Control-group feature normalization for multivariate pattern analysis of structural MRI data using the support vector machine

Predicting treatment outcomes in patients with panic disorder: Cross-sectional and two-year longitudinal structural connectome analysis using machine learning methods

PURPOSE

This study examined the relationship between structural brain networks and long-term treatment outcomes in patients with panic disorder (PD) using machine learning methods.

METHOD

The study involved 80 participants (53 PD patients and 27 healthy controls) and included clinical assessments and MRI scans at baseline and after two years (160 MRIs). Patients were categorized based on their response to two-year pharmacotherapy. Brain networks were analyzed using white matter tractography and network-based statistics.

RESULTS

Results showed structural network changes in PD patients, particularly in the extended fear network, including frontal regions, thalamus, and cingulate gyrus. Longitudinal analysis revealed that increased connections to the amygdala, hippocampus, and insula were associated with better treatment response. Conversely, overconnectivity in the amygdala and insula at baseline was associated with poor response, and similar patterns were found in the insula and parieto-occipital cortex related to non-remission. This study found that SVM and CPM could effectively predict treatment outcomes based on network pattern changes in PD.

CONCLUSIONS

These findings suggest that monitoring structural connectome changes in limbic and paralimbic regions is critical for understanding PD and tailoring treatment. The study highlights the potential of using personalized biomarkers to develop individualized treatment strategies for PD.

Intracranial EEG signals disentangle multi-areal neural dynamics of vicarious pain perception

Empathy enables understanding and sharing of others' feelings. Human neuroimaging studies have identified critical brain regions supporting empathy for pain, including the anterior insula (AI), anterior cingulate (ACC), amygdala, and inferior frontal gyrus (IFG). However, to date, the precise spatio-temporal profiles of empathic neural responses and inter-regional communications remain elusive. Here, using intracranial electroencephalography, we investigated electrophysiological signatures of vicarious pain perception. Others' pain perception induced early increases in high-gamma activity in IFG, beta power increases in ACC, but decreased beta power in AI and amygdala. Vicarious pain perception also altered the beta-band-coordinated coupling between ACC, AI, and amygdala, as well as increased modulation of IFG high-gamma amplitudes by beta phases of amygdala/AI/ACC. We identified a necessary combination of neural features for decoding vicarious pain perception. These spatio-temporally specific regional activities and inter-regional interactions within the empathy network suggest a neurodynamic model of human pain empathy.

Examining the Most Important Risk Factors for Predicting Youth Persistent and Distressing Psychotic-Like Experiences

BACKGROUND

Persistence and distress distinguish more clinically significant psychotic-like experiences (PLEs) from those that are less likely to be associated with impairment and/or need for care. Identifying risk factors that identify clinically relevant PLEs early in development is important for improving our understanding of the etiopathogenesis of these experiences. Machine learning analyses were used to examine the most important baseline factors distinguishing persistent distressing PLEs.

METHODS

Using Adolescent Brain Cognitive Development (ABCD) Study data on PLEs from 3 time points (ages 9-13 years), we created the following groups: individuals with persistent distressing PLEs (n = 305), individuals with transient distressing PLEs (n = 374), and individuals with low-level PLEs demographically matched to either the persistent distressing PLEs group (n = 305) or the transient distressing PLEs group (n = 374). Random forest classification models were trained to distinguish persistent distressing PLEs from low-level PLEs, transient distressing PLEs from low-level PLEs, and persistent distressing PLEs from transient distressing PLEs. Models were trained using identified baseline predictors as input features (i.e., cognitive, neural [cortical thickness, resting-state functional connectivity], developmental milestone delays, internalizing symptoms, adverse childhood experiences).

RESULTS

The model distinguishing persistent distressing PLEs from low-level PLEs showed the highest accuracy (test sample accuracy = 69.33%; 95% CI, 61.29%-76.59%). The most important predictors included internalizing symptoms, adverse childhood experiences, and cognitive functioning. Models for distinguishing persistent PLEs from transient distressing PLEs generally performed poorly.

