Joint prediction of multiple scores captures better individual traits from brain images

Cross-dataset Evaluation of Dementia Longitudinal Progression Prediction Models

Accurate Alzheimer's Disease (AD) progression prediction is essential for early intervention. The TADPOLE challenge, involving 92 algorithms, used multimodal biomarkers to predict future clinical diagnosis, cognition, and ventricular volume. The winning algorithm, FROG, utilized a Longitudinal-to-Cross-sectional (L2C) transformation to convert variable longitudinal histories into fixed-length feature vectors, which contrasted with most existing approaches that fitted models to entire longitudinal histories, e.g., AD Course Map (AD-Map) and minimal recurrent neural networks (MinimalRNN). The TADPOLE challenge only utilized the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. To evaluate FROG's generalizability, we trained it on the ADNI dataset and tested it on three external datasets covering 2,312 participants and 13,200 timepoints. We also introduced two FROG variants. One variant, L2C feedforward neural network (L2C-FNN), unified all XGBoost models used by the original FROG with an FNN. Across external datasets, L2C-FNN and AD-Map were the best for predicting cognition and ventricular volume. For clinical diagnosis prediction, L2C-FNN was the best, while AD-Map was the worst. L2C-FNN compared favorably with other approaches regardless of the number of observed timepoints, and when predicting from 0 to 6 years into the future, underscoring its potential for long-term dementia progression prediction. Pretrained ADNI models are publicly available: GITHUB_LINK.

On the stability of canonical correlation analysis and partial least squares with application to brain-behavior associations

Associations between datasets can be discovered through multivariate methods like Canonical Correlation Analysis (CCA) or Partial Least Squares (PLS). A requisite property for interpretability and generalizability of CCA/PLS associations is stability of their feature patterns. However, stability of CCA/PLS in high-dimensional datasets is questionable, as found in empirical characterizations. To study these issues systematically, we developed a generative modeling framework to simulate synthetic datasets. We found that when sample size is relatively small, but comparable to typical studies, CCA/PLS associations are highly unstable and inaccurate; both in their magnitude and importantly in the feature pattern underlying the association. We confirmed these trends across two neuroimaging modalities and in independent datasets with n ≈ 1000 and n = 20,000, and found that only the latter comprised sufficient observations for stable mappings between imaging-derived and behavioral features. We further developed a power calculator to provide sample sizes required for stability and reliability of multivariate analyses. Collectively, we characterize how to limit detrimental effects of overfitting on CCA/PLS stability, and provide recommendations for future studies.

Prediction of cognitive performance differences in older age from multimodal neuroimaging data

Differences in brain structure and functional and structural network architecture have been found to partly explain cognitive performance differences in older ages. Thus, they may serve as potential markers for these differences. Initial unimodal studies, however, have reported mixed prediction results of selective cognitive variables based on these brain features using machine learning (ML). Thus, the aim of the current study was to investigate the general validity of cognitive performance prediction from imaging data in healthy older adults. In particular, the focus was with examining whether (1) multimodal information, i.e., region-wise grey matter volume (GMV), resting-state functional connectivity (RSFC), and structural connectivity (SC) estimates, may improve predictability of cognitive targets, (2) predictability differences arise for global cognition and distinct cognitive profiles, and (3) results generalize across different ML approaches in 594 healthy older adults (age range: 55-85 years) from the 1000BRAINS study. Prediction potential was examined for each modality and all multimodal combinations, with and without confound (i.e., age, education, and sex) regression across different analytic options, i.e., variations in algorithms, feature sets, and multimodal approaches (i.e., concatenation vs. stacking). Results showed that prediction performance differed considerably between deconfounding strategies. In the absence of demographic confounder control, successful prediction of cognitive performance could be observed across analytic choices. Combination of different modalities tended to marginally improve predictability of cognitive performance compared to single modalities. Importantly, all previously described effects vanished in the strict confounder control condition. Despite a small trend for a multimodal benefit, developing a biomarker for cognitive aging remains challenging.

