ABSTRACT
In this chapter, it is discussed how deep transfer learning strategies can be utilised as a way of overcoming automated diagnosis challenges of tumours encountered in the digital pathology field. Through several experiments, two strategies of deep transfer learning for image classification are investigated. The first strategy consists of using a network initialised with pre-trained weights and partially retraining it on a new dataset while the second strategy uses features extracted from pre-trained convolutional network model without retraining the network and use them for training a third-party classifier. The experiments aimed to compare the performance of different convolutional networks based on deep transfer learning strategies for classifying Squamous Cell Carcinoma tumours of the head and neck cancer from histopathological images.