CONCLUSIONS

Model performance metrics indicated that while most important factors overlapped across models (e.g., internalizing symptoms), adverse childhood experiences were especially important for predicting persistent distressing PLEs. Machine learning analyses proved useful for distinguishing the most clinically relevant group from the least clinically relevant group but showed limited ability to distinguish among clinically relevant groups that differed in PLE persistence.

Changes in dynamic and static brain fluctuation distinguish minimal hepatic encephalopathy and cirrhosis patients and predict the severity of liver damage

Purpose

Minimal hepatic encephalopathy (MHE) is characterized by mild neuropsychological and neurophysiological alterations that are not detectable by routine clinical examination. Abnormal brain activity (in terms of the amplitude of low-frequency fluctuation (ALFF) has been observed in MHE patients. However, little is known concerning temporal dynamics of intrinsic brain activity. The present study aimed to investigate the abnormal dynamics of brain activity (dynamic ALFF; dALFF) and static measures [static ALFF; (sALFF)] in MHE patients and to strive for a reliable imaging neuromarkers for distinguishing MHE patients from cirrhosis patients. In addition, the present study also investigated whether intrinsic brain activity predicted the severity of liver damage.

Methods

Thirty-four cirrhosis patients with MHE, 28 cirrhosis patients without MHE, and 33 age-, sex-, and education-matched healthy controls (HCs) underwent resting-state magnetic resonance imaging (rs-fMRI). dALFF was estimated by combining the ALFF method with the sliding-window method, in which temporal variability was quantized over the whole-scan timepoints and then compared among the three groups. Additionally, dALFF, sALFF and both two features were utilized as classification features in a support vector machine (SVM) to distinguish MHE patients from cirrhosis patients. The severity of liver damage was reflected by the Child–Pugh score. dALFF, sALFF and both two features were used to predict Child–Pugh scores in MHE patients using a general linear model.

Results

Compared with HCs, MHE patients showed significantly increased dALFF in the left inferior occipital gyrus, right middle occipital gyrus, and right insula; increased dALFF was also observed in the right posterior lobe of the cerebellum (CPL) and right thalamus. Compared with HCs, noMHE patients exhibited decreased dALFF in the right precuneus. In contrast, compared with noMHE patients, MHE patients showed increased dALFF in the right precuneus, right superior frontal gyrus, and right superior occipital gyrus. Furthermore, the increased dALFF values in the left precuneus were positively associated with poor digit-symbol test (DST) scores (r = 0.356, p = 0.038); however, dALFF in the right inferior temporal gyrus (ITG) was negatively associated with the number connection test–A (NCT-A) scores (r = -0.784, p = 0.000). A significant positive correlation was found between dALFF in the left inferior occipital gyrus (IOG) and high blood ammonia levels (r = 0.424, p = 0.012). Notably, dALFF values yielded a higher classification accuracy than sALFF values in distinguishing MHE patients from cirrhosis patients. Importantly, the dALFF values predicted the Child–Pugh score (r = 0.140, p = 0.030), whereas sALFF values did not in the current dataset. Combining two features had high accuracy in classification in distinguishing MHE patients from cirrhotic patients and yielded prediction in the severity of liver damage.

Conclusion

These findings suggest that combining dALFF and sALFF features is a useful neuromarkers for distinguishing MHE patients from cirrhosis patients and highlights the important role of dALFF feature in predicting the severity of liver damage in MHE.

