fMRI) to examine whether differences in behavioral learning rates derived from computational modeling, particularly for punishment, are reflected in aberrant neural responses in youths with CD compared with typically developing control participants (TDCs).

METHODS

A total of 75 youths with CD and 99 TDCs (9-18 years, 47% girls) performed a probabilistic RL task with punishment, reward, and neutral contingencies. Using fMRI data in conjunction with computational modeling indices (learning rate α), we investigated group differences for the 3 learning conditions in whole-brain and region of interest (ROI) analyses, including the ventral striatum and insula.

RESULTS

Whole-brain analysis revealed typical neural responses for RL in both groups. However, linear regression models for the ROI analyses revealed that only the response pattern of the (anterior) insula during punishment learning was different in participants with CD compared with TDCs.

CONCLUSIONS

Youths with CD have atypical neural responses to learning from punishment (but not from reward), specifically in the insula. This suggests a selective dysfunction of RL mechanisms in CD that contributes to punishment insensitivity/hyposensitivity as a hallmark of the disorder. Because the (anterior) insula is involved in avoidance behaviors related to negative affect or arousal, insula dysfunction in CD may contribute to inappropriate behavioral decision making, which increases the risk for reckless, antisocial, and aggressive behaviors in affected youth.

Metastatic spine disease alters spinal load-to-strength ratios in patients compared to healthy individuals

Pathologic vertebral fractures (PVF) are common and serious complications in patients with metastatic lesions affecting the spine. Accurate assessment of cancer patients' PVF risk is an unmet clinical need. Load-to-strength ratios (LSRs) evaluated in vivo by estimating vertebral loading from biomechanical modeling and strength from computed tomography imaging (CT) have been associated with osteoporotic vertebral fractures in older adults. Here, for the first time, we investigate LSRs of thoracic and lumbar vertebrae of 135 spine metastases patients compared to LSRs of 246 healthy adults, comparable by age and sex, from the Framingham Heart Study under four loading tasks. Findings include: (1) Osteolytic vertebrae have higher LSRs than osteosclerotic and mixed vertebrae; (2). In patients' vertebrae without CT observed metastases, LSRs were greater than healthy controls. (3) LSRs depend on the spinal region (Thoracic, Thoracolumbar, Lumbar). These findings suggest that LSRs may contribute to identifying patients at risk of incident PVF in metastatic spine disease patients. The lesion-mediated difference suggests that risk thresholds should be established based on spinal region, simulated task, and metastatic lesion type.

Computational Methods for Data Integration and Imputation of Missing Values in Omics Datasets

Molecular profiling of different omic-modalities (e.g., DNA methylomics, transcriptomics, proteomics) in biological systems represents the basis for research and clinical decision-making. Measurement-specific biases, so-called batch effects, often hinder the integration of independently acquired datasets, and missing values further hamper the applicability of typical data processing algorithms. In addition to careful experimental design, well-defined standards in data acquisition and data exchange, the alleviation of these phenomena particularly requires a dedicated data integration and preprocessing pipeline. This review aims to give a comprehensive overview of computational methods for data integration and missing value imputation for omic data analyses. We provide formal definitions for missing value mechanisms and propose a novel statistical taxonomy for batch effects, especially in the presence of missing data. Based on an automated document search and systematic literature review, we describe 32 distinct data integration methods from five main methodological categories, as well as 37 algorithms for missing value imputation from five separate categories. Additionally, this review highlights multiple quantitative evaluation methods to aid researchers in selecting a suitable set of methods for their work. Finally, this work provides an integrated discussion of the relevance of batch effects and missing values in omics with corresponding method recommendations. We then propose a comprehensive three-step workflow from the study conception to final data analysis and deduce perspectives for future research. Eventually, we present a comprehensive flow chart as well as exemplary decision trees to aid practitioners in the selection of specific approaches for imputation and data integration in their studies.

DeepResBat: Deep residual batch harmonization accounting for covariate distribution differences

Pooling MRI data from multiple datasets requires harmonization to reduce undesired inter-site variabilities, while preserving effects of biological variables (or covariates). The popular harmonization approach ComBat uses a mixed effect regression framework that explicitly accounts for covariate distribution differences across datasets. There is also significant interest in developing harmonization approaches based on deep neural networks (DNNs), such as conditional variational autoencoder (cVAE). However, current DNN approaches do not explicitly account for covariate distribution differences across datasets. Here, we provide mathematical results, suggesting that not accounting for covariates can lead to suboptimal harmonization. We propose two DNN-based covariate-aware harmonization approaches: covariate VAE (coVAE) and DeepResBat. The coVAE approach is a natural extension of cVAE by concatenating covariates and site information with site- and covariate-invariant latent representations. DeepResBat adopts a residual framework inspired by ComBat. DeepResBat first removes the effects of covariates with nonlinear regression trees, followed by eliminating site differences with cVAE. Finally, covariate effects are added back to the harmonized residuals. Using three datasets from three continents with a total of 2787 participants and 10,085 anatomical T1 scans, we find that DeepResBat and coVAE outperformed ComBat, CovBat and cVAE in terms of removing dataset differences, while enhancing biological effects of interest. However, coVAE hallucinates spurious associations between anatomical MRI and covariates even when no association exists. Future studies proposing DNN-based harmonization approaches should be aware of this false positive pitfall. Overall, our results suggest that DeepResBat is an effective deep learning alternative to ComBat. Code for DeepResBat can be found here: https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/harmonization/An2024_DeepResBat.