Multilayer meta-matching: translating phenotypic prediction models from multiple datasets to small data

Resting-state functional connectivity (RSFC) is widely used to predict phenotypic traits in individuals. Large sample sizes can significantly improve prediction accuracies. However, for studies of certain clinical populations or focused neuroscience inquiries, small-scale datasets often remain a necessity. We have previously proposed a "meta-matching" approach to translate prediction models from large datasets to predict new phenotypes in small datasets. We demonstrated large improvement of meta-matching over classical kernel ridge regression (KRR) when translating models from a single source dataset (UK Biobank) to the Human Connectome Project Young Adults (HCP-YA) dataset. In the current study, we propose two meta-matching variants ("meta-matching with dataset stacking" and "multilayer meta-matching") to translate models from multiple source datasets across disparate sample sizes to predict new phenotypes in small target datasets. We evaluate both approaches by translating models trained from five source datasets (with sample sizes ranging from 862 participants to 36,834 participants) to predict phenotypes in the HCP-YA and HCP-Aging datasets. We find that multilayer meta-matching modestly outperforms meta-matching with dataset stacking. Both meta-matching variants perform better than the original "meta-matching with stacking" approach trained only on the UK Biobank. All meta-matching variants outperform classical KRR and transfer learning by a large margin. In fact, KRR is better than classical transfer learning when less than 50 participants are available for finetuning, suggesting the difficulty of classical transfer learning in the very small sample regime. The multilayer meta-matching model is publicly available at GITHUB_LINK.

Predicting sex, age, general cognition and mental health with machine learning on brain structural connectomes

There is an increasing expectation that advanced, computationally expensive machine learning (ML) techniques, when applied to large population-wide neuroimaging datasets, will help to uncover key differences in the human brain in health and disease. We take a comprehensive approach to explore how multiple aspects of brain structural connectivity can predict sex, age, general cognitive function and general psychopathology, testing different ML algorithms from deep learning (DL) model (BrainNetCNN) to classical ML methods. We modelled N = 8183 structural connectomes from UK Biobank using six different structural network weightings obtained from diffusion MRI. Streamline count generally provided the highest prediction accuracies in all prediction tasks. DL did not improve on prediction accuracies from simpler linear models. Further, high correlations between gradient attribution coefficients from DL and model coefficients from linear models suggested the models ranked the importance of features in similar ways, which indirectly suggested the similarity in models' strategies for making predictive decision to some extent. This highlights that model complexity is unlikely to improve detection of associations between structural connectomes and complex phenotypes with the current sample size.

Evidence for embracing normative modeling

In this work, we expand the normative model repository introduced in Rutherford et al., 2022a to include normative models charting lifespan trajectories of structural surface area and brain functional connectivity, measured using two unique resting-state network atlases (Yeo-17 and Smith-10), and an updated online platform for transferring these models to new data sources. We showcase the value of these models with a head-to-head comparison between the features output by normative modeling and raw data features in several benchmarking tasks: mass univariate group difference testing (schizophrenia versus control), classification (schizophrenia versus control), and regression (predicting general cognitive ability). Across all benchmarks, we show the advantage of using normative modeling features, with the strongest statistically significant results demonstrated in the group difference testing and classification tasks. We intend for these accessible resources to facilitate the wider adoption of normative modeling across the neuroimaging community.

Supervised Phenotype Discovery From Multimodal Brain Imaging

Data-driven discovery of image-derived phenotypes (IDPs) from large-scale multimodal brain imaging data has enormous potential for neuroscientific and clinical research by linking IDPs to subjects' demographic, behavioural, clinical and cognitive measures (i.e., non-imaging derived phenotypes or nIDPs). However, current approaches are primarily based on unsupervised approaches, without the use of information in nIDPs. In this paper, we proposed a semi-supervised, multimodal, and multi-task fusion approach, termed SuperBigFLICA, for IDP discovery, which simultaneously integrates information from multiple imaging modalities as well as multiple nIDPs. SuperBigFLICA is computationally efficient and largely avoids the need for parameter tuning. Using the UK Biobank brain imaging dataset with around 40,000 subjects and 47 modalities, along with more than 17,000 nIDPs, we showed that SuperBigFLICA enhances the prediction power of nIDPs, benchmarked against IDPs derived by conventional expert-knowledge and unsupervised-learning approaches (with average nIDP prediction accuracy improvements of up to 46%). It also enables the learning of generic imaging features that can predict new nIDPs. Further empirical analysis of the SuperBigFLICA algorithm demonstrates its robustness in different prediction tasks and the ability to derive biologically meaningful IDPs in predicting health outcomes and cognitive nIDPs, such as fluid intelligence and hypertension.

