ABSTRACT
MRI has been extensively used to identify anatomical and functional differences in Autism Spectrum Disorder (ASD). Yet, many of these findings have proven difficult to replicate because studies rely on small cohorts and are built on many complex, undisclosed, analytic choices. We conducted an international challenge to predict ASD diagnosis from MRI data, where we provided preprocessed anatomical and functional MRI data from > 2,000 individuals. Evaluation of the predictions was rigorously blinded. 146 challengers submitted prediction algorithms, which were evaluated at the end of the challenge using unseen data and an additional acquisition site. On the best algorithms, we studied the importance of MRI modalities, brain regions, and sample size. We found evidence that MRI could predict ASD diagnosis: the 10 best algorithms reliably predicted diagnosis with AUCâ¼0.80 - far superior to what can be currently obtained using genotyping data in cohorts 20-times larger. We observed that functional MRI was more important for prediction than anatomical MRI, and that increasing sample size steadily increased prediction accuracy, providing an efficient strategy to improve biomarkers. We also observed that despite a strong incentive to generalise to unseen data, model development on a given dataset faces the risk of overfitting: performing well in cross-validation on the data at hand, but not generalising. Finally, we were able to predict ASD diagnosis on an external sample added after the end of the challenge (EU-AIMS), although with a lower prediction accuracy (AUC=0.72). This indicates that despite being based on a large multisite cohort, our challenge still produced biomarkers fragile in the face of dataset shifts.
Subject(s)
Autism Spectrum Disorder , Autistic Disorder , Autism Spectrum Disorder/diagnostic imaging , Autistic Disorder/diagnostic imaging , Biomarkers , Brain/diagnostic imaging , Humans , Magnetic Resonance Imaging/methodsABSTRACT
PURPOSE: Patients receiving treatment for acromegaly often experience significant associated comorbidities for which they are prescribed additional medications. We aimed to determine the real-world prevalence of comorbidities and concomitant medications in patients with acromegaly, and to investigate the association between frequency of comorbidities and number of concomitantly prescribed medications. METHODS: Administrative claims data were obtained from the IBM® MarketScan® database for a cohort of patients with acromegaly, identified by relevant diagnosis codes and acromegaly treatments, and a matched control cohort of patients without acromegaly from January 2010 through April 2020. Comorbidities were identified based on relevant claims and assessed for both cohorts. RESULTS: Overall, 1175 patients with acromegaly and 5875 matched patients without acromegaly were included. Patients with acromegaly had significantly more comorbidities and were prescribed concomitant medications more so than patients without acromegaly. In the acromegaly and control cohorts, respectively, 67.6% and 48.4% of patients had cardiovascular disorders, the most prevalent comorbidities, and 89.0% and 68.3% were prescribed > 3 concomitant medications (p < 0.0001). Hypopituitarism and hypothalamic disorders, sleep apnea, malignant neoplasms and cancer, and arthritis and musculoskeletal disorders were also highly prevalent in the acromegaly cohort. A moderate, positive correlation (Spearman correlation coefficient 0.60) was found between number of comorbidities and number of concomitant medications in the acromegaly cohort. CONCLUSION: Compared with patients without acromegaly, patients with acromegaly have significantly more comorbidities and are prescribed significantly more concomitant medications. Physicians should consider the number and type of ongoing medications for individual patients before prescribing additional acromegaly treatments.
Subject(s)
Acromegaly , Acromegaly/complications , Acromegaly/drug therapy , Acromegaly/epidemiology , Cohort Studies , Comorbidity , Databases, Factual , Humans , Prevalence , Retrospective Studies , United States/epidemiologyABSTRACT
Despite significant progress in the field, the detection of fMRI signal changes during hallucinatory events remains difficult and time-consuming. This article first proposes a machine-learning algorithm to automatically identify resting-state fMRI periods that precede hallucinations versus periods that do not. When applied to whole-brain fMRI data, state-of-the-art classification methods, such as support vector machines (SVM), yield dense solutions that are difficult to interpret. We proposed to extend the existing sparse classification methods by taking the spatial structure of brain images into account with structured sparsity using the total variation penalty. Based on this approach, we obtained reliable classifying performances associated with interpretable predictive patterns, composed of two clearly identifiable clusters in speech-related brain regions. The variation in transition-to-hallucination functional patterns not only from one patient to another but also from one occurrence to the next (e.g., also depending on the sensory modalities involved) appeared to be the major difficulty when developing effective classifiers. Consequently, second, this article aimed to characterize the variability within the prehallucination patterns using an extension of principal component analysis with spatial constraints. The principal components (PCs) and the associated basis patterns shed light on the intrinsic structures of the variability present in the dataset. Such results are promising in the scope of innovative fMRI-guided therapy for drug-resistant hallucinations, such as fMRI-based neurofeedback.
