Accurate brain age prediction with lightweight deep neural networks.

Han Peng,Weikang Gong,Stephen M Smith,Christian F Beckmann,Andrea Vedaldi

doi:10.1016/j.media.2020.101871

Abstract

Deep learning has huge potential for accurate disease prediction with neuroimaging data, but the prediction performance is often limited by training-dataset size and computing memory requirements. To address this, we propose a deep convolutional neural network model, Simple Fully Convolutional Network (SFCN), for accurate prediction of brain age using T1-weighted structural MRI data. Compared with other popular deep network architectures, SFCN has fewer parameters, so is more compatible with small dataset size and 3D volume data. The network architecture was combined with several techniques for boosting performance, including data augmentation, pre-training, model regularization, model ensemble and prediction bias correction. We compared our overall SFCN approach with several widely-used machine learning models. It achieved state-of-the-art performance in UK Biobank data (N = 14,503), with mean absolute error (MAE) = 2.14y in brain age prediction and 99.5% in sex classification. SFCN also won (both parts of) the 2019 Predictive Analysis Challenge for brain age prediction, involving 79 competing teams (N = 2,638, MAE = 2.90y). We describe here the details of our approach, and its optimisation and validation. Our approach can easily be generalised to other tasks using different image modalities, and is released on GitHub.

Highlights

The emergence of machine learning techniques has made automatic disease prediction from medical imaging data possible
We proposed Simple Fully Convolutional Network (SFCN), a lightweight deep neural network architecture, which achieved state-of-the-art brain age prediction using T1-weighted structural magnetic resonance imaging (MRI) images
We investigated different approaches for boosting the performance of the deep learning model, and tested three factors that are valuable for improving the performance of a single deep learning model in a neuroimaging dataset: 1) the lightweight model structure, 2) data augmentation and regularisation techniques, 3) large dataset size

Summary

Introduction

The emergence of machine learning techniques has made automatic disease prediction from medical imaging data possible. Deep learning has had some successes, and yet faces several challenges (Arslan et al, 2018; Baumgartner et al, 2018; Cole et al, 2017; Kawahara et al, 2017). 3D neuroimaging data requires much more GPU memory than most 2D images, meaning that models successful in 2D data (e.g., ImageNet classification (Krizhevsky et al, 2012; Simonyan and Zisserman, 2014)) are infeasible in the 3D scenario. Deep networks usually require a large sample size for model fitting, but neuroimaging datasets often have relatively few samples compared to existing million-sample natural image datasets (Russakovsky et al, 2015), which could limit the ability to learn image features effectively, and result in overfitting problems.

Methods

Results

Conclusion