Predicting future cognitive decline from non-brain and multimodal brain imaging data in healthy and pathological aging

Previous literature has focused on predicting a diagnostic label from structural brain imaging. Since subtle changes in the brain precede a cognitive decline in healthy and pathological aging, our study predicts future decline as a continuous trajectory instead. Here, we tested whether baseline multimodal neuroimaging data improve the prediction of future cognitive decline in healthy and pathological aging. Nonbrain data (demographics, clinical, and neuropsychological scores), structural MRI, and functional connectivity data from OASIS-3 (N = 662; age = 46-96 years) were entered into cross-validated multitarget random forest models to predict future cognitive decline (measured by CDR and MMSE), on average 5.8 years into the future. The analysis was preregistered, and all analysis code is publicly available. Combining non-brain with structural data improved the continuous prediction of future cognitive decline (best test-set performance: R2 = 0.42). Cognitive performance, daily functioning, and subcortical volume drove the performance of our model. Including functional connectivity did not improve predictive accuracy. In the future, the prognosis of age-related cognitive decline may enable earlier and more effective individualized cognitive, pharmacological, and behavioral interventions.

A benchmark for prediction of psychiatric multimorbidity from resting EEG data in a large pediatric sample

Psychiatric disorders are among the most common and debilitating illnesses across the lifespan and begin usually during childhood and adolescence, which emphasizes the importance of studying the developing brain. Most of the previous pediatric neuroimaging studies employed traditional univariate statistics on relatively small samples. Multivariate machine learning approaches have a great potential to overcome the limitations of these approaches. On the other hand, the vast majority of existing multivariate machine learning studies have focused on differentiating between children with an isolated psychiatric disorder and typically developing children. However, this line of research does not reflect the real-life situation as the majority of children with a clinical diagnosis have multiple psychiatric disorders (multimorbidity), and consequently, a clinician has the task to choose between different diagnoses and/or the combination of multiple diagnoses. Thus, the goal of the present benchmark is to predict psychiatric multimorbidity in children and adolescents. For this purpose, we implemented two kinds of machine learning benchmark challenges: The first challenge targets the prediction of the seven most prevalent DSM-V psychiatric diagnoses for the available data set, of which each individual can exhibit multiple ones concurrently (i.e. multi-task multi-label classification). Based on behavioral and cognitive measures, a second challenge focuses on predicting psychiatric symptom severity on a dimensional level (i.e. multiple regression task). For the present benchmark challenges, we will leverage existing and future data from the biobank of the Healthy Brain Network (HBN) initiative, which offers a unique large-sample dataset (N = 2042) that provides a wide array of different psychiatric developmental disorders and true hidden data sets. Due to limited real-world practicability and economic viability of MRI measurements, the present challenge will permit only resting state EEG data and demographic information to derive predictive models. We believe that a community driven effort to derive predictive markers from these data using advanced machine learning algorithms can help to improve the diagnosis of psychiatric developmental disorders.

A framework for deep multitask learning with multiparametric magnetic resonance imaging for the joint prediction of histological characteristics in breast cancer

The clinical management and decision-making process related to breast cancer are based on multiple histological indicators. This study aims to jointly predict the Ki-67 expression level, luminal A subtype and histological grade molecular biomarkers using a new deep multitask learning method with multiparametric magnetic resonance imaging. A multitask learning network structure was proposed by introducing a common-task layer and task-specific layers to learn the high-level features that are common to all tasks and related to a specific task, respectively. A network pretrained with knowledge from the ImageNet dataset was used and fine-tuned with MRI data. Information from multiparametric MR images was fused using the strategy at the feature and decision levels. The area under the receiver operating characteristic curve (AUC) was used to measure model performance. For single-task learning using a single image series, the deep learning model generated AUCs of 0.752, 0.722, and 0.596 for the Ki-67, luminal A and histological grade prediction tasks, respectively. The performance was improved by freezing the first 5 convolutional layers, using 20% shared layers and fusing multiparametric series at the feature level, which achieved AUCs of 0.819, 0.799 and 0.747 for Ki-67, luminal A and histological grade prediction tasks, respectively. Our study showed advantages in jointly predicting correlated clinical biomarkers using a deep multitask learning framework with an appropriate number of fine-tuned convolutional layers by taking full advantage of common and complementary imaging features. Multiparametric image series-based multitask learning could be a promising approach for the multiple clinical indicator-based management of breast cancer.

