Local and Deep Features Based Convolutional Neural Network Frameworks for Brain MRI Anomaly Detection
Algorithm 1
Pseudocode of DBP-DAE.
Initialize mini-batch size, epochs number (EP), pretraining learning rate (PLR), number of layers (NL), dimension (D), total number of classes (C), and neurons in each hidden layer n[L]
(1)
Input DBP as a input feature Vector with D dimensions
(2)
For each layer (NL):
(3)
1< L < D
(4)
D-input and D-hidden
(5)
If (L = 1)
(6)
n [1] = D-hidden
(7)
D = D-input
(8)
Else
(9)
Dimension of visible layer n[L − 1]
(10)
Dimension of hidden layer n[L]
(11)
End
(12)
Initialize ,,
(13)
For each pretraining epoch
(14)
For each mini batch
(15)
Compute reconstruction
(16)
(17)
Compute Cost
(18)
(19)
Update ,,
(20)
End
(21)
Freeze reconstruction layer
(22)
End
(23)
Initial parameters of logistic regression layer
(24)
Input of classifier layer = n[D]
(25)
Output of classifier layer = C
(26)
For each fine-tuning
(27)
For each mini batch
(28)
Compute probability function of each class regarding to equation (6)