Multi-site harmonization of MRI data uncovers machine-learning discrimination capability in barely separable populations: An example from the ABIDE dataset

Machine Learning (ML) techniques have been widely used in Neuroimaging studies of Autism Spectrum Disorders (ASD) both to identify possible brain alterations related to this condition and to evaluate the predictive power of brain imaging modalities. The collection and public sharing of large imaging samples has favored an even greater diffusion of the use of ML-based analyses. However, multi-center data collections may suffer the batch effect, which, especially in case of Magnetic Resonance Imaging (MRI) studies, should be curated to avoid confounding effects for ML classifiers and masking biases. This is particularly important in the study of barely separable populations according to MRI data, such as subjects with ASD compared to controls with typical development (TD). Here, we show how the implementation of a harmo- nization protocol on brain structural features unlocks the case-control ML separation capability in the analysis of a multi-center MRI dataset. This effect is demonstrated on the ABIDE data collection, involving subjects encompassing a wide age range. After data harmonization, the overall ASD vs. TD discrimination capability by a Random Forest (RF) classifier improves from a very low performance (AUC = 0.58 ± 0.04) to a still low, but reasonably significant AUC = 0.67 ± 0.03. The performances of the RF classifier have been evaluated also in the age-specific subgroups of children, adolescents and adults, obtaining AUC = 0.62 ± 0.02, AUC = 0.65 ± 0.03 and AUC = 0.69 ± 0.06, respectively. Specific and consistent patterns of anatomical differences related to the ASD condition have been identified for the three different age subgroups.

Feature Optimization Method of Material Identification for Loose Particles Inside Sealed Relays

Existing material identification for loose particles inside sealed relays focuses on the selection and optimization of classification algorithms, which ignores the features in the material dataset. In this paper, we propose a feature optimization method of material identification for loose particles inside sealed relays. First, for the missing value problem, multiple methods were used to process the material dataset. By comparing the identification accuracy achieved by a Random-Forest-based classifier (RF classifier) on the different processed datasets, the optimal direct-discarding method was obtained. Second, for the uneven data distribution problem, multiple methods were used to process the material dataset. By comparing the achieved identification accuracy, the optimal min–max standardization method was obtained. Then, for the feature selection problem, an innovative multi-index–fusion feature selection method was designed, and its superiority was verified through several tests. Test results show that the identification accuracy achieved by RF classifier on the dataset was improved from 59.63% to 63.60%. Test results of ten material verification datasets show that the identification accuracies achieved by RF classifier were greatly improved, with an average improvement of 3.01%. This strongly promotes research progress in loose particle material identification and is an important supplement to existing loose particle detection research. This is also the highest loose particle material identification accuracy achieved to in aerospace engineering, which has important practical value for improving the reliability of aerospace systems. Theoretically, it can be applied to feature optimization in machine learning.

Artificial intelligence in brain MRI analysis of Alzheimer's disease over the past 12 years: A systematic review

INTRODUCTION

Multiple structural brain changes in Alzheimer's disease (AD) and mild cognitive impairment (MCI) have been revealed on magnetic resonance imaging (MRI). There is a fast-growing effort in applying artificial intelligence (AI) to analyze these data. Here, we review and evaluate the AI studies in brain MRI analysis with synthesis.

METHODS

A systematic review of the literature, spanning the years from 2009 to 2020, was completed using the PubMed database. AI studies using MRI imaging to investigate normal aging, mild cognitive impairment, and AD-dementia were retrieved for review. Bias assessment was completed using the PROBAST criteria.

RESULTS

97 relevant studies were included in the review. The studies were typically focused on the classification of AD, MCI, and normal aging (71% of the reported studies) and the prediction of MCI conversion to AD (25%). The best performance was achieved by using the deep learning-based convolution neural network algorithms (weighted average accuracy 89%), in contrast to 76-86% using Logistic Regression, Support Vector Machines, and other AI methods.

DISCUSSION

The synthesized evidence is paramount to developing sophisticated AI approaches to reliably capture and quantify multiple subtle MRI changes in the whole brain that exemplify the complexity and heterogeneity of AD and brain aging.