Subject(s)
Brain Mapping/methods , Brain/diagnostic imaging , Hallucinations/diagnostic imaging , Machine Learning , Magnetic Resonance Imaging/methods , Schizophrenia/diagnostic imaging , Adult , Auditory Perception/physiology , Brain/physiopathology , Female , Hallucinations/physiopathology , Humans , Male , Neural Pathways/diagnostic imaging , Neural Pathways/physiopathology , Neurofeedback , Pattern Recognition, Automated/methods , Principal Component Analysis , Schizophrenia/physiopathologyABSTRACT
BACKGROUND: Functional magnetic resonance imaging (fMRI) capture aims at detecting auditory-verbal hallucinations (AVHs) from continuously recorded brain activity. Establishing efficient capture methods with low computational cost that easily generalize between patients remains a key objective in precision psychiatry. To address this issue, we developed a novel automatized fMRI-capture procedure for AVHs in patients with schizophrenia (SCZ). METHODS: We used a previously validated but labor-intensive personalized fMRI-capture method to train a linear classifier using machine learning techniques. We benchmarked the performances of this classifier on 2320 AVH periods versus resting-state periods obtained from SCZ patients with frequent symptoms (n = 23). We characterized patterns of blood oxygen level-dependent activity that were predictive of AVH both within and between subjects. Generalizability was assessed with a second independent sample gathering 2000 AVH labels (n = 34 patients with SCZ), while specificity was tested with a nonclinical control sample performing an auditory imagery task (840 labels, n = 20). RESULTS: Our between-subject classifier achieved high decoding accuracy (area under the curve = 0.85) and discriminated AVH from rest and verbal imagery. Optimizing the parameters on the first schizophrenia dataset and testing its performance on the second dataset led to an out-of-sample area under the curve of 0.85 (0.88 for the converse test). We showed that AVH detection critically depends on local blood oxygen level-dependent activity patterns within Broca's area. CONCLUSIONS: Our results demonstrate that it is possible to reliably detect AVH states from fMRI blood oxygen level-dependent signals in patients with SCZ using a multivariate decoder without performing complex preprocessing steps. These findings constitute a crucial step toward brain-based treatments for severe drug-resistant hallucinations.
Subject(s)
Broca Area , Schizophrenia , Hallucinations , Humans , Magnetic Resonance Imaging , Oxygen SaturationABSTRACT
Autism spectrum disorder (ASD) is a developmental disorder underdiagnosed in adults. To date, no consistent evidence of alterations in brain structure has been reported in adults with ASD and few studies were conducted at that age. We analyzed structural magnetic resonance imaging data from 167 high functioning adults with ASD and 195 controls. We ran our analyses on a discovery (n = 301) and a replication sample (n = 61). The right caudal anterior cingulate cortical thickness was significantly thinner in adults with ASD compared to controls in both the discovery and the replication sample. Our work underlines the relevance of studying the brain anatomy of an adult ASD population.