Pattern learning reveals brain asymmetry to be linked to socioeconomic status

Socioeconomic status (SES) anchors individuals in their social network layers. Our embedding in the societal fabric resonates with habitus, world view, opportunity, and health disparity. It remains obscure how distinct facets of SES are reflected in the architecture of the central nervous system. Here, we capitalized on multivariate multi-output learning algorithms to explore possible imprints of SES in gray and white matter structure in the wider population (n ≈ 10,000 UK Biobank participants). Individuals with higher SES, compared with those with lower SES, showed a pattern of increased region volumes in the left brain and decreased region volumes in the right brain. The analogous lateralization pattern emerged for the fiber structure of anatomical white matter tracts. Our multimodal findings suggest hemispheric asymmetry as an SES-related brain signature, which was consistent across six different indicators of SES: degree, education, income, job, neighborhood and vehicle count. Hence, hemispheric specialization may have evolved in human primates in a way that reveals crucial links to SES.

Correlation Between Brain Structure Atrophy and Plasma Amyloid-β and Phosphorylated Tau in Patients With Alzheimer’s Disease and Amnestic Mild Cognitive Impairment Explored by Surface-Based Morphometry

Objective

To investigate the changes in the cortical thickness of the region of interest (ROI) and plasma Aβ40, Aβ42, and phosphorylated Tau (P-Tau) concentrations in patients with Alzheimer’s disease (AD) and amnestic mild cognitive impairment (aMCI) as the disease progressed with surface-based morphometry (SBM), to analyze the correlation between ROI cortical thickness and measured plasma indexes and neuropsychological scales, and to explore the clinical value of ROI cortical thickness combined with plasma Aβ40, Aβ42, and P-Tau in the early recognition and diagnosis of AD.

Methods

This study enrolled 33 patients with AD, 48 patients with aMCI, and 33 healthy controls (normal control, NC). Concentration changes in plasma Aβ42, Aβ40, and P-Tau collected in each group were analyzed. Meanwhile, the whole brain T1 structure images (T1WI-3D-MPRAGE) of each group of patients were collected, and T1 image in AD-aMCI, AD-NC, and aMCI-NC group were analyzed and processed by SBM technology to obtain brain regions with statistical differences as clusters, and the cortical thickness of each cluster was extracted. Multivariate ordered logistic regression analysis was used to screen out the measured plasma indexes and the indexes with independent risk factors in the cortical thickness of each cluster. Three comparative receiver operating characteristic (ROC) curves of AD-aMCI, AD-NC, and aMCI-NC groups were plotted, respectively, to explore the diagnostic value of multi-factor combined prediction for cognitive impairment. The relationship between cortical thickness and plasma indexes, and between cortical thickness and Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) scores were clarified by Pearson correlation analysis.

Results

Plasma Aβ40, Aβ42, and P-Tau proteins in the NC, aMCI, and AD groups increased with the progression of AD (P < 0.01); cortical thickness reductions in the AD-aMCI groups and AD-NC groups mainly involved the bilateral superior temporal gyrus, transverse temporal gyrus, superior marginal gyrus, insula, right entorhinal cortex, right fusiform gyrus, and cingulate gyrus. However, there were no statistical significances in cortical thickness reductions in the aMCI and NC groups. The cortical thickness of the ROI was negatively correlated with plasma Aβ40, Aβ42, and P-Tau concentrations (P < 0.05), and the cortical thickness of the ROI was positively correlated with MMSE and MoCA scores. Independent risk factors such as Aβ40, Aβ42, P-Tau, and AD-NC cluster 1R (right superior temporal gyrus, temporal pole, entorhinal cortex, transverse temporal gyrus, fusiform gyrus, superior marginal gyrus, middle temporal gyrus, and inferior temporal gyrus) were combined to plot ROC curves. The diagnostic efficiency of plasma indexes was higher than that of cortical thickness indexes, the diagnostic efficiency of ROC curves after the combination of cortical thickness and plasma indexes was higher than that of cortical thickness or plasma indexes alone.

