Figure 4.

Hybrid model to distinguish CVD from non-CVD using Retinal Images and DXA tabular data. CNN stem for Retinal Image was based on ResNet-34 architecture [68]. The MLP stem includes Linear (Lin), Batch Normalization (BN), Dropout (Dr) layers as shown in the diagram. Both stems were integrated, and their output was fed into the Classification Head having multiple layers of BN, Dr, Lin, ReLU, BN, Dr, and finally a single linear layer as output layer (CVD or non-CVD).