Table 1 Predictive accuracy of models employing the combined CNN-MLP algorithm and CNN-only algorithm.

From: Brain age prediction using combined deep convolutional neural network and multi-layer perceptron algorithms

 

Algorithm

Input

Performance metrics

MAE (years)

RMSE (years)

R2

10-fold cross validation

(Training set, n = 2,703)

Combined CNN-MLP

Minimally preprocessed whole brain T1-weighted image + sex information

3.494 ± 0.228

4.689 ± 0.570

0.933 ± 0.012

CNN-only

Minimally preprocessed whole brain T1-weighted image

3.563 ± 0.193

4.839 ± 0.299

0.932 ± 0.009

Internal validation

(Test set, n = 301)

Combined CNN-MLP

Minimally preprocessed

whole brain T1-weighted image + sex information

3.184

4.687

0.936

CNN-only

Minimally preprocessed whole brain T1-weighted image

3.342

4.659

0.937

External validation

(CamCAN set, n = 645)

Combined CNN-MLP

Minimally preprocessed whole brain T1-weighted image + sex information

4.910

6.148

0.891

CNN-only

Minimally preprocessed whole brain T1-weighted image

5.064

6.295

0.885

  1. The results were obtained following the application of hyperparameter tuning, utilizing Adam as the chosen optimizer. Performance metrics from 10-fold cross-validation are presented as mean ± standard deviation.
  2. Adam adaptive moment estimation, CamCAN Cambridge Centre for Ageing and Neuroscience, CNN convolutional neural network, MAE mean absolute error, MLP multi-layer perceptron, RMSE root mean squared error, R2 coefficient of determination.
Back to article page