Conclusion

Plasma Aβ40, Aβ42, and P-Tau may be potential biomarkers for early prediction of AD. As the disease progressed, AD patients developed cortical atrophy characterized by atrophy of the medial temporal lobe. The combined prediction of these region and plasma Aβ40, Aβ42, and P-Tau had a higher diagnostic value than single-factor prediction for cognitive decline.

Precision medicine in stroke: towards personalized outcome predictions using artificial intelligence

Stroke ranks among the leading causes for morbidity and mortality worldwide. New and continuously improving treatment options such as thrombolysis and thrombectomy have revolutionized acute stroke treatment in recent years. Following modern rhythms, the next revolution might well be the strategic use of the steadily increasing amounts of patient-related data for generating models enabling individualized outcome predictions. Milestones have already been achieved in several health care domains, as big data and artificial intelligence have entered everyday life. The aim of this review is to synoptically illustrate and discuss how artificial intelligence approaches may help to compute single-patient predictions in stroke outcome research in the acute, subacute and chronic stage. We will present approaches considering demographic, clinical and electrophysiological data, as well as data originating from various imaging modalities and combinations thereof. We will outline their advantages, disadvantages, their potential pitfalls and the promises they hold with a special focus on a clinical audience. Throughout the review we will highlight methodological aspects of novel machine-learning approaches as they are particularly crucial to realize precision medicine. We will finally provide an outlook on how artificial intelligence approaches might contribute to enhancing favourable outcomes after stroke.

Tackling the Complexity of Lesion-Symptoms Mapping: How to Bridge the Gap Between Data Scientists and Clinicians?

Accurate and predictive lesion-symptoms mapping is a major goal in the field of clinical neurosciences. Recent studies have called for a reappraisal of the results given by the standard univariate voxel-based lesion-symptom mapping technique, emphasizing the need of developing multivariate methods. While the organization of large datasets and their analysis with machine learning (ML) approaches represents an opportunity to increase prediction accuracy, the complexity and dimensionality of the problem remain a major obstacle. Acknowledging the difficulty of inferring individual outcomes from the observation of spatial patterns of lesions, we propose here to base prediction on new individuals on models of brain connectivity, whereby the disruption of a given network predicts the occurrence of selective deficits. Well-suited ML tools are necessary to capture the relevant information from limited datasets and perform reliable inference.

Toward Robust Anxiety Biomarkers: A Machine Learning Approach in a Large-Scale Sample

BACKGROUND

The field of psychiatry has long sought biomarkers that can objectively diagnose patients, predict treatment response, or identify individuals at risk of illness onset. However, reliable psychiatric biomarkers have yet to emerge. The recent application of machine learning techniques to develop neuroimaging-based biomarkers has yielded promising preliminary results. However, much of the work in this domain has not met best practice standards from the field of machine learning. This is especially true for studies of anxiety, creating uncertainty about the potential for anxiety biomarker development.

METHODS

We applied machine learning tools to predict trait anxiety from neuroimaging measurements in humans. Using publicly available data from the Brain Genomics Superstruct Project, we compared a suite of neuroimaging-based machine learning models predicting anxiety within a discovery sample (n = 531, 307 women) via k-fold cross-validation, and we tested the final model (a stacked model incorporating region-to-region functional connectivity, amygdala seed-to-voxel connectivity, and volumetric and cortical thickness data) in a held-out, unseen test sample (n = 348, 209 women).

RESULTS

Though the best model was able to predict anxiety within the discovery sample (cross-validated R2 of .06, permutation test p < .001), the generalization test within the holdout sample failed (R2 of -.04, permutation test p > .05).

CONCLUSIONS

In this study, we did not find evidence of a generalizable anxiety biomarker. However, we encourage other researchers to investigate this topic, utilizing large samples and proper methodology, to clarify the potential of neuroimaging-based anxiety biomarkers.

