Simple 1-D Convolutional Networks for Resting-State fMRI Based Classification in Autism

Abstract

Deep learning methods are increasingly being used with neuroimaging data like structural and function magnetic resonance imaging (MRI) to predict the diagnosis of neuropsychiatric and neurological disorders. For psychiatric disorders in particular, it is believed that one of the most promising modality is the resting-state functional MRI (rsfMRI), which captures the intrinsic connectivity between regions in the brain. Because rsfMRI data points are inherently high-dimensional (∼1M), it is impossible to process the entire input in its raw form. In this paper, we propose a very simple transformation of the rsfMRI images that captures all of the temporal dynamics of the signal but sub-samples its spatial extent. As a result, we use a very simple 1-D convolutional network which is fast to train, requires minimal preprocessing and performs at par with the state-of-theart on the classification of Autism spectrum disorders.

A Hybrid 3DCNN and 3DC-LSTM Based Model for 4D Spatio-Temporal fMRI Data: An ABIDE Autism Classification Study

“…Model Accuracy F1-score NYU AE MLP [10] 0.64 ± 0.1 0.67 SVM [7] 0.6 ± 0.13 0.59 1D Conv [9] 0.64 ± 0.11 0.62 CNN3D TC* [5] 0.57 0.61 CNN3D MS* [5] 0.60 0.65 convGRU-CNN3D* [5] 0.67 0.71 CNN4D* [5] 0.60 0.68 3DCNN 1D (ours) 0.59 ± 0.07 0.58 3DCNN C-LSTM (ours) 0.77±0.05 0.78 UM AE MLP [10] 0.56 ± 0.11 0.59 SVM [7] 0.54 ± 0.11 0.56 1D Conv [9] 0.63 ± 0.1 0.62 3DCNN 1D (ours) 0.66 ± 0.09 0.58 3DCNN C-LSTM (ours) 0.71±0.06 0.70 ABIDE-I AE MLP [10] 0.63 ± 0.02 0.64 SVM [7] 0.58 ± 0.04 0.6 1D Conv [9] 0.64±0.06 0.64 3DCNN 1D (ours) 0.54 ± 0.02 0.50 3DCNN C-LSTM (ours) 0.58 ± 0.03 0.53 * Results as reported by Bengs et al [5] on NYU data.…”

Machine Learning Methods for Diagnosing Autism Spectrum Disorder and Attention- Deficit/Hyperactivity Disorder Using Functional and Structural MRI: A Survey

“…They investigated 6 different approaches, such as repeating phenotypic information along the time dimension, concatenating it to the time series and feeding it to the network, or feeding the phenotypic data and the final output of LSTM to the dense layer. CNN networks are also used in different studies for diagnosing autism (Brown et al, 2018 ; Khosla et al, 2018 ; Li G. et al, 2018 ; Parisot et al, 2018 ; Anirudh and Thiagarajan, 2019 ; El-Gazzar et al, 2019a , b ). Khosla et al ( 2018 ) proposed a multi-channel CNN network in which each channel represents the connectivity of each voxel with specific regions of interest.…”

Classifying Autism Spectrum Disorder Using the Temporal Statistics of Resting-State Functional MRI Data With 3D Convolutional Neural Networks

“…The preprocessing was done using the Configurable Pipeline for the Analysis of Connectomes (C-PAC, (12). We followed a preprocessing strategy adopted by the Preprocessed Connectome Project initiative 3 . This will allow others to replicate and extend the findings in this paper.…”

“…In previous work of our group, we maintained the time dimension while summarizing the spatial dimension using ROIs (Harvard Oxford atlas). This approach obtained an accuracy of 68% using a simple 1D-CNN model on the ABIDE I+II dataset (3).…”

“…In this approach, about one million voxels in space are locally averaged in about 100 to 400 different brain regions depending on the atlas, thus considerably reducing the dimension of the input data. These 100 to 400 averaged time courses are then either used in a 1-D convolutional network (3) or are first used to calculate a crosscorrelation matrix which subsequently is used as a feature in various machine learning methods (4,5). For example, (4) reached an accuracy of 70% on ABIDE-I using the cross-correlation matrix on time courses extracted using the CC400 atlas.…”