Ensemble Deep Learning on Large, Mixed-Site fMRI Datasets in Autism and Other Tasks

Abstract

Deep learning models for MRI classification face two recurring problems: they are typically limited by low sample size, and are abstracted by their own complexity (the "black box problem"). In this paper, we train a convolutional neural network (CNN) with the largest multi-source, functional MRI (fMRI) connectomic dataset ever compiled, consisting of 43,858 datapoints. We apply this model to a crosssectional comparison of autism (ASD) vs typically developing (TD) controls that has proved difficult to characterise with inferential statistics. To contextualise these findings, we additionally perform classifications of gender and task vs rest. Employing class-balancing to build a training set, we trained 3×300 modified CNNs in an ensemble model to classify fMRI connectivity matrices with overall AUROCs of 0.6774, 0.7680, and 0.9222 for ASD vs TD, gender, and task vs rest, respectively. Additionally, we aim to address the black box problem in this context using two visualization methods. First, class activation maps show which functional connections of the brain our models focus on when performing classification. Second, by analyzing maximal activations of the hidden layers, we were also able to explore how the model organizes a large and mixed-centre dataset, finding that it dedicates specific areas of its hidden layers to processing different covariates of data (depending on the independent variable analyzed), and other areas to mix data from different sources. Our study finds that deep learning models that distinguish ASD from TD controls focus broadly on temporal and cerebellar connections, with a particularly high focus on the right caudate nucleus and paracentral sulcus.

Domain Adaptation Using a Three-Way Decision Improves the Identification of Autism Patients from Multisite fMRI Data

“…For example, Lu et al [ 65 ] proposed a multi-kernel-based subspace clustering algorithm for identifying ASD patients, which still has a good clustering effect on high-dimensional network datasets. Leming et al [ 66 ] trained a convolutional neural network and applied it to ASD recognition, and their experiments showed that deep learning models that distinguish ASD from NC controls focus broadly on temporal and cerebellar connections. However, the problem of small size fMRI data prevented the generalization of the above research works [ 67 ].…”

Beyond linear regression: mapping models in cognitive neuroscience should align with research goals

“…A number of studies have leveraged the power of ML methods to build flexible nonlinear mapping models and use them to identify neural correlates of brain disorders (e.g., Hasanzadeh et al, 2019;Kazemi & Houghten, 2018;Kim et al, 2016;Leming et al, 2020) and behavioral traits (e.g., Kumar et al, 2019;Morioka et al, 2020;Xiao et al, 2019). Yet the vast majority of cognitive neuroscience studies use linear mapping models (such as linear regression), resulting in a gap between different neuroscience subfields.…”

Basic Statistical Estimation Outperforms Machine Learning in Monthly Prediction of Seasonal Climatic Parameters

“…These 12 feature set variants were applied to both the raw and de-seasonalized training data. [89][90][91][92]. For all ML tools except for MLR, parameters were optimized via grid search with three fold validation, using the time series cross-validator implemented in scikit-learn [93].…”