Cross-scanner harmonization methods for structural MRI may need further work: A comparison study

Evaluation of Mean Shift, ComBat, and CycleGAN for Harmonizing Brain Connectivity Matrices Across Sites

Connectivity matrices derived from diffusion MRI (dMRI) provide an interpretable and generalizable way of understanding the human brain connectome. However, dMRI suffers from inter-site and between-scanner variation, which impedes analysis across datasets to improve robustness and reproducibility of results. To evaluate different harmonization approaches on connectivity matrices, we compared graph measures derived from these matrices before and after applying three harmonization techniques: mean shift, ComBat, and CycleGAN. The sample comprises 168 age-matched, sex-matched normal subjects from two studies: the Vanderbilt Memory and Aging Project (VMAP) and the Biomarkers of Cognitive Decline Among Normal Individuals (BIOCARD). First, we plotted the graph measures and used coefficient of variation (CoV) and the Mann-Whitney U test to evaluate different methods' effectiveness in removing site effects on the matrices and the derived graph measures. ComBat effectively eliminated site effects for global efficiency and modularity and outperformed the other two methods. However, all methods exhibited poor performance when harmonizing average betweenness centrality. Second, we tested whether our harmonization methods preserved correlations between age and graph measures. All methods except for CycleGAN in one direction improved correlations between age and global efficiency and between age and modularity from insignificant to significant with p-values less than 0.05.

A three-dimensional deep learning model for inter-site harmonization of structural MR images of the brain: Extensive validation with a multicenter dataset

In multicenter MRI studies, pooling the imaging data can introduce site-related variabilities and can therefore bias the subsequent analyses. To harmonize the intensity distributions of brain MR images in a multicenter dataset, unsupervised deep learning methods can be employed. Here, we developed a model based on cycle-consistent adversarial networks for the harmonization of T1-weighted brain MR images. In contrast to previous works, it was designed to process three-dimensional whole-brain images in a stable manner while optimizing computation resources. Using six different MRI datasets for healthy adults (n=1525 in total) with different acquisition parameters, we tested the model in (i) three pairwise harmonizations with site effects of various sizes, (ii) an overall harmonization of the six datasets with different age distributions, and (iii) a traveling-subject dataset. Our results for intensity distributions, brain volumes, image quality metrics and radiomic features indicated that the MRI characteristics at the various sites had been effectively homogenized. Next, brain age prediction experiments and the observed correlation between the gray-matter volume and age showed that thanks to an appropriate training strategy and despite biological differences between the dataset populations, the model reinforced biological patterns. Furthermore, radiologic analyses of the harmonized images attested to the conservation of the radiologic information in the original images. The robustness of the harmonization model (as judged with various datasets and metrics) demonstrates its potential for application in retrospective multicenter studies.

Effects of MRI scanner manufacturers in classification tasks with deep learning models

Deep learning has become a leading subset of machine learning and has been successfully employed in diverse areas, ranging from natural language processing to medical image analysis. In medical imaging, researchers have progressively turned towards multi-center neuroimaging studies to address complex questions in neuroscience, leveraging larger sample sizes and aiming to enhance the accuracy of deep learning models. However, variations in image pixel/voxel characteristics can arise between centers due to factors including differences in magnetic resonance imaging scanners. Such variations create challenges, particularly inconsistent performance in machine learning-based approaches, often referred to as domain shift, where the trained models fail to achieve satisfactory or improved results when confronted with dissimilar test data. This study analyzes the performance of multiple disease classification tasks using multi-center MRI data obtained from three widely used scanner manufacturers (GE, Philips, and Siemens) across several deep learning-based networks. Furthermore, we investigate the efficacy of mitigating scanner vendor effects using ComBat-based harmonization techniques when applied to multi-center datasets of 3D structural MR images. Our experimental results reveal a substantial decline in classification performance when models trained on one type of scanner manufacturer are tested with data from different manufacturers. Moreover, despite applying ComBat-based harmonization, the harmonized images do not demonstrate any noticeable performance enhancement for disease classification tasks.

