Multi-Task Pre-Training of Deep Neural Networks for Digital Pathology

In this work, we investigate multi-task learning as a way of pre-training models for classification tasks in digital pathology. It is motivated by the fact that many small and medium-size datasets have been released community over years whereas there no large scale dataset similar to ImageNet domain. We first assemble transform pathology into pool 22 almost 900k images. Then, propose simple architecture training scheme creating transferable model robust evaluation selection protocol order evaluate our method. Depending on target task, show used feature extractors either improve significantly pre-trained or provide comparable performance. Fine-tuning improves performance extraction able recover lack specificity features, both sources yield