Model-based whole-brain effective connectivity to study distributed cognition in health and disease

Neuroimaging techniques are now widely used to study human cognition. The functional associations between brain areas have become a standard proxy to describe how cognitive processes are distributed across the brain network. Among the many analysis tools available, dynamic models of brain activity have been developed to overcome the limitations of original connectivity measures such as functional connectivity. This goes in line with the many efforts devoted to the assessment of directional interactions between brain areas from the observed neuroimaging activity. This opinion article provides an overview of our model-based whole-brain effective connectivity to analyze fMRI data, while discussing the pros and cons of our approach with respect to other established approaches. Our framework relies on the multivariate Ornstein-Uhlenbeck (MOU) process and is thus referred to as MOU-EC. Once tuned, the model provides a directed connectivity estimate that reflects the dynamical state of BOLD activity, which can be used to explore cognition. We illustrate this approach using two applications on task-evoked fMRI data. First, as a connectivity measure, MOU-EC can be used to extract biomarkers for task-specific brain coordination, understood as the patterns of areas exchanging information. The multivariate nature of connectivity measures raises several challenges for whole-brain analysis, for which machine-learning tools present some advantages over statistical testing. Second, we show how to interpret changes in MOU-EC connections in a collective and model-based manner, bridging with network analysis. Our framework provides a comprehensive set of tools that open exciting perspectives to study distributed cognition, as well as neuropathologies.

A joint network optimization framework to predict clinical severity from resting state functional MRI data

We propose a novel optimization framework to predict clinical severity from resting state fMRI (rs-fMRI) data. Our model consists of two coupled terms. The first term decomposes the correlation matrices into a sparse set of representative subnetworks that define a network manifold. These subnetworks are modeled as rank-one outer-products which correspond to the elemental patterns of co-activation across the brain; the subnetworks are combined via patient-specific non-negative coefficients. The second term is a linear regression model that uses the patient-specific coefficients to predict a measure of clinical severity. We validate our framework on two separate datasets in a ten fold cross validation setting. The first is a cohort of fifty-eight patients diagnosed with Autism Spectrum Disorder (ASD). The second dataset consists of sixty three patients from a publicly available ASD database. Our method outperforms standard semi-supervised frameworks, which employ conventional graph theoretic and statistical representation learning techniques to relate the rs-fMRI correlations to behavior. In contrast, our joint network optimization framework exploits the structure of the rs-fMRI correlation matrices to simultaneously capture group level effects and patient heterogeneity. Finally, we demonstrate that our proposed framework robustly identifies clinically relevant networks characteristic of ASD.

Deep neural networks and kernel regression achieve comparable accuracies for functional connectivity prediction of behavior and demographics

There is significant interest in the development and application of deep neural networks (DNNs) to neuroimaging data. A growing literature suggests that DNNs outperform their classical counterparts in a variety of neuroimaging applications, yet there are few direct comparisons of relative utility. Here, we compared the performance of three DNN architectures and a classical machine learning algorithm (kernel regression) in predicting individual phenotypes from whole-brain resting-state functional connectivity (RSFC) patterns. One of the DNNs was a generic fully-connected feedforward neural network, while the other two DNNs were recently published approaches specifically designed to exploit the structure of connectome data. By using a combined sample of almost 10,000 participants from the Human Connectome Project (HCP) and UK Biobank, we showed that the three DNNs and kernel regression achieved similar performance across a wide range of behavioral and demographic measures. Furthermore, the generic feedforward neural network exhibited similar performance to the two state-of-the-art connectome-specific DNNs. When predicting fluid intelligence in the UK Biobank, performance of all algorithms dramatically improved when sample size increased from 100 to 1000 subjects. Improvement was smaller, but still significant, when sample size increased from 1000 to 5000 subjects. Importantly, kernel regression was competitive across all sample sizes. Overall, our study suggests that kernel regression is as effective as DNNs for RSFC-based behavioral prediction, while incurring significantly lower computational costs. Therefore, kernel regression might serve as a useful baseline algorithm for future studies.

Joint Prediction of Breast Cancer Histological Grade and Ki-67 Expression Level Based on DCE-MRI and DWI Radiomics

OBJECTIVE

Histologic grade and Ki-67 proliferation status are important clinical indictors for breast cancer prognosis and treatment. The purpose of this study is to improve prediction accuracy of these clinical indicators based on tumor radiomic analysis.