The orbitofrontal cortex represents advantageous choice in the Iowa gambling task

A good‐based model, the central neurobiological model of economic decision‐making, proposes that the orbitofrontal cortex (OFC) represents binary choice outcome, that is, the chosen good. A good is defined by a group of determinants characterizing the conditions in which the commodity is offered, including commodity type, cost, risk, time delay, and ambiguity. Previous studies have found that the OFC represents the binary choice outcome in decision‐making tasks involving commodity type, cost, risk, and delay. Real‐life decisions are often complex and involve uncertainty, rewards, and penalties; however, whether the OFC represents binary choice outcomes in a complex decision‐making situation, for example, Iowa gambling task (IGT), remains unclear. Here, we propose that the OFC represents binary choice outcome, that is, advantageous choice versus disadvantageous choice, in the IGT. We propose two hypotheses: first, the activity pattern in the human OFC represents an advantageous choice; and second, choice induces an OFC‐related functional network. Using functional magnetic resonance imaging and advanced machine‐learning tools, we found that the OFC represented an advantageous choice in the IGT. The OFC representation of advantageous choice was related to decision‐making performance. Choice modulated the functional connectivity between the OFC and the superior medial gyrus. In conclusion, the OFC represents an advantageous choice during the IGT. In the framework of a good‐based model, the results extend the role of the OFC to complex decision‐making situation when making a binary choice.

Classification, Categorization, and Algorithms for Articular Cartilage Defects

There is a critical unmet need in the clinical implementation of valid preventative and therapeutic strategies for patients with articular cartilage pathology based on the significant gap in understanding of the relationships between diagnostic data, disease progression, patient-related variables, and symptoms. In this article, the current state of classification and categorization for articular cartilage pathology is discussed with particular focus on machine learning methods and the authors propose a bedside-bench-bedside approach with highly quantitative techniques as a solution to these hurdles. Leveraging computational learning with available data toward articular cartilage pathology patient phenotyping holds promise for clinical research and will likely be an important tool to identify translational solutions into evidence-based clinical applications to benefit patients. Recommendations for successful implementation of these approaches include using standardized definitions of articular cartilage, to include characterization of depth, size, location, and number; using measurements that minimize subjectivity or validated patient-reported outcome measures; considering not just the articular cartilage pathology but the whole joint, and the patient perception and perspective. Application of this approach through a multistep process by a multidisciplinary team of clinicians and scientists holds promise for validating disease mechanism-based phenotypes toward clinically relevant understanding of articular cartilage pathology for evidence-based application to orthopaedic practice.

Stratifying patients using fast multiple kernel learning framework: case studies of Alzheimer's disease and cancers

BACKGROUND

Predictive patient stratification is greatly emerging, because it allows us to prospectively identify which patients will benefit from what interventions before their condition worsens. In the biomedical research, a number of stratification methods have been successfully applied and have assisted treatment process. Because of heterogeneity and complexity of medical data, it is very challenging to integrate them and make use of them in practical clinic. There are two major challenges of data integration. Firstly, since the biomedical data has a high number of dimensions, combining multiple data leads to the hard problem of vast dimensional space handling. The computation is enormously complex and time-consuming. Secondly, the disparity of different data types causes another critical problem in machine learning for biomedical data. It has a great need to develop an efficient machine learning framework to handle the challenges.

METHODS

In this paper, we propose a fast-multiple kernel learning framework, referred to as fMKL-DR, that optimise equations to calculate matrix chain multiplication and reduce dimensions in data space. We applied our framework to two case studies, Alzheimer's disease (AD) patient stratification and cancer patient stratification. We performed several comparative evaluations on various biomedical datasets.

RESULTS

In the case study of AD patients, we enhanced significantly the multiple-ROIs approach based on MRI image data. The method could successfully classify not only AD patients and non-AD patients but also different phases of AD patients with AUC close to 1. In the case study of cancer patients, the framework was applied to six types of cancers, i.e., glioblastoma multiforme cancer, ovarian cancer, lung cancer, breast cancer, kidney cancer, and liver cancer. We efficiently integrated gene expression, miRNA expression, and DNA methylation. The results showed that the classification model basing on integrated datasets was much more accurate than classification model basing on the single data type.

