Ensemble Learning of Convolutional Neural Network, Support Vector Machine, and Best Linear Unbiased Predictor for Brain Age Prediction: ARAMIS Contribution to the Predictive Analytics Competition 2019 Challenge

Baptiste Couvy-Duchesne,Olivier Colliot,Elina Thibeau–Sutre,Ninon Burgos,Adam Wild,Manon Ansart,Stanley Durrleman,Didier Dormont,Johann Faouzi,Benoît Martin

doi:10.3389/fpsyt.2020.593336

Abstract

We ranked third in the Predictive Analytics Competition (PAC) 2019 challenge by achieving a mean absolute error (MAE) of 3.33 years in predicting age from T1-weighted MRI brain images. Our approach combined seven algorithms that allow generating predictions when the number of features exceeds the number of observations, in particular, two versions of best linear unbiased predictor (BLUP), support vector machine (SVM), two shallow convolutional neural networks (CNNs), and the famous ResNet and Inception V1. Ensemble learning was derived from estimating weights via linear regression in a hold-out subset of the training sample. We further evaluated and identified factors that could influence prediction accuracy: choice of algorithm, ensemble learning, and features used as input/MRI image processing. Our prediction error was correlated with age, and absolute error was greater for older participants, suggesting to increase the training sample for this subgroup. Our results may be used to guide researchers to build age predictors on healthy individuals, which can be used in research and in the clinics as non-specific predictors of disease status.

Highlights

Chronological age is an important risk factor for many conditions such as neurological disorders (e.g., Alzheimer’s and Parkinson’s), chronic disorders, cancer, or stroke, to name a few
Influence of the Type of Brain Features on Prediction Accuracy We investigated the impact of the input features by training the best linear unbiased predictor (BLUP) and support vector machine (SVM) models on the gray matter maps, in replacement of the vertex-wise surfaces used previously
For the first challenge of minimizing the mean absolute error (MAE), the deep learning models performed significantly better than BLUP or SVM (p-value < 3.1E−4, paired t-test) with a MAE between 3.82 (Inception) and 4.18 years, compared with a MAE >4.90 years (BLUP-quantiles, Table 1)

Summary

Introduction

Chronological age is an important risk factor for many conditions such as neurological disorders (e.g., Alzheimer’s and Parkinson’s), chronic (including cardiovascular) disorders, cancer, or stroke, to name a few. It is an imperfect predictor of disease risk or of healthy individuals’ functional capability [1]. The interested reader may refer to Le et al [15] and Smith et al [16] for further discussion on PAD analyses and possible pitfalls Overall, these results indicate that brain age is associated with disorders, mortality, and function beyond what can be explained by chronological age. Combining brain age and methylation age [21] resulted in an increased prediction of the mortality risk, suggesting that brain age and the epigenetic clock capture different mechanisms of aging [6]

Methods

Results

Conclusion