HACA3: A unified approach for multi-site MR image harmonization

The lack of standardization and consistency of acquisition is a prominent issue in magnetic resonance (MR) imaging. This often causes undesired contrast variations in the acquired images due to differences in hardware and acquisition parameters. In recent years, image synthesis-based MR harmonization with disentanglement has been proposed to compensate for the undesired contrast variations. The general idea is to disentangle anatomy and contrast information from MR images to achieve cross-site harmonization. Despite the success of existing methods, we argue that major improvements can be made from three aspects. First, most existing methods are built upon the assumption that multi-contrast MR images of the same subject share the same anatomy. This assumption is questionable, since different MR contrasts are specialized to highlight different anatomical features. Second, these methods often require a fixed set of MR contrasts for training (e.g., both T1-weighted and T2-weighted images), limiting their applicability. Lastly, existing methods are generally sensitive to imaging artifacts. In this paper, we present Harmonization with Attention-based Contrast, Anatomy, and Artifact Awareness (HACA3), a novel approach to address these three issues. HACA3 incorporates an anatomy fusion module that accounts for the inherent anatomical differences between MR contrasts. Furthermore, HACA3 can be trained and applied to any combination of MR contrasts and is robust to imaging artifacts. HACA3 is developed and evaluated on diverse MR datasets acquired from 21 sites with varying field strengths, scanner platforms, and acquisition protocols. Experiments show that HACA3 achieves state-of-the-art harmonization performance under multiple image quality metrics. We also demonstrate the versatility and potential clinical impact of HACA3 on downstream tasks including white matter lesion segmentation for people with multiple sclerosis and longitudinal volumetric analyses for normal aging subjects. Code is available at https://github.com/lianruizuo/haca3.

DSMRI: Domain Shift Analyzer for Multi-Center MRI Datasets

In medical research and clinical applications, the utilization of MRI datasets from multiple centers has become increasingly prevalent. However, inherent variability between these centers presents challenges due to domain shift, which can impact the quality and reliability of the analysis. Regrettably, the absence of adequate tools for domain shift analysis hinders the development and validation of domain adaptation and harmonization techniques. To address this issue, this paper presents a novel Domain Shift analyzer for MRI (DSMRI) framework designed explicitly for domain shift analysis in multi-center MRI datasets. The proposed model assesses the degree of domain shift within an MRI dataset by leveraging various MRI-quality-related metrics derived from the spatial domain. DSMRI also incorporates features from the frequency domain to capture low- and high-frequency information about the image. It further includes the wavelet domain features by effectively measuring the sparsity and energy present in the wavelet coefficients. Furthermore, DSMRI introduces several texture features, thereby enhancing the robustness of the domain shift analysis process. The proposed framework includes visualization techniques such as t-SNE and UMAP to demonstrate that similar data are grouped closely while dissimilar data are in separate clusters. Additionally, quantitative analysis is used to measure the domain shift distance, domain classification accuracy, and the ranking of significant features. The effectiveness of the proposed approach is demonstrated using experimental evaluations on seven large-scale multi-site neuroimaging datasets.

Temporo-basal sulcal connections: a manual annotation protocol and an investigation of sexual dimorphism and heritability

The temporo-basal region of the human brain is composed of the collateral, the occipito-temporal, and the rhinal sulci. We manually rated (using a novel protocol) the connections between rhinal/collateral (RS-CS), collateral/occipito-temporal (CS-OTS) and rhinal/occipito-temporal (RS-OTS) sulci, using the MRI of nearly 3400 individuals including around 1000 twins. We reported both the associations between sulcal polymorphisms as well with a wide range of demographics (e.g. age, sex, handedness). Finally, we also estimated the heritability, and the genetic correlation between sulcal connections. We reported the frequency of the sulcal connections in the general population, which were hemisphere dependent. We found a sexual dimorphism of the connections, especially marked in the right hemisphere, with a CS-OTS connection more frequent in females (approximately 35-40% versus 20-25% in males) and an RS-CS connection more common in males (approximately 40-45% versus 25-30% in females). We confirmed associations between sulcal connections and characteristics of incomplete hippocampal inversion (IHI). We estimated the broad sense heritability to be 0.28-0.45 for RS-CS and CS-OTS connections, with hints of dominant contribution for the RS-CS connection. The connections appeared to share some of their genetic causing factors as indicated by strong genetic correlations. Heritability appeared much smaller for the (rarer) RS-OTS connection.