CONCLUSIONS

The results demonstrated that the fMKL-DR remarkably improves computational cost and accuracy for both AD patient and cancer patient stratification. We optimised the data integration, dimension reduction, and kernel fusion. Our framework has great potential for mining large-scale cohort data and aiding personalised prevention.

Congruency and reactivation aid memory integration through reinstatement of prior knowledge

Building knowledge schemas that organize information and guide future learning is of great importance in everyday life. Such knowledge building is suggested to occur through reinstatement of prior knowledge during new learning, yielding integration of new with old memories supported by the medial prefrontal cortex (mPFC) and medial temporal lobe (MTL). Congruency with prior knowledge is also known to enhance subsequent memory. Yet, how reactivation and congruency interact to optimize memory integration is unknown. To investigate this question, we used an adapted AB-AC inference paradigm in combination with functional Magnetic Resonance Imaging (fMRI). Participants first studied an AB-association followed by an AC-association, so B (a scene) and C (an object) were indirectly linked through A (a pseudoword). BC-associations were either congruent or incongruent with prior knowledge (e.g. bathduck or hammer in a bathroom), and participants reported subjective B-reactivation strength while learning AC. Behaviorally, both congruency and reactivation enhanced memory integration. In the brain, these behavioral effects related to univariate and multivariate parametric effects in the MTL, mPFC, and Parahippocampal Place Area (PPA). Moreover, mPFC exhibited larger PPA-connectivity for more congruent associations. These outcomes provide insights into the neural mechanisms underlying memory enhancement, which has value for educational learning.

Identifying brain regions associated with the neuropathology of chronic low back pain: a resting-state amplitude of low-frequency fluctuation study

BACKGROUND

Previous studies have found widespread pain processing alterations in the brain in chronic low back pain (cLBP) patients. We aimed to (1) identify brain regions showing altered amplitude of low-frequency fluctuations (ALFF) using MRI and use these regions to discriminate cLBP patients from healthy controls (HCs) and (2) identify brain regions that are sensitive to cLBP pain intensity changes.

METHODS

We compared ALFF differences by MRI between cLBP subjects (90) and HCs (74), conducted a discriminative analysis to validate the results, and explored structural changes in key brain regions of cLBP. We also compared ALFF changes in cLBP patients after pain-exacerbating manoeuvres.

RESULTS

ALFF was increased in the post-/precentral gyrus (PoG/PrG), paracentral lobule (PCL)/supplementary motor area (SMA), and anterior cingulate cortex (ACC), and grey matter volume was increased in the left ACC in cLBP patients. PCL/SMA ALFF reliably discriminated cLBP patients from HCs in an independent cohort. cLBP patients showed increased ALFF in the insula, amygdala, hippocampal/parahippocampal gyrus, and thalamus and decreased ALFF in the default mode network (DMN) when their spontaneous low back pain intensity increased after the pain-exacerbating manoeuvre.

CONCLUSIONS

Brain low-frequency oscillations in the PCL, SMA, PoG, PrG, and ACC may be associated with the neuropathology of cLBP. Low-frequency oscillations in the insula, amygdala, hippocampal/parahippocampal gyrus, thalamus, and DMN are sensitive to manoeuvre-induced spontaneous back pain intensity changes.