IGUANe: A 3D generalizable CycleGAN for multicenter harmonization of brain MR images

In MRI studies, the aggregation of imaging data from multiple acquisition sites enhances sample size but may introduce site-related variabilities that hinder consistency in subsequent analyses. Deep learning methods for image translation have emerged as a solution for harmonizing MR images across sites. In this study, we introduce IGUANe (Image Generation with Unified Adversarial Networks), an original 3D model that leverages the strengths of domain translation and straightforward application of style transfer methods for multicenter brain MR image harmonization. IGUANe extends CycleGAN by integrating an arbitrary number of domains for training through a many-to-one architecture. The framework based on domain pairs enables the implementation of sampling strategies that prevent confusion between site-related and biological variabilities. During inference, the model can be applied to any image, even from an unknown acquisition site, making it a universal generator for harmonization. Trained on a dataset comprising T1-weighted images from 11 different scanners, IGUANe was evaluated on data from unseen sites. The assessments included the transformation of MR images with traveling subjects, the preservation of pairwise distances between MR images within domains, the evolution of volumetric patterns related to age and Alzheimer's disease (AD), and the performance in age regression and patient classification tasks. Comparisons with other harmonization and normalization methods suggest that IGUANe better preserves individual information in MR images and is more suitable for maintaining and reinforcing variabilities related to age and AD. Future studies may further assess IGUANe in other multicenter contexts, either using the same model or retraining it for applications to different image modalities. Codes and the trained IGUANe model are available at https://github.com/RocaVincent/iguane_harmonization.git.

Standardization and clinical applications of retinal imaging biomarkers for cardiovascular disease: a Roadmap from an NHLBI workshop

The accessibility of the retina with the use of non-invasive and relatively low-cost ophthalmic imaging techniques and analytics provides a unique opportunity to improve the detection, diagnosis and monitoring of systemic diseases. The National Heart, Lung, and Blood Institute conducted a workshop in October 2022 to examine this concept. On the basis of the discussions at that workshop, this Roadmap describes current knowledge gaps and new research opportunities to evaluate the relationships between the eye (in particular, retinal biomarkers) and the risk of cardiovascular diseases, including coronary artery disease, heart failure, stroke, hypertension and vascular dementia. Identified gaps include the need to simplify and standardize the capture of high-quality images of the eye by non-ophthalmic health workers and to conduct longitudinal studies using multidisciplinary networks of diverse at-risk populations with improved implementation and methods to protect participant and dataset privacy. Other gaps include improving the measurement of structural and functional retinal biomarkers, determining the relationship between microvascular and macrovascular risk factors, improving multimodal imaging 'pipelines', and integrating advanced imaging with 'omics', lifestyle factors, primary care data and radiological reports, by using artificial intelligence technology to improve the identification of individual-level risk. Future research on retinal microvascular disease and retinal biomarkers might additionally provide insights into the temporal development of microvascular disease across other systemic vascular beds.

Electroconvulsive Therapy Regulates Brain Connectome Dynamics in Patients With Major Depressive Disorder

BACKGROUND

Electroconvulsive therapy (ECT) is an effective treatment for patients with major depressive disorder (MDD), but its underlying neural mechanisms remain largely unknown. The aim of this study was to identify changes in brain connectome dynamics after ECT in MDD and to explore their associations with treatment outcome.

METHODS

We collected longitudinal resting-state functional magnetic resonance imaging data from 80 patients with MDD (50 with suicidal ideation [MDD-SI] and 30 without [MDD-NSI]) before and after ECT and 37 age- and sex-matched healthy control participants. A multilayer network model was used to assess modular switching over time in functional connectomes. Support vector regression was used to assess whether pre-ECT network dynamics could predict treatment response in terms of symptom severity.

RESULTS

At baseline, patients with MDD had lower global modularity and higher modular variability in functional connectomes than control participants. Network modularity increased and network variability decreased after ECT in patients with MDD, predominantly in the default mode and somatomotor networks. Moreover, ECT was associated with decreased modular variability in the left dorsal anterior cingulate cortex of MDD-SI but not MDD-NSI patients, and pre-ECT modular variability significantly predicted symptom improvement in the MDD-SI group but not in the MDD-NSI group.

CONCLUSIONS

We highlight ECT-induced changes in MDD brain network dynamics and their predictive value for treatment outcome, particularly in patients with SI. This study advances our understanding of the neural mechanisms of ECT from a dynamic brain network perspective and suggests potential prognostic biomarkers for predicting ECT efficacy in patients with MDD.

Generative models of MRI-derived neuroimaging features and associated dataset of 18,000 samples

Availability of large and diverse medical datasets is often challenged by privacy and data sharing restrictions. Successful application of machine learning techniques for disease diagnosis, prognosis, and precision medicine, requires large amounts of data for model building and optimization. To help overcome such limitations in the context of brain MRI, we present GenMIND: a collection of generative models of normative regional volumetric features derived from structural brain imaging. GenMIND models are trained on real brain imaging regional volumetric measures from the iSTAGING consortium, which encompasses over 40,000 MRI scans across 13 studies, incorporating covariates such as age, sex, and race. Leveraging GenMIND, we produce and offer 18,000 synthetic samples spanning the adult lifespan (ages 22-90 years), alongside the model's capability to generate unlimited data. Experimental results indicate that samples generated from GenMIND align well with the distributions observed in real data. Most importantly, the generated normative data significantly enhances the accuracy of downstream machine learning models on tasks such as disease classification. Dataset and the generative models are publicly available.

Deep feature batch correction using ComBat for machine learning applications in computational pathology

Background

Developing artificial intelligence (AI) models for digital pathology requires large datasets from multiple sources. However, without careful implementation, AI models risk learning confounding site-specific features in datasets instead of clinically relevant information, leading to overestimated performance, poor generalizability to real-world data, and potential misdiagnosis.

Methods

Whole-slide images (WSIs) from The Cancer Genome Atlas (TCGA) colon (COAD), and stomach adenocarcinoma datasets were selected for inclusion in this study. Patch embeddings were obtained using three feature extraction models, followed by ComBat harmonization. Attention-based multiple instance learning models were trained to predict tissue-source site (TSS), as well as clinical and genetic attributes, using raw, Macenko normalized, and Combat-harmonized patch embeddings.

