Figure 6
Overview of the proposed combined CNN-MLP algorithm for brain age prediction. The CNN architecture, designed for minimally preprocessed T1-weighted images, consists of repeated convolutional blocks, each with 3D convolutional layers, batch normalization, ReLU activations, and max pooling. After these blocks, the sequence includes a flattening layer, two dense layers interspersed with ReLU activations, batch normalization, and a dropout layer. The MLP, tailored for categorical sex information, features dense layers with ReLU activations. Both algorithms' outputs are merged by a concatenation layer, processed through two dense layers, with the final layer using a linear activation for brain age prediction. Abbreviations: 3D, 3-dimensional; BatchNorm, batch normalization; CNN, convolutional neural network; Conv, Convolution; MaxPool, max pooling; MLP, multi-layer perceptron; ReLU, rectified linear unit.
