Reference no: EM132292685
Creating a 3D DenseNet.
Information:
The first step in the assignment is to get a 3D densenet for classification working in Pytroch.
You will have to adapt its layers (conv, maxpool, etc..) to 3D, the number of output units to 2 (as we have a 2-class classification problem) and change the input size to (batch_size,1,512,512,256).
Please, remember that the length of the var block_config=(n,n,n,n,n) indicates how many times you will downsample from the original size.
In order to change the number of slices of each image to 256 (input size for 3rd dimension), you can probably use the attached function
The inputs size are 1000 CT images and the input size should be [512,512,256]
We will also need metrics to evaluate the algorithm and its training performance. The measurements we are interested (for training, validation and testing sets) are: loss function value, accuracy (for the whole dataset and also for each of the classes), sensitivity, specificity, positive predictive value and negative predictive value.
Train the model with all available images and plot:
Training loss
Training accuracy
training AUC (ROC Curve)
training positive predictive value
training negative predictive value
training sensitivity
training specificity
Attachment:- Assignment.rar