Results

TSS prediction achieved high accuracy (AUROC > 0.95) with all three feature extraction models. ComBat harmonization significantly reduced the AUROC for TSS prediction, with mean AUROCs dropping to approximately 0.5 for most models, indicating successful mitigation of batch effects (e.g., CCL-ResNet50 in TCGA-COAD: Pre-ComBat AUROC = 0.960, Post-ComBat AUROC = 0.506, p < 0.001). Clinical attributes associated with TSS, such as race and treatment response, showed decreased predictability post-harmonization. Notably, the prediction of genetic features like MSI status remained robust after harmonization (e.g., MSI in TCGA-COAD: Pre-ComBat AUROC = 0.667, Post-ComBat AUROC = 0.669, p=0.952), indicating the preservation of true histological signals.

Conclusion

ComBat harmonization of deep learning-derived histology features effectively reduces the risk of AI models learning confounding features in WSIs, ensuring more reliable performance estimates. This approach is promising for the integration of large-scale digital pathology datasets.

MICCAI-CDMRI 2023 QuantConn Challenge Findings on Achieving Robust Quantitative Connectivity through Harmonized Preprocessing of Diffusion MRI

White matter alterations are increasingly implicated in neurological diseases and their progression. International-scale studies use diffusion-weighted magnetic resonance imaging (DW-MRI) to qualitatively identify changes in white matter microstructure and connectivity. Yet, quantitative analysis of DW-MRI data is hindered by inconsistencies stemming from varying acquisition protocols. Specifically, there is a pressing need to harmonize the preprocessing of DW-MRI datasets to ensure the derivation of robust quantitative diffusion metrics across acquisitions. In the MICCAI-CDMRI 2023 QuantConn challenge, participants were provided raw data from the same individuals collected on the same scanner but with two different acquisitions and tasked with preprocessing the DW-MRI to minimize acquisition differences while retaining biological variation. Harmonized submissions are evaluated on the reproducibility and comparability of cross-acquisition bundle-wise microstructure measures, bundle shape features, and connectomics. The key innovations of the QuantConn challenge are that (1) we assess bundles and tractography in the context of harmonization for the first time, (2) we assess connectomics in the context of harmonization for the first time, and (3) we have 10x additional subjects over prior harmonization challenge, MUSHAC and 100x over SuperMUDI. We find that bundle surface area, fractional anisotropy, connectome assortativity, betweenness centrality, edge count, modularity, nodal strength, and participation coefficient measures are most biased by acquisition and that machine learning voxel-wise correction, RISH mapping, and NeSH methods effectively reduce these biases. In addition, microstructure measures AD, MD, RD, bundle length, connectome density, efficiency, and path length are least biased by these acquisition differences. A machine learning approach that learned voxel-wise cross-acquisition relationships was the most effective at harmonizing connectomic, microstructure, and macrostructure features, but requires the same subject be scanned at each site co-registered. NeSH, a spatial and angular resampling method, was also effective and has generalizable framework not reliant co-registration. Our code is available at https://github.com/nancynewlin-masi/QuantConn/.

A medical image classification method based on self-regularized adversarial learning

BACKGROUND

Deep learning (DL) techniques have been extensively applied in medical image classification. The unique characteristics of medical imaging data present challenges, including small labeled datasets, severely imbalanced class distribution, and significant variations in imaging quality. Recently, generative adversarial network (GAN)-based classification methods have gained attention for their ability to enhance classification accuracy by incorporating realistic GAN-generated images as data augmentation. However, the performance of these GAN-based methods often relies on high-quality generated images, while large amounts of training data are required to train GAN models to achieve optimal performance.

PURPOSE

In this study, we propose an adversarial learning-based classification framework to achieve better classification performance. Innovatively, GAN models are employed as supplementary regularization terms to support classification, aiming to address the challenges described above.

METHODS

The proposed classification framework, GAN-DL, consists of a feature extraction network (F-Net), a classifier, and two adversarial networks, specifically a reconstruction network (R-Net) and a discriminator network (D-Net). The F-Net extracts features from input images, and the classifier uses these features for classification tasks. R-Net and D-Net have been designed following the GAN architecture. R-Net employs the extracted feature to reconstruct the original images, while D-Net is tasked with the discrimination between the reconstructed image and the original images. An iterative adversarial learning strategy is designed to guide model training by incorporating multiple network-specific loss functions. These loss functions, serving as supplementary regularization, are automatically derived during the reconstruction process and require no additional data annotation.

RESULTS

To verify the model's effectiveness, we performed experiments on two datasets, including a COVID-19 dataset with 13 958 chest x-ray images and an oropharyngeal squamous cell carcinoma (OPSCC) dataset with 3255 positron emission tomography images. Thirteen classic DL-based classification methods were implemented on the same datasets for comparison. Performance metrics included precision, sensitivity, specificity, andF 1 $F_1$ -score. In addition, we conducted ablation studies to assess the effects of various factors on model performance, including the network depth of F-Net, training image size, training dataset size, and loss function design. Our method achieved superior performance than all comparative methods. On the COVID-19 dataset, our method achieved95.4 % ± 0.6 % $95.4\%\pm 0.6\%$ ,95.3 % ± 0.9 % $95.3\%\pm 0.9\%$ ,97.7 % ± 0.4 % $97.7\%\pm 0.4\%$ , and95.3 % ± 0.9 % $95.3\%\pm 0.9\%$ in terms of precision, sensitivity, specificity, andF 1 $F_1$ -score, respectively. It achieved96.2 % ± 0.7 % $96.2\%\pm 0.7\%$ across all these metrics on the OPSCC dataset. The study to investigate the effects of two adversarial networks highlights the crucial role of D-Net in improving model performance. Ablation studies further provide an in-depth understanding of our methodology.

CONCLUSION

Our adversarial-based classification framework leverages GAN-based adversarial networks and an iterative adversarial learning strategy to harness supplementary regularization during training. This design significantly enhances classification accuracy and mitigates overfitting issues in medical image datasets. Moreover, its modular design not only demonstrates flexibility but also indicates its potential applicability to various clinical contexts and medical imaging applications.

