Search

Abstract The prospective study of youths at clinical high risk (CHR) for psychosis, including neuroimaging, can identify neural signatures predictive of psychosis outcomes using algorithms that integrate complex information. Here, to identify risk and psychosis conversion, we implemented multiple kernel learning (MKL), a multimodal machine learning approach allowing patterns from each modality to inform each other. Baseline multimodal scans (n = 74, 11 converters) included structural, resting-state functional imaging, and diffusion-weighted data. Multimodal MKL outperformed unimodal models (AUC = 0.73 vs. 0.66 in predicting conversion). Moreover, patterns learned by MKL were robust to training set variations, suggesting it can identify cross-modality redundancies and synergies to stabilize the predictive pattern. We identified many predictors consistent with the literature, including frontal cortices, cingulate, thalamus, and striatum. This highlights the advantage of methods that leverage the complex pathophysiology of psychosis.

Data collection, curation, and cleaning constitute a crucial phase in Machine Learning (ML) projects. In biomedical ML, it is often desirable to leverage multiple datasets to increase sample size and diversity, but this poses unique challenges, which arise from heterogeneity in study design, data descriptors, file system organization, and metadata. In this study, we present an approach to the integration of multiple brain MRI datasets with a focus on homogenization of their organization and preprocessing for ML. We use our own fusion example (approximately 84,000 images from 54,000 subjects, 12 studies, and 88 individual scanners) to illustrate and discuss the issues faced by study fusion efforts, and we examine key decisions necessary during dataset homogenization, presenting in detail a database structure flexible enough to accommodate multiple observational MRI datasets. We believe our approach can provide a basis for future similarly-minded biomedical ML projects.

Abstract Aberrant pauses are characteristic of schizophrenia and are robustly associated with its negative symptoms. Here, we found that pause behavior was associated with negative symptoms in individuals at clinical high risk (CHR) for psychosis, and with measures of syntactic complexity—phrase length and usage of determiners that introduce clauses—that we previously showed in this same CHR cohort to help comprise a classifier that predicted psychosis. These findings suggest a common impairment in discourse planning and verbal self-monitoring that affects both speech and language, and which is detected in clinical ratings of negative symptoms.

It is now possible to predict what a chemical smells like based on its chemical structure, however to date, this has only been done for a small number of odor descriptors. Here, using natural-language semantic representations, the authors demonstrate prediction of a much wider range of descriptors.

People vary in the extent to which they feel better after taking an inert, placebo, treatment, but the basis for individual placebo response is unclear. Here, the authors show how brain structural and functional variables, as well as personality traits, predict placebo response in those with chronic back pain.

In Huntington's disease (HD), accurate estimates of expected future motor impairments are key for clinical trials. Individual prognosis is only partially explained by genetics. However, studies so far have focused on predicting the time to clinical diagnosis based on fixed impairment levels, as opposed to predicting impairment in time windows comparable to the duration of a clinical trial. Here we evaluate an approach to both detect atrophy patterns associated with early degeneration and provide a prognosis of motor impairment within 3 years, using data from the TRACK-HD study on 80 premanifest HD (pre-HD) individuals and 85 age- and sex-matched healthy controls. We integrate anatomical MRI information from gray matter concentrations (estimated via voxel-based morphometry) together with baseline data from demographic, genetic and motor domains to distinguish individuals at high risk of developing pronounced future motor impairment from those at low risk. We evaluate the ability of models to distinguish between these two groups solely using baseline imaging data, as well as in combination with longitudinal imaging or non-imaging data. Our models show improved performance for motor prognosis through the incorporation of imaging features to non-imaging data, reaching 88% cross-validated accuracy when using baseline non-longitudinal information, and detect informative correlates in the caudate nucleus and the thalamus both for motor prognosis and early atrophy detection. These results show the plausibility of using baseline imaging and basic demographic/genetic measures for early detection of individuals at high risk of severe future motor impairment in relatively short timeframes. Keywords: Future motor impairment prediction, Premanifest Huntington's disease, Classification, Structural MRI, TRACK-HD

Neuroimaging: Brain connectivity pattern predicts symptom severity Brain network analyses from functional magnetic resonance imaging (fMRI) data may help diagnose schizophrenia and predict symptom severity. Detecting neuroimaging patterns requires large-scale analysis across multiple data sets. Mina Gheiratmand and colleagues from the University of Alberta, along with researchers at the IBM T.J. Watson Research Center analyzed brain imaging data from the Function Biomedical Informatics Research Network, a study designed to test the reproducibility of brain scan results taken on different fMRI machines from people with schizophrenia and schizoaffective disorders, as well as healthy controls. They studied brain networks at different levels of resolution from data gathered while study participants conducted a common auditory test. The researchers showed that they could discriminate between patients with schizophrenia and controls with 74% accuracy across multiple neuroimaging sites using the strength of connection in a functional network. They observed the most robust and discriminative connectivity differences between the thalamus and primary motor and sensory cortices as well as between the precuneus and other brain regions. Moreover, they could determine symptom severity based on the connectivity changes involving these areas. This new approach towards finding objective, reliable neuroimaging biomarkers for schizophrenia and its severity could be used for diagnosis and to assess disease progression and therapeutic efficacy.

The menstrual cycle affects many aspects of female physiology, from the immune system to behavioral and emotional regulation. It is unclear however if these physiological changes are reflected in everyday, naturalistic language production, and moreover whether these putative effects can be consistently quantified. Using a novel approach based on social networks, we characterized linguistic expression differences in female and male volunteers over the course of several months, while having no physiological or reported information of the female participants' menstrual cycles. We used a simple algorithm to quantify the linguistic affect intensity of 418 (184 females and 234 males) subjects using their social networks production and found a 7-day modulatory cycle of affect intensity that corresponds to labor-week fluctuations, with no significant difference by biological sex, and a 28-day cycle over which females are significantly different than males. Our results are consistent with the hypothesis that the menstrual cycle modulates affective features of naturalistic linguistic production.

In this work we analyze electro-corticography (ECoG) recordings in human sub- jects as they are anesthetized. We hypothesize that the decrease in responsiveness that defines anesthesia induction is concomitant with the stabilization of neuronal dynamics. To test this hypothesis, we performed a moving vector autoregressive analysis and quantified stability of neuronal dynamics using eigenmode decompo- sition of the autoregressive matrices, independently fitted to short sliding temporal windows. Consistent with the hypothesis we show that while the subject is awake, many modes of oscillations of neuronal activity are found at the edge of instabil- ity, but as the subject becomes anesthetized the fitted dynamics becomes more damped. Analysis of eigenmode distributions in the awake and anesthetized brain revealed statistically significant stabilization not present in surrogate data. Sta- bility analysis thus offer a novel way of quantifying changes in neuronal activity that characterize loss of consciousness induced by general anesthetics. Specifically, our analysis suggests that the effect of the anesthetic procedure is to damp out high frequency activity while still allowing for low frequency modes to perform a function.