Highly Accurate FMRI ADHD Classification using time distributed multi modal 3D CNNs

Authors: Christopher Sims

Published: 2022-05-24 11:39:11+00:00

Comment: 8 pages, 5 Figures, 1 Table

AI Summary

This paper proposes an algorithm for ADHD classification from fMRI data using time-distributed multi-modal 3D CNNs. It leverages 3D Generative Adversarial Networks (GANs) to augment data with synthetic MRI images, comparing this approach against a single-modal 3D CNN. The aim is to enhance accuracy by incorporating structural brain information via 'deepfake' MRI data.

Abstract

This work proposes an algorithm for fMRI data analysis for the classification of ADHD disorders. There have been several breakthroughs in the analysis of fMRI via 3D convolutional neural networks (CNNs). With these new techniques it is possible to preserve the 3D spatial data of fMRI data. Additionally there have been recent advances in the use of 3D generative adversarial neural networks (GANs) for the generation of normal MRI data. This work utilizes multi modal 3D CNNs with data augmentation from 3D GAN for ADHD prediction from fMRI. By leveraging a 3D-GAN it would be possible to use deepfake data to enhance the accuracy of 3D CNN classification of brain disorders. A comparison will be made between a time distributed single modal 3D CNN model for classification and the modified multi modal model with MRI data as well.

Key findings

Multi-modal 3D CNNs, particularly with GRU recurrent layers and augmented by 3D GAN-generated MRI data, achieved significantly higher ADHD classification accuracy (up to 99.69%) compared to single-modal GRU models (75.55%). This indicates that leveraging general 3D brain structure from synthetic MRI is highly effective, preventing overfitting to specific subject brain structures and boosting classification performance.

Approach

The authors employ time-distributed 3D CNNs, incorporating either LSTM or GRU recurrent layers, to process fMRI data while preserving its temporal component for ADHD classification. A multi-modal extension integrates randomly generated 3D MRI data (synthesized by a 3D-GAN) alongside fMRI as an 'informer' of brain structure, comparing its performance against single-modal models.

Datasets

ADHD-200 dataset (including PEKING 1, KKI, NEUROIMAGE, NYU, OHSU, PITTSBURGH, WASHINGTONU)

Model(s)

Time-distributed 3D Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), 3D Generative Adversarial Networks (GANs) (specifically 3D-StyleGAN, an adaptation of StyleGAN2).

Author countries

USA

← Previous