Current best practices and future opportunities for reproducible findings using large-scale neuroimaging in psychiatry

Research into the brain basis of psychopathology is challenging due to the heterogeneity of psychiatric disorders, extensive comorbidities, underdiagnosis or overdiagnosis, multifaceted interactions with genetics and life experiences, and the highly multivariate nature of neural correlates. Therefore, increasingly larger datasets that measure more variables in larger cohorts are needed to gain insights. In this review, we present current "best practice" approaches for using existing databases, collecting and sharing new repositories for big data analyses, and future directions for big data in neuroimaging and psychiatry with an emphasis on contributing to collaborative efforts and the challenges of multi-study data analysis.

ComBat models for harmonization of resting-state EEG features in multisite studies

OBJECTIVE

Pooling multisite resting-state electroencephalography (rsEEG) datasets may introduce bias due to batch effects (i.e., cross-site differences in the rsEEG related to scanner/sample characteristics). The Combining Batches (ComBat) models, introduced for microarray expression and adapted for neuroimaging, can control for batch effects while preserving the variability of biological covariates. We aim to evaluate four ComBat harmonization methods in a pooled sample from five independent rsEEG datasets of young and old adults.

METHODS

RsEEG signals (n = 374) were automatically preprocessed. Oscillatory and aperiodic rsEEG features were extracted in sensor space. Features were harmonized using neuroCombat (standard ComBat used in neuroimaging), neuroHarmonize (variant with nonlinear adjustment of covariates), OPNested-GMM (variant based on Gaussian Mixture Models to fit bimodal feature distributions), and HarmonizR (variant based on resampling to handle missing feature values). Relationships between rsEEG features and age were explored before and after harmonizing batch effects.

RESULTS

Batch effects were identified in rsEEG features. All ComBat methods reduced batch effects and features' dispersion; HarmonizR and OPNested-GMM ComBat achieved the greatest performance. Harmonized Beta power, individual Alpha peak frequency, Aperiodic exponent, and offset in posterior electrodes showed significant relations with age. All ComBat models maintained the direction of observed relationships while increasing the effect size.

CONCLUSIONS

ComBat models, particularly HarmonizeR and OPNested-GMM ComBat, effectively control for batch effects in rsEEG spectral features.

SIGNIFICANCE

This workflow can be used in multisite studies to harmonize batch effects in sensor-space rsEEG spectral features while preserving biological associations.

Generative models of MRI-derived neuroimaging features and associated dataset of 18,000 samples

Availability of large and diverse medical datasets is often challenged by privacy and data sharing restrictions. For successful application of machine learning techniques for disease diagnosis, prognosis, and precision medicine, large amounts of data are necessary for model building and optimization. To help overcome such limitations in the context of brain MRI, we present GenMIND: a collection of generative models of normative regional volumetric features derived from structural brain imaging. GenMIND models are trained on real brain imaging regional volumetric measures from the iSTAGING consortium, which encompasses over 40,000 MRI scans across 13 studies, incorporating covariates such as age, sex, and race. Leveraging GenMIND, we produce and offer 18,000 synthetic samples spanning the adult lifespan (ages 22-90 years), alongside the model's capability to generate unlimited data. Experimental results indicate that samples generated from GenMIND agree with the distributions obtained from real data. Most importantly, the generated normative data significantly enhance the accuracy of downstream machine learning models on tasks such as disease classification. Data and models are available at: https://huggingface.co/spaces/rongguangw/GenMIND.

Cone beam computed tomography (CBCT) and megavoltage computed tomography (MVCT)-based radiomics in head and neck cancers: a systematic review and radiomics quality score assessment

Background

Cone beam computed tomography (CBCT) and megavoltage computed tomography (MVCT)-based images demonstrate measurable radiomics features that are potentially prognostic. This study aims to systematically synthesize the current research applying radiomics in head and neck cancers for outcome prediction and to assess the radiomics quality score (RQS) of the studies.

Methods

A systematic search was performed to identify available studies on PubMed, Web of Science, and Scopus databases. Studies related to radiomics in oncology/radiotherapy fields and based on predefined Patient, Intervention, Comparator, Outcome, and Study design (PICOS) criteria were included. The methodological quality of the included study was evaluated independently by two reviewers according to the RQS. The Mann-Whitney U test was performed according to subgroups. The P values <0.05 were considered statistically significant. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 reporting guidelines were adhered to.

Results

From a total of 743 identified studies, six original studies were eligible for inclusion in the systematic review (median =97 patients). The intraclass correlation coefficient (ICC) for inter-reviewer on total RQS was excellent with 0.99 [95% confidence interval (CI) of 0.946< ICC <0.999]. There were no significant differences in the analyses between each RQS domain and subgroup components (P always >0.05). Numerically higher RQS domains score for publication year ≤2022 than 2023 and number of patients > median than ≤ median but not statistically significant.

Conclusions

The number of radiomics studies involving CBCT and MVCT is still very limited. Self-reported RQS assessments should be encouraged for all radiomics studies.

Neuroimage analysis using artificial intelligence approaches: a systematic review

In the contemporary era, artificial intelligence (AI) has undergone a transformative evolution, exerting a profound influence on neuroimaging data analysis. This development has significantly elevated our comprehension of intricate brain functions. This study investigates the ramifications of employing AI techniques on neuroimaging data, with a specific objective to improve diagnostic capabilities and contribute to the overall progress of the field. A systematic search was conducted in prominent scientific databases, including PubMed, IEEE Xplore, and Scopus, meticulously curating 456 relevant articles on AI-driven neuroimaging analysis spanning from 2013 to 2023. To maintain rigor and credibility, stringent inclusion criteria, quality assessments, and precise data extraction protocols were consistently enforced throughout this review. Following a rigorous selection process, 104 studies were selected for review, focusing on diverse neuroimaging modalities with an emphasis on mental and neurological disorders. Among these, 19.2% addressed mental illness, and 80.7% focused on neurological disorders. It is found that the prevailing clinical tasks are disease classification (58.7%) and lesion segmentation (28.9%), whereas image reconstruction constituted 7.3%, and image regression and prediction tasks represented 9.6%. AI-driven neuroimaging analysis holds tremendous potential, transforming both research and clinical applications. Machine learning and deep learning algorithms outperform traditional methods, reshaping the field significantly.

