Multi-View Ensemble Classification of Brain Connectivity Images for Neurodegeneration Type Discrimination

Abstract

Brain connectivity analyses using voxels as features are not robust enough for single-patient classification because of the inter-subject anatomical and functional variability. To construct more robust features, voxels can be aggregated into clusters that are maximally coherent across subjects. Moreover, combining multi-modal neuroimaging and multi-view data integration techniques allows generating multiple independent connectivity features for the same patient. Structural and functional connectivity features were extracted from multi-modal MRI images with a clustering technique, and used for the multi-view classification of different phenotypes of neurodegeneration by an ensemble learning method (random forest). Two different multi-view models (intermediate and late data integration) were trained on, and tested for the classification of, individual whole-brain default-mode network (DMN) and fractional anisotropy (FA) maps, from 41 amyotrophic lateral sclerosis (ALS) patients, 37 Parkinson’s disease (PD) patients and 43 healthy control (HC) subjects. Both multi-view data models exhibited ensemble classification accuracies significantly above chance. In ALS patients, multi-view models exhibited the best performances (intermediate: 82.9%, late: 80.5% correct classification) and were more discriminative than each single-view model. In PD patients and controls, multi-view models’ performances were lower (PD: 59.5%, 62.2%; HC: 56.8%, 59.1%) but higher than at least one single-view model. Training the models only on patients, produced more than 85% patients correctly discriminated as ALS or PD type and maximal performances for multi-view models. These results highlight the potentials of mining complementary information from the integration of multiple data views in the classification of connectivity patterns from multi-modal brain images in the study of neurodegenerative diseases.

Linking structural and effective brain connectivity: structurally informed Parametric Empirical Bayes (si-PEB)

“…Measures of white-matter integrity have been associated to altered functional connectivity in patients with amyotrophic lateral sclerosis (Douaud et al 2011 ), high functioning autism (Mueller et al 2013 ), schizophrenia (Schlösser et al 2007 ; Pomarol-Clotet et al 2010 ; Motzkin et al 2011 ), bipolar disorder (Motzkin et al 2011 ), anorexia and bulimia nervosa (Frank et al 2016 ) and temporal lobe epilepsy (Voets et al 2009 ). Given the volume of data afforded by multimodal imaging, computational approaches towards integration of structure and function appear indispensable to optimally inform diagnosis and management, as demonstrated by connectivity-based classification of patients with movement disorders (Fratello et al 2017 ) and cognitive decline (Pineda-Pardo et al 2014 ). Overall, such approaches may afford a more comprehensive understanding of the subtle alterations underlying pathologies of brain structure and function in neurological and psychiatric conditions.…”

Ant genera identification using an ensemble of convolutional neural networks

“…Although the main methods to deal with this problem are co-training, multiple kernel learning and subspace learning, there are some similarities with ensemble learning [ 15 ]. Still, ensemble is applied to multi-view data by aggregating the votes of random forests trained to each view [ 16 ]. Multi-view CNN approaches involve appending the multiple views as channels [ 17 ], using a fine-tune ImageNet-trained CNN for each view, for concatenating the penultimate layer output for each view, training a classifier on top of it [ 18 ] and simultaneously training a CNN for which the initial layers are view-specific convolutional layers [ 19 ].…”

Stacked penalized logistic regression for selecting views in multi-view learning

“…genomics, transcriptomics, proteomics, metabolomics) [3], but also as the same profiling data summarized at different levels [4], or as different gene sets or genetic pathways [5]. In neuroimaging, views may present themselves as different MRI modalities, such as functional MRI and diffusion-weighted MRI [6], but also as different feature sets computed from the same structural image [7]. As there is growing interest in integrating multi-omics and imaging data with other sources of information -electronic health records, patient databases, and even social media, wearables, and games -the abundance of multi-view data in biomedical research can only be expected to increase [8,9].…”