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

View on arXiv ← Back to list

Authors: Christopher Sims

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

AI Summary

This paper proposes a multimodal 3D convolutional neural network (CNN) for classifying ADHD from fMRI data, incorporating data augmentation from a 3D generative adversarial network (GAN) to improve accuracy. The approach compares single-modal and multimodal models using LSTM and GRU recurrent neural networks.

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

The GRU-based model outperformed the LSTM in single-modality classification. The multimodal model, incorporating GAN-generated MRI data, achieved significantly higher accuracy (over 95%) compared to single-modality and previous methods. The study suggests that the general 3D brain structure is crucial for ADHD classification.

Approach

The authors utilize a multimodal 3D CNN architecture, combining fMRI and GAN-generated MRI data. Time-distributed layers preserve temporal information in fMRI data, and LSTM and GRU RNNs handle the time series aspect. The model is trained and validated on the ADHD-200 dataset.

Datasets

ADHD-200 dataset (preprocessed with Athena pipeline), including Peking, KKI, Neuroimage, NYU, OHSU, Pittsburgh, and WashingtonU datasets.

Model(s)

3D Convolutional Neural Networks (CNNs) with time-distributed layers; Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) recurrent neural networks; 3D Generative Adversarial Network (GAN) for data augmentation.

Author countries

USA

← Previous