HarmonyTM: multi-center data harmonization applied to distributed learning for Parkinson's disease classification

Purpose

Distributed learning is widely used to comply with data-sharing regulations and access diverse datasets for training machine learning (ML) models. The traveling model (TM) is a distributed learning approach that sequentially trains with data from one center at a time, which is especially advantageous when dealing with limited local datasets. However, a critical concern emerges when centers utilize different scanners for data acquisition, which could potentially lead models to exploit these differences as shortcuts. Although data harmonization can mitigate this issue, current methods typically rely on large or paired datasets, which can be impractical to obtain in distributed setups.

Approach

We introduced HarmonyTM, a data harmonization method tailored for the TM. HarmonyTM effectively mitigates bias in the model's feature representation while retaining crucial disease-related information, all without requiring extensive datasets. Specifically, we employed adversarial training to "unlearn" bias from the features used in the model for classifying Parkinson's disease (PD). We evaluated HarmonyTM using multi-center three-dimensional (3D) neuroimaging datasets from 83 centers using 23 different scanners.

Results

Our results show that HarmonyTM improved PD classification accuracy from 72% to 76% and reduced (unwanted) scanner classification accuracy from 53% to 30% in the TM setup.

Conclusion

HarmonyTM is a method tailored for harmonizing 3D neuroimaging data within the TM approach, aiming to minimize shortcut learning in distributed setups. This prevents the disease classifier from leveraging scanner-specific details to classify patients with or without PD-a key aspect for deploying ML models for clinical applications.

Deep learning for the harmonization of structural MRI scans: a survey

Medical imaging datasets for research are frequently collected from multiple imaging centers using different scanners, protocols, and settings. These variations affect data consistency and compatibility across different sources. Image harmonization is a critical step to mitigate the effects of factors like inherent differences between various vendors, hardware upgrades, protocol changes, and scanner calibration drift, as well as to ensure consistent data for medical image processing techniques. Given the critical importance and widespread relevance of this issue, a vast array of image harmonization methodologies have emerged, with deep learning-based approaches driving substantial advancements in recent times. The goal of this review paper is to examine the latest deep learning techniques employed for image harmonization by analyzing cutting-edge architectural approaches in the field of medical image harmonization, evaluating both their strengths and limitations. This paper begins by providing a comprehensive fundamental overview of image harmonization strategies, covering three critical aspects: established imaging datasets, commonly used evaluation metrics, and characteristics of different scanners. Subsequently, this paper analyzes recent structural MRI (Magnetic Resonance Imaging) harmonization techniques based on network architecture, network learning algorithm, network supervision strategy, and network output. The underlying architectures include U-Net, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), flow-based generative models, transformer-based approaches, as well as custom-designed network architectures. This paper investigates the effectiveness of Disentangled Representation Learning (DRL) as a pivotal learning algorithm in harmonization. Lastly, the review highlights the primary limitations in harmonization techniques, specifically the lack of comprehensive quantitative comparisons across different methods. The overall aim of this review is to serve as a guide for researchers and practitioners to select appropriate architectures based on their specific conditions and requirements. It also aims to foster discussions around ongoing challenges in the field and shed light on promising future research directions with the potential for significant advancements.

DeepResBat: deep residual batch harmonization accounting for covariate distribution differences

Pooling MRI data from multiple datasets requires harmonization to reduce undesired inter-site variabilities, while preserving effects of biological variables (or covariates). The popular harmonization approach ComBat uses a mixed effect regression framework that explicitly accounts for covariate distribution differences across datasets. There is also significant interest in developing harmonization approaches based on deep neural networks (DNNs), such as conditional variational autoencoder (cVAE). However, current DNN approaches do not explicitly account for covariate distribution differences across datasets. Here, we provide mathematical results, suggesting that not accounting for covariates can lead to suboptimal harmonization. We propose two DNN-based covariate-aware harmonization approaches: covariate VAE (coVAE) and DeepResBat. The coVAE approach is a natural extension of cVAE by concatenating covariates and site information with site- and covariate-invariant latent representations. DeepResBat adopts a residual framework inspired by ComBat. DeepResBat first removes the effects of covariates with nonlinear regression trees, followed by eliminating site differences with cVAE. Finally, covariate effects are added back to the harmonized residuals. Using three datasets from three continents with a total of 2787 participants and 10085 anatomical T1 scans, we find that DeepResBat and coVAE outperformed ComBat, CovBat and cVAE in terms of removing dataset differences, while enhancing biological effects of interest. However, coVAE hallucinates spurious associations between anatomical MRI and covariates even when no association exists. Future studies proposing DNN-based harmonization approaches should be aware of this false positive pitfall. Overall, our results suggest that DeepResBat is an effective deep learning alternative to ComBat. Code for DeepResBat can be found here: https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/harmonization/An2024_DeepResBat.

Deep Learning-Driven Transformation: A Novel Approach for Mitigating Batch Effects in Diffusion MRI Beyond Traditional Harmonization

BACKGROUND

"Batch effect" in MR images, due to vendor-specific features, MR machine generations, and imaging parameters, challenges image quality and hinders deep learning (DL) model generalizability.

PURPOSE

We aim to develop a DL model using contrast adjustment and super-resolution to reduce diffusion-weighted images (DWIs) diversity across magnetic field strengths and imaging parameters.

STUDY TYPE

Retrospective.

SUBJECTS

The DL model was built using an open dataset from one individual. The MR machine identification model was trained and validated on a dataset of 1134 adults (54% females, 46% males), with 1050 subjects showing no DWI abnormalities and 84 with conditions like stroke and tumors. The 21,000 images were divided into 80% for training, 20% for validation, and 3500 for testing.

FIELD STRENGTH/SEQUENCE

Seven MR scanners from four manufacturers with 1.5 T and 3 T magnetic field strengths. DWIs were acquired using spin-echo sequences and high-resolution T2WIs using the T2-SPACE sequence.