Age Prediction Based on Brain MRI Image: A Survey

Human age prediction is an interesting and applicable issue in different fields. It can be based on various criteria such as face image, DNA methylation, chest plate radiographs, knee radiographs, dental images and etc. Most of the age prediction researches have mainly been based on images. Since the image processing and Machine Learning (ML) techniques have grown up, the investigations were led to use them in age prediction problem. The implementations would be used in different fields, especially in medical applications. Brain Age Estimation (BAE) has attracted more attention in recent years and it would be so helpful in early diagnosis of some neurodegenerative diseases such as Alzheimer, Parkinson, Huntington, etc. BAE is performed on Magnetic Resonance Imaging (MRI) images to compute the brain ages. Studies based on brain MRI shows that there is a relation between accelerated aging and accelerated brain atrophy. This refers to the effects of neurodegenerative diseases on brain structure while making the whole of it older. This paper reviews and summarizes the main approaches for age prediction based on brain MRI images including preprocessing methods, useful tools used in different research works and the estimation algorithms. We categorize the BAE methods based on two factors, first the way of processing MRI images, which includes pixel-based, surface-based, or voxel-based methods and second, the generation of ML algorithms that includes traditional or Deep Learning (DL) methods. The modern techniques as DL methods help MRI based age prediction to get results that are more accurate. In recent years, more precise and statistical ML approaches have been utilized with the help of related tools for simplifying computations and getting accurate results. Pros and cons of each research and the challenges in each work are expressed and some guidelines and deliberations for future research are suggested.

Utilization of Neurally Adjusted Ventilatory Assist (NAVA) Mode in Infants and Children Undergoing Congenital Heart Surgery: A Retrospective Review

We assessed the feasibility and the impact of NAVA compared to conventional modes of mechanical ventilation in ventilatory and gas exchange parameters in post-operative children with congenital heart disease. Infants and children (age < 18 years) that underwent congenital heart surgery were enrolled. Patients were ventilated with conventional synchronized intermittent mechanical ventilation (SIMV) and subsequently transitioned to NAVA during their cardiovascular intensive care unit (CVICU) stay. The ventilatory and gas exchange parameters for the 24 h pre- and post-transition to NAVA were compared. Additional parameters assessed included pain scores and sedation requirements. Eighty-one patients met inclusion criteria with a median age of 21 days (interquartile range 13 days-2 months). The majority of patients enrolled (75.3%) had complex congenital heart disease with high surgical severity scores. The transition to NAVA was tolerated by all patients without complications. The mean peak inspiratory pressure (PIP) was 1.8 cm H₂O lower (p < 0.001) and mean airway pressure (P_aw) was 0.5 cm H₂O lower (p = 0.009) on NAVA compared to conventional modes of mechanical ventilation. There was no significant difference in patients' respiratory rate, tidal volume, arterial pH, pCO₂, and lactate levels between the two modes of ventilation. There was a decreased sedation requirement during the time of NAVA ventilation. Comfort scores did not differ significantly with ventilator mode change. We concluded that NAVA is safe and well-tolerated mode of mechanical ventilation for our cohort of patients after congenital heart surgery. Compared to conventional ventilation there was a statistically significant decrease in PIP and P_aw on NAVA.

Improving Alzheimer's Disease Classification by Combining Multiple Measures

Several anatomical magnetic resonance imaging (MRI) markers for Alzheimer's disease (AD) have been identified. Cortical gray matter volume, cortical thickness, and subcortical volume have been used successfully to assist the diagnosis of Alzheimer's disease including its early warning and developing stages, e.g., mild cognitive impairment (MCI) including MCI converted to AD (MCIc) and MCI not converted to AD (MCInc). Currently, these anatomical MRI measures have mainly been used separately. Thus, the full potential of anatomical MRI scans for AD diagnosis might not yet have been used optimally. Meanwhile, most studies currently only focused on morphological features of regions of interest (ROIs) or interregional features without considering the combination of them. To further improve the diagnosis of AD, we propose a novel approach of extracting ROI features and interregional features based on multiple measures from MRI images to distinguish AD, MCI (including MCIc and MCInc), and health control (HC). First, we construct six individual networks based on six different anatomical measures (i.e., CGMV, CT, CSA, CC, CFI, and SV) and Automated Anatomical Labeling (AAL) atlas for each subject. Then, for each individual network, we extract all node (ROI) features and edge (interregional) features, and denoted as node feature set and edge feature set, respectively. Therefore, we can obtain six node feature sets and six edge feature sets from six different anatomical measures. Next, each feature within a feature set is ranked by -score in descending order, and the top ranked features of each feature set are applied to MKBoost algorithm to obtain the best classification accuracy. After obtaining the best classification accuracy, we can get the optimal feature subset and the corresponding classifier for each node or edge feature set. Afterwards, to investigate the classification performance with only node features, we proposed a weighted multiple kernel learning (wMKL) framework to combine these six optimal node feature subsets, and obtain a combined classifier to perform AD classification. Similarly, we can obtain the classification performance with only edge features. Finally, we combine both six optimal node feature subsets and six optimal edge feature subsets to further improve the classification performance. Experimental results show that the proposed method outperforms some state-of-the-art methods in AD classification, and demonstrate that different measures contain complementary information.