METHODS

We jointly predicted Ki-67 and tumor grade with a multitask learning framework by separately utilizing radiomics from tumor MRI series. Additionally, we showed how multitask learning models (MTLs) could be extended to combined radiomics from the MRI series for a better prediction based on the assumption that features from different sources of images share common patterns while providing complementary information. Tumor radiomic analysis was performed with morphological, statistical and textural features extracted on the DWI and dynamic contrast-enhanced MRI (DCE-MRI) series of the precontrast and subtraction images, respectively.

RESULTS

Joint prediction of Ki-67 status and tumor grade on MR images using the MTL achieved performance improvements over that of single-task-based predictive models. Similarly, for the prediction tasks of Ki-67 and tumor grade, the MTL for combined precontrast and apparent diffusion coefficient (ADC) images achieved AUCs of 0.811 and 0.816, which were significantly better than that of the single-task- based model with p values of 0.005 and 0.017, respectively.

CONCLUSION

Mapping MRI radiomics to two related clinical indicators improves prediction performance for both Ki-67 expression level and tumor grade.

SIGNIFICANCE

Joint prediction of indicators by multitask learning that combines correlations of MRI radiomics is important for optimal tumor therapy and treatment because clinical decisions are made by integrating multiple clinical indicators.

Extracting orthogonal subject- and condition-specific signatures from fMRI data using whole-brain effective connectivity

The study of brain communication based on fMRI data is often limited because such measurements are a mixture of session-to-session variability with subject- and condition-related information. Disentangling these contributions is crucial for real-life applications, in particular when only a few recording sessions are available. The present study aims to define a reliable standard for the extraction of multiple signatures from fMRI data, while verifying that they do not mix information about the different modalities (e.g., subjects and conditions such as tasks performed by them). In particular, condition-specific signatures should not be contaminated by subject-related information, since they aim to generalize over subjects. Practically, signatures correspond to subnetworks of directed interactions between brain regions (typically 100 covering the whole brain) supporting the subject and condition identification for single fMRI sessions. The key for robust prediction is using effective connectivity instead of functional connectivity. Our method demonstrates excellent generalization capabilities for subject identification in two datasets, using only a few sessions per subject as reference. Using another dataset with resting state and movie viewing, we show that the two signatures related to subjects and tasks correspond to distinct subnetworks, which are thus topologically orthogonal. Our results set solid foundations for applications tailored to individual subjects, such as clinical diagnostic.

Enhanced estimations of post-stroke aphasia severity using stacked multimodal predictions

The severity of post-stroke aphasia and the potential for recovery are highly variable and difficult to predict. Evidence suggests that optimal estimation of aphasia severity requires the integration of multiple neuroimaging modalities and the adoption of new methods that can detect multivariate brain-behavior relationships. We created and tested a multimodal framework that relies on three information sources (lesion maps, structural connectivity, and functional connectivity) to create an array of unimodal predictions which are then fed into a final model that creates "stacked multimodal predictions" (STAMP). Crossvalidated predictions of four aphasia scores (picture naming, sentence repetition, sentence comprehension, and overall aphasia severity) were obtained from 53 left hemispheric chronic stroke patients (age: 57.1 ± 12.3 yrs, post-stroke interval: 20 months, 25 female). Results showed accurate predictions for all four aphasia scores (correlation true vs. predicted: r = 0.79-0.88). The accuracy was slightly smaller but yet significant (r = 0.66) in a full split crossvalidation with each patient considered as new. Critically, multimodal predictions produced more accurate results that any single modality alone. Topological maps of the brain regions involved in the prediction were recovered and compared with traditional voxel-based lesion-to-symptom maps, revealing high spatial congruency. These results suggest that neuroimaging modalities carry complementary information potentially useful for the prediction of aphasia scores. More broadly, this study shows that the translation of neuroimaging findings into clinically useful tools calls for a shift in perspective from unimodal to multimodal neuroimaging, from univariate to multivariate methods, from linear to nonlinear models, and, conceptually, from inferential to predictive brain mapping. Hum Brain Mapp 38:5603-5615, 2017. © 2017 Wiley Periodicals, Inc.