ASSESSMENT

An experienced, board-certified radiologist evaluated the effectiveness of restoring high-resolution T2WI and harmonizing diverse DWI with metrics such as PSNR and SSIM, and the texture and frequency attributes were further analyzed using gray-level co-occurrence matrix and 1-dimensional power spectral density. The model's impact on machine-specific characteristics was gauged through the performance metrics of a ResNet-50 model. Comprehensive statistical tests were employed for statistical robustness, including McNemar's test and the Dice index.

RESULTS

Our DL protocol reduced DWI contrast and resolution variation. ResNet-50 model's accuracy decreased from 0.9443 to 0.5786, precision from 0.9442 to 0.6494, recall from 0.9443 to 0.5786, and F1 score from 0.9438 to 0.5587. The t-SNE visualization indicated more consistent image features across multiple MR devices. Autoencoder halved learning iterations; Dice coefficient >0.74 confirmed signal reproducibility in 84 lesions.

CONCLUSION

This study presents a DL strategy to mitigate batch effects in diffusion MR images, improving their quality and generalizability.

EVIDENCE LEVEL

3 TECHNICAL EFFICACY: Stage 1.

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.

ComBatLS: A location- and scale-preserving method for multi-site image harmonization

Recent work has leveraged massive datasets and advanced harmonization methods to construct normative models of neuroanatomical features and benchmark individuals' morphology. However, current harmonization tools do not preserve the effects of biological covariates including sex and age on features' variances; this failure may induce error in normative scores, particularly when such factors are distributed unequally across sites. Here, we introduce a new extension of the popular ComBat harmonization method, ComBatLS, that preserves biological variance in features' locations and scales. We use UK Biobank data to show that ComBatLS robustly replicates individuals' normative scores better than other ComBat methods when subjects are assigned to sex-imbalanced synthetic "sites". Additionally, we demonstrate that ComBatLS significantly reduces sex biases in normative scores compared to traditional methods. Finally, we show that ComBatLS successfully harmonizes consortium data collected across over 50 studies. R implementation of ComBatLS is available at https://github.com/andy1764/ComBatFamily.

Multivariate testing and effect size measures for batch effect evaluation in radiomic features

While precision medicine applications of radiomics analysis are promising, differences in image acquisition can cause "batch effects" that reduce reproducibility and affect downstream predictive analyses. Harmonization methods such as ComBat have been developed to correct these effects, but evaluation methods for quantifying batch effects are inconsistent. In this study, we propose the use of the multivariate statistical test PERMANOVA and the Robust Effect Size Index (RESI) to better quantify and characterize batch effects in radiomics data. We evaluate these methods in both simulated and real radiomics features extracted from full-field digital mammography (FFDM) data. PERMANOVA demonstrated higher power than standard univariate statistical testing, and RESI was able to interpretably quantify the effect size of site at extremely large sample sizes. These methods show promise as more powerful and interpretable methods for the detection and quantification of batch effects in radiomics studies.

Functional connectome through the human life span

The lifespan growth of the functional connectome remains unknown. Here, we assemble task-free functional and structural magnetic resonance imaging data from 33,250 individuals aged 32 postmenstrual weeks to 80 years from 132 global sites. We report critical inflection points in the nonlinear growth curves of the global mean and variance of the connectome, peaking in the late fourth and late third decades of life, respectively. After constructing a fine-grained, lifespan-wide suite of system-level brain atlases, we show distinct maturation timelines for functional segregation within different systems. Lifespan growth of regional connectivity is organized along a primary-to-association cortical axis. These connectome-based normative models reveal substantial individual heterogeneities in functional brain networks in patients with autism spectrum disorder, major depressive disorder, and Alzheimer's disease. These findings elucidate the lifespan evolution of the functional connectome and can serve as a normative reference for quantifying individual variation in development, aging, and neuropsychiatric disorders.

SAN: Mitigating spatial covariance heterogeneity in cortical thickness data collected from multiple scanners or sites

In neuroimaging studies, combining data collected from multiple study sites or scanners is becoming common to increase the reproducibility of scientific discoveries. At the same time, unwanted variations arise by using different scanners (inter-scanner biases), which need to be corrected before downstream analyses to facilitate replicable research and prevent spurious findings. While statistical harmonization methods such as ComBat have become popular in mitigating inter-scanner biases in neuroimaging, recent methodological advances have shown that harmonizing heterogeneous covariances results in higher data quality. In vertex-level cortical thickness data, heterogeneity in spatial autocorrelation is a critical factor that affects covariance heterogeneity. Our work proposes a new statistical harmonization method called spatial autocorrelation normalization (SAN) that preserves homogeneous covariance vertex-level cortical thickness data across different scanners. We use an explicit Gaussian process to characterize scanner-invariant and scanner-specific variations to reconstruct spatially homogeneous data across scanners. SAN is computationally feasible, and it easily allows the integration of existing harmonization methods. We demonstrate the utility of the proposed method using cortical thickness data from the Social Processes Initiative in the Neurobiology of the Schizophrenia(s) (SPINS) study. SAN is publicly available as an R package.

Functional connectivity development along the sensorimotor-association axis enhances the cortical hierarchy

Human cortical maturation has been posited to be organized along the sensorimotor-association axis, a hierarchical axis of brain organization that spans from unimodal sensorimotor cortices to transmodal association cortices. Here, we investigate the hypothesis that the development of functional connectivity during childhood through adolescence conforms to the cortical hierarchy defined by the sensorimotor-association axis. We tested this pre-registered hypothesis in four large-scale, independent datasets (total n = 3355; ages 5-23 years): the Philadelphia Neurodevelopmental Cohort (n = 1207), Nathan Kline Institute-Rockland Sample (n = 397), Human Connectome Project: Development (n = 625), and Healthy Brain Network (n = 1126). Across datasets, the development of functional connectivity systematically varied along the sensorimotor-association axis. Connectivity in sensorimotor regions increased, whereas connectivity in association cortices declined, refining and reinforcing the cortical hierarchy. These consistent and generalizable results establish that the sensorimotor-association axis of cortical organization encodes the dominant pattern of functional connectivity development.