Subject(s)
Autistic Disorder/diagnostic imaging , Cerebral Cortex/diagnostic imaging , Adolescent , Adult , Case-Control Studies , Female , Humans , Magnetic Resonance Imaging , MaleABSTRACT
Principal component analysis (PCA) is an exploratory tool widely used in data analysis to uncover the dominant patterns of variability within a population. Despite its ability to represent a data set in a low-dimensional space, PCA's interpretability remains limited. Indeed, the components produced by PCA are often noisy or exhibit no visually meaningful patterns. Furthermore, the fact that the components are usually non-sparse may also impede interpretation, unless arbitrary thresholding is applied. However, in neuroimaging, it is essential to uncover clinically interpretable phenotypic markers that would account for the main variability in the brain images of a population. Recently, some alternatives to the standard PCA approach, such as sparse PCA (SPCA), have been proposed, their aim being to limit the density of the components. Nonetheless, sparsity alone does not entirely solve the interpretability problem in neuroimaging, since it may yield scattered and unstable components. We hypothesized that the incorporation of prior information regarding the structure of the data may lead to improved relevance and interpretability of brain patterns. We therefore present a simple extension of the popular PCA framework that adds structured sparsity penalties on the loading vectors in order to identify the few stable regions in the brain images that capture most of the variability. Such structured sparsity can be obtained by combining, e.g., and total variation (TV) penalties, where the TV regularization encodes information on the underlying structure of the data. This paper presents the structured SPCA (denoted SPCA-TV) optimization framework and its resolution. We demonstrate SPCA-TV's effectiveness and versatility on three different data sets. It can be applied to any kind of structured data, such as, e.g., -dimensional array images or meshes of cortical surfaces. The gains of SPCA-TV over unstructured approaches (such as SPCA and ElasticNet PCA) or structured approach (such as GraphNet PCA) are significant, since SPCA-TV reveals the variability within a data set in the form of intelligible brain patterns that are easier to interpret and more stable across different samples.
Subject(s)
Image Processing, Computer-Assisted/methods , Magnetic Resonance Imaging/methods , Principal Component Analysis/methods , Algorithms , Brain/diagnostic imaging , Humans , Neuroimaging , Unsupervised Machine LearningABSTRACT
Mild traumatic brain injuries (mTBIs) are often associated with posttraumatic stress disorder (PTSD). In cases of chronic mTBI, accurate diagnosis can be challenging due to the overlapping symptoms this condition shares with PTSD. Furthermore, mTBIs are heterogeneous and not easily observed using conventional neuroimaging tools, despite the fact that diffuse axonal injuries are the most common injury. Diffusion tensor imaging (DTI) is sensitive to diffuse axonal injuries and is thus more likely to detect mTBIs, especially when analyses account for the inter-individual variability of these injuries. Using a subject-specific approach, we compared fractional anisotropy (FA) abnormalities between groups with a history of mTBI (n = 35), comorbid mTBI and PTSD (mTBI + PTSD; n = 22), and healthy controls (n = 37). We compared all three groups on the number of abnormal FA clusters derived from subject-specific injury profiles (i.e., individual z-score maps) along a common white matter skeleton. The mTBI + PTSD group evinced a greater number of abnormally low FA clusters relative to both the healthy controls and the mTBI group without PTSD (p < .05). Across the groups with a history of mTBI, increased numbers of abnormally low FA clusters were significantly associated with PTSD symptom severity, depression, post-concussion symptoms, and reduced information processing speed (p < .05). These findings highlight the utility of subject-specific microstructural analyses when searching for mTBI-related brain abnormalities, particularly in patients with PTSD. This study also suggests that patients with a history of mTBI and comorbid PTSD, relative to those without PTSD, are at increased risk of FA abnormalities.
Subject(s)
Brain Concussion/diagnostic imaging , Brain/diagnostic imaging , Diffusion Tensor Imaging , Stress Disorders, Post-Traumatic/diagnostic imaging , White Matter/diagnostic imaging , Adult , Brain Concussion/complications , Brain Concussion/epidemiology , Comorbidity , Female , Humans , Male , Stress Disorders, Post-Traumatic/complications , Stress Disorders, Post-Traumatic/epidemiologyABSTRACT
Harmonizing diffusion MRI (dMRI) images across multiple sites is imperative for joint analysis of the data to significantly increase the sample size and statistical power of neuroimaging studies. In this work, we develop a method to harmonize diffusion MRI data across multiple sites and scanners that incorporates two main novelties: i) we take into account the spatial variability of the signal (for different sites) in different parts of the brain as opposed to existing methods, which consider one linear statistical covariate for the entire brain; ii) our method is model-free, in that no a-priori model of diffusion (e.g., tensor, compartmental models, etc.) is assumed and the signal itself is corrected for scanner related differences. We use spherical harmonic basis functions to represent the signal and compute several rotation invariant features, which are used to estimate a regionally specific linear mapping between signal from different sites (and scanners). We validate our method on diffusion data acquired from four different sites (including two GE and two Siemens scanners) on a group of healthy subjects. Diffusion measures such fractional anisotropy, mean diffusivity and generalized fractional anisotropy are compared across multiple sites before and after the mapping. Our experimental results demonstrate that, for identical acquisition protocol across sites, scanner-specific differences can be accurately removed using the proposed method.