An Empirical Evaluation of a Novel Ensemble Deep Neural Network Model and Explainable AI for Accurate Segmentation and Classification of Ovarian Tumors Using CT Images

. 2024 Mar 4;14(5):543. doi: 10.3390/diagnostics14050543

Algorithm 1: Algorithm for Segmentation using UNet and Transformers

Input: X as the input image,

c_{i}

as the output of the ith convolutional layer in the UNet model,

f_{i}

as the ith feature map,

p_{i}

as the ith pooling layer,

u_{i}

as the ith up-convolutional (transposed convolutional) layer in the U-Net model,

t_{i}

as the output of the transformer model,

m_{i}

as the multi-head attention mechanism in the transformer model,

d_{i}

as the ith dense layer in the transformer model.
Output: Segmented Lesion
1. for i in range(N) do:
           E = sigma(Conv(E-1))
           D = sigma(Conv(D-1)
           S = E + D
           L_bce = sum(y_i × log(O_i) + (1 − y_i) × log(1 − O_i) for i in range(N))
           L_dice = (2 × TP)/((TP + FP) + (TP + FN))
           L_jaccard = TP/(TP + FP + FN)
end for
# Transformer-style segmentation with MultiHeadAttention, Position-wise Feed-forward Network, and Normalization
2. for i in range(N) do:
                     E = MultiHeadAttention_Layer(x) + x
                     D = MultiHeadAttention(Y) + Y + MultiHeadAttention(Encoder_Output)
           FFN = ReLU(Conv1D(Z, W_1 + b_1)) @ W_2 + b_2
end for