Maximum Classifier Discrepancy Generative Adversarial Network for Jointly Harmonizing Scanner Effects and Improving Reproducibility of Downstream Tasks

OBJECTIVE

Multi-site collaboration is essential for overcoming small-sample problems when exploring reproducible biomarkers in MRI studies. However, various scanner-specific factors dramatically reduce the cross-scanner replicability. Moreover, existing harmony methods mostly could not guarantee the improved performance of downstream tasks.

METHODS

we proposed a new multi-scanner harmony framework, called 'maximum classifier discrepancy generative adversarial network', or MCD-GAN, for removing scanner effects in the original feature space while preserving substantial biological information for downstream tasks. Specifically, the adversarial generative network was utilized for persisting the structural layout of each sample, and the maximum classifier discrepancy module was introduced for regulating GAN generators by incorporating the downstream tasks.

RESULTS

We compared the MCD-GAN with other state-of-the-art data harmony approaches (e.g., ComBat, CycleGAN) on simulated data and the Adolescent Brain Cognitive Development (ABCD) dataset. Results demonstrate that MCD-GAN outperformed other approaches in improving cross-scanner classification performance while preserving the anatomical layout of the original images.

SIGNIFICANCE

To the best of our knowledge, the proposed MCD-GAN is the first generative model which incorporates downstream tasks while harmonizing, and is a promising solution for facilitating cross-site reproducibility in various tasks such as classification and regression.

Feasibility of deep learning-reconstructed thin-slice single-breath-hold HASTE for detecting pancreatic lesions: A comparison with two conventional T2-weighted imaging sequences

Objective

The objective of this study was to evaluate the clinical feasibility of deep learning reconstruction-accelerated thin-slice single-breath-hold half-Fourier single-shot turbo spin echo imaging (HASTEDL) for detecting pancreatic lesions, in comparison with two conventional T2-weighted imaging sequences: compressed-sensing HASTE (HASTECS) and BLADE.

Methods

From March 2022 to January 2023, a total of 63 patients with suspected pancreatic-related disease underwent the HASTEDL, HASTECS, and BLADE sequences were enrolled in this retrospectively study. The acquisition time, the pancreatic lesion conspicuity (LCP), respiratory motion artifact (RMA), main pancreatic duct conspicuity (MPDC), overall image quality (OIQ), signal-to-noise ratio (SNR), and contrast-noise-ratio (CNR) of the pancreatic lesions were compared among the three sequences by two readers.

Results

The acquisition time of both HASTEDL and HASTECS was 16 s, which was significantly shorter than that of 102 s for BLADE. In terms of qualitative parameters, Reader 1 and Reader 2 assigned significantly higher scores to the LCP, RMA, MPDC, and OIQ for HASTEDL compared to HASTECS and BLADE sequences; As for the quantitative parameters, the SNR values of the pancreatic head, body, tail, and lesions, the CNR of the pancreatic lesion measured by the two readers were also significantly higher for HASTEDL than for HASTECS and BLADE sequences.

Conclusions

Compared to conventional T2WI sequences (HASTECS and BLADE), deep-learning reconstructed HASTE enables thin slice and single-breath-hold acquisition with clinical acceptable image quality for detection of pancreatic lesions.

Deep conditional generative model for longitudinal single-slice abdominal computed tomography harmonization

Purpose

Two-dimensional single-slice abdominal computed tomography (CT) provides a detailed tissue map with high resolution allowing quantitative characterization of relationships between health conditions and aging. However, longitudinal analysis of body composition changes using these scans is difficult due to positional variation between slices acquired in different years, which leads to different organs/tissues being captured.

Approach

To address this issue, we propose C-SliceGen, which takes an arbitrary axial slice in the abdominal region as a condition and generates a pre-defined vertebral level slice by estimating structural changes in the latent space.

Results

Our experiments on 2608 volumetric CT data from two in-house datasets and 50 subjects from the 2015 Multi-Atlas Abdomen Labeling Challenge Beyond the Cranial Vault (BTCV) dataset demonstrate that our model can generate high-quality images that are realistic and similar. We further evaluate our method's capability to harmonize longitudinal positional variation on 1033 subjects from the Baltimore longitudinal study of aging dataset, which contains longitudinal single abdominal slices, and confirmed that our method can harmonize the slice positional variance in terms of visceral fat area.

Conclusion

This approach provides a promising direction for mapping slices from different vertebral levels to a target slice and reducing positional variance for single-slice longitudinal analysis. The source code is available at: https://github.com/MASILab/C-SliceGen.

Empirical assessment of the assumptions of ComBat with diffusion tensor imaging

Purpose

Diffusion tensor imaging (DTI) is a magnetic resonance imaging technique that provides unique information about white matter microstructure in the brain but is susceptible to confounding effects introduced by scanner or acquisition differences. ComBat is a leading approach for addressing these site biases. However, despite its frequent use for harmonization, ComBat's robustness toward site dissimilarities and overall cohort size have not yet been evaluated in terms of DTI.

Approach

As a baseline, we match N = 358 participants from two sites to create a "silver standard" that simulates a cohort for multi-site harmonization. Across sites, we harmonize mean fractional anisotropy and mean diffusivity, calculated using participant DTI data, for the regions of interest defined by the JHU EVE-Type III atlas. We bootstrap 10 iterations at 19 levels of total sample size, 10 levels of sample size imbalance between sites, and 6 levels of mean age difference between sites to quantify (i) β AGE , the linear regression coefficient of the relationship between FA and age; (ii) γ ^ s f * , the ComBat-estimated site-shift; and (iii) δ ^ s f * , the ComBat-estimated site-scaling. We characterize the reliability of ComBat by evaluating the root mean squared error in these three metrics and examine if there is a correlation between the reliability of ComBat and a violation of assumptions.

