Deep neural networks have been shown to be very powerful for image recognition. They have provided significant increases in accuracy when applied to color images for object recognition using the ImageNet corpus. They have worked effectively when trained on big data, a million or more camera images and as many as 1000 classes of objects. This talk examines using pre-trained deep neural networks in medical imaging. How to leverage pre-trained large convolutional neural networks to extract features to enable predictions of 'disease' prognosis for two types of cancer will be shown. Results for CT images of lung nodules/tumors and MR images of brain cancer will be discussed. Transfer learning through the extraction of deep features will be shown to be a promising feature selection strategy, enabling the most accurate prognosis predictions on the data sets to be discussed.