Algorithm 1. The 5-CNN model description

Input: thyroidal images of dimension (500 px, 500 px) from the train dataset.

Output: CNN model weights

1. for each image in the dataset

2. Resize image to (224 px, 224 px)

3. Normalize the image pixels values between [0, 1].

4. end

5. Add a first convolutional layer with a RELU activation function.

6. Add a second convolutional layer with a RELU activation function.

7. Apply a max pooling layer for down-sampling feature map from the previous layer.

8. Repeat the steps 4 and 5 for three times.

9. Add a Flatten layer on the output obtained from the last max-pooling layer.

10. Add a fully connected layer with 256 hidden units.

11. Apply a dropout for inactivate neurons in the previous layer.

12. Add a fully connected layer with 4 hidden units and a softmax activation function.

13. Optimize the model with RMSProp optimizer with a learning rate of 0.0001.

14. Train the model for 100 epochs.

15. Save the final model.