Yes, it's possible. Create a new network in the same structure as the pre-trained one, but with one extra layer before the output layer. this should be added to the config file. The extra layer is type 'fc' with 2 output neurons (numFm=2).
Then, copy all the layers content from the pre-trained net to the new net (except for the added layer and output layer)
i.e
netNet.layers{1} = net.layers{1} ;
netNet.layers{2} = net.layers{2};
...
Afterwards, you can start training your new network using 'Train' function.