SOSPCNN: Structurally Optimized Stochastic Pooling Convolutional Neural Network for Tetralogy of Fallot Recognition

Algorithm 2

Pseudocode of proposed 18-way DA on th training image

Input: import raw preprocessed th training image .

  geometric or photometric or noise-injection DA transforms are utilized on .

Step 1: we obtain . See Equation (12)

 Each enhanced dataset contains new images. See Equation (13).

Step 2: a horizontal mirror image is produced as . See Equation (14).

Step 3: -way data augmentation methods are carried out on ,

 We obtain . See Equation (15).

Step 4: , , , and are merged via . See Equation (16).

Output A new dataset is produced. The number of images is . See Equation (17).