Results

ComBat remains well behaved for β AGE when N > 162 and when the mean age difference is less than 4 years. The assumptions of the ComBat model regarding the normality of residual distributions are not violated as the model becomes unstable.

Conclusion

Prior to harmonization of DTI data with ComBat, the input cohort should be examined for size and covariate distributions of each site. Direct assessment of residual distributions is less informative on stability than bootstrap analysis. We caution use ComBat of in situations that do not conform to the above thresholds.

Toward individualized connectomes of brain morphology

The morphological brain connectome (MBC) delineates the coordinated patterns of local morphological features (such as cortical thickness) across brain regions. While classically constructed using population-based approaches, there is a growing trend toward individualized modeling. Currently, the methods for individualized MBCs are varied, posing challenges for method selection and cross-study comparisons. Here, we summarize how individualized MBCs are modeled through low-order methods (correlation-, divergence-, distance-, and deviation-based methods) describing relations in brain morphology, as well as high-order methods capturing similarities in these low-order relations. We discuss the merits and limitations of different methods, examining them in the context of robustness, reproducibility, and reliability. We highlight the importance of elucidating the cellular and molecular mechanisms underlying the individualized connectome, and establishing normative benchmarks to assess individual variation in development, aging, and neuropsychiatric disorders.

Merging or ensembling: integrative analysis in multiple neuroimaging studies

The aim of this paper is to systematically investigate merging and ensembling methods for spatially varying coefficient mixed effects models (SVCMEM) in order to carry out integrative learning of neuroimaging data obtained from multiple biomedical studies. The "merged" approach involves training a single learning model using a comprehensive dataset that encompasses information from all the studies. Conversely, the "ensemble" approach involves creating a weighted average of distinct learning models, each developed from an individual study. We systematically investigate the prediction accuracy of the merged and ensemble learners under the presence of different degrees of interstudy heterogeneity. Additionally, we establish asymptotic guidelines for making strategic decisions about when to employ either of these models in different scenarios, along with deriving optimal weights for the ensemble learner. To validate our theoretical results, we perform extensive simulation studies. The proposed methodology is also applied to 3 large-scale neuroimaging studies.

Deep learning based joint fusion approach to exploit anatomical and functional brain information in autism spectrum disorders

BACKGROUND

The integration of the information encoded in multiparametric MRI images can enhance the performance of machine-learning classifiers. In this study, we investigate whether the combination of structural and functional MRI might improve the performances of a deep learning (DL) model trained to discriminate subjects with Autism Spectrum Disorders (ASD) with respect to typically developing controls (TD).

MATERIAL AND METHODS

We analyzed both structural and functional MRI brain scans publicly available within the ABIDE I and II data collections. We considered 1383 male subjects with age between 5 and 40 years, including 680 subjects with ASD and 703 TD from 35 different acquisition sites. We extracted morphometric and functional brain features from MRI scans with the Freesurfer and the CPAC analysis packages, respectively. Then, due to the multisite nature of the dataset, we implemented a data harmonization protocol. The ASD vs. TD classification was carried out with a multiple-input DL model, consisting in a neural network which generates a fixed-length feature representation of the data of each modality (FR-NN), and a Dense Neural Network for classification (C-NN). Specifically, we implemented a joint fusion approach to multiple source data integration. The main advantage of the latter is that the loss is propagated back to the FR-NN during the training, thus creating informative feature representations for each data modality. Then, a C-NN, with a number of layers and neurons per layer to be optimized during the model training, performs the ASD-TD discrimination. The performance was evaluated by computing the Area under the Receiver Operating Characteristic curve within a nested 10-fold cross-validation. The brain features that drive the DL classification were identified by the SHAP explainability framework.

RESULTS

The AUC values of 0.66±0.05 and of 0.76±0.04 were obtained in the ASD vs. TD discrimination when only structural or functional features are considered, respectively. The joint fusion approach led to an AUC of 0.78±0.04. The set of structural and functional connectivity features identified as the most important for the two-class discrimination supports the idea that brain changes tend to occur in individuals with ASD in regions belonging to the Default Mode Network and to the Social Brain.

CONCLUSIONS

Our results demonstrate that the multimodal joint fusion approach outperforms the classification results obtained with data acquired by a single MRI modality as it efficiently exploits the complementarity of structural and functional brain information.

Exercise-Related Physical Activity Relates to Brain Volumes in 10,125 Individuals

BACKGROUND

The potential neuroprotective effects of regular physical activity on brain structure are unclear, despite links between activity and reduced dementia risk.

OBJECTIVE

To investigate the relationships between regular moderate to vigorous physical activity and quantified brain volumes on magnetic resonance neuroimaging.

METHODS

A total of 10,125 healthy participants underwent whole-body MRI scans, with brain sequences including isotropic MP-RAGE. Three deep learning models analyzed axial, sagittal, and coronal views from the scans. Moderate to vigorous physical activity, defined by activities increasing respiration and pulse rate for at least 10 continuous minutes, was modeled with brain volumes via partial correlations. Analyses adjusted for age, sex, and total intracranial volume, and a 5% Benjamini-Hochberg False Discovery Rate addressed multiple comparisons.

RESULTS

Participant average age was 52.98±13.04 years (range 18-97) and 52.3% were biologically male. Of these, 7,606 (75.1%) reported engaging in moderate or vigorous physical activity approximately 4.05±3.43 days per week. Those with vigorous activity were slightly younger (p < 0.00001), and fewer women compared to men engaged in such activities (p = 3.76e-15). Adjusting for age, sex, body mass index, and multiple comparisons, increased days of moderate to vigorous activity correlated with larger normalized brain volumes in multiple regions including: total gray matter (Partial R = 0.05, p = 1.22e-7), white matter (Partial R = 0.06, p = 9.34e-11), hippocampus (Partial R = 0.05, p = 5.96e-7), and frontal, parietal, and occipital lobes (Partial R = 0.04, p≤1.06e-5).

CONCLUSIONS

Exercise-related physical activity is associated with increased brain volumes, indicating potential neuroprotective effects.