PAIR Comparison between Two Within-Group Conditions of Resting-State fMRI Improves Classification Accuracy

Classification approaches have been increasingly applied to differentiate patients and normal controls using resting-state functional magnetic resonance imaging data (RS-fMRI). Although most previous classification studies have reported promising accuracy within individual datasets, achieving high levels of accuracy with multiple datasets remains challenging for two main reasons: high dimensionality, and high variability across subjects. We used two independent RS-fMRI datasets (n = 31, 46, respectively) both with eyes closed (EC) and eyes open (EO) conditions. For each dataset, we first reduced the number of features to a small number of brain regions with paired t-tests, using the amplitude of low frequency fluctuation (ALFF) as a metric. Second, we employed a new method for feature extraction, named the PAIR method, examining EC and EO as paired conditions rather than independent conditions. Specifically, for each dataset, we obtained EC minus EO (EC—EO) maps of ALFF from half of subjects (n = 15 for dataset-1, n = 23 for dataset-2) and obtained EO—EC maps from the other half (n = 16 for dataset-1, n = 23 for dataset-2). A support vector machine (SVM) method was used for classification of EC RS-fMRI mapping and EO mapping. The mean classification accuracy of the PAIR method was 91.40% for dataset-1, and 92.75% for dataset-2 in the conventional frequency band of 0.01–0.08 Hz. For cross-dataset validation, we applied the classifier from dataset-1 directly to dataset-2, and vice versa. The mean accuracy of cross-dataset validation was 94.93% for dataset-1 to dataset-2 and 90.32% for dataset-2 to dataset-1 in the 0.01–0.08 Hz range. For the UNPAIR method, classification accuracy was substantially lower (mean 69.89% for dataset-1 and 82.97% for dataset-2), and was much lower for cross-dataset validation (64.69% for dataset-1 to dataset-2 and 64.98% for dataset-2 to dataset-1) in the 0.01–0.08 Hz range. In conclusion, for within-group design studies (e.g., paired conditions or follow-up studies), we recommend the PAIR method for feature extraction. In addition, dimensionality reduction with strong prior knowledge of specific brain regions should also be considered for feature selection in neuroimaging studies.

Harmonization of multi-site diffusion tensor imaging data

Diffusion tensor imaging (DTI) is a well-established magnetic resonance imaging (MRI) technique used for studying microstructural changes in the white matter. As with many other imaging modalities, DTI images suffer from technical between-scanner variation that hinders comparisons of images across imaging sites, scanners and over time. Using fractional anisotropy (FA) and mean diffusivity (MD) maps of 205 healthy participants acquired on two different scanners, we show that the DTI measurements are highly site-specific, highlighting the need of correcting for site effects before performing downstream statistical analyses. We first show evidence that combining DTI data from multiple sites, without harmonization, may be counter-productive and negatively impacts the inference. Then, we propose and compare several harmonization approaches for DTI data, and show that ComBat, a popular batch-effect correction tool used in genomics, performs best at modeling and removing the unwanted inter-site variability in FA and MD maps. Using age as a biological phenotype of interest, we show that ComBat both preserves biological variability and removes the unwanted variation introduced by site. Finally, we assess the different harmonization methods in the presence of different levels of confounding between site and age, in addition to test robustness to small sample size studies.