Implementing Transfer Learning for Mitotic Cell Detection

Abstract

The primary objective of this project is to use a neural network (deep learning) model that will be trained on datasets of human cells in mitosis and then apply it on animal cells using transfer learning techniques. The purpose of this project is to speed up the process for the pathologist to detect mitotic activity for cancer diagnosis and determine transfer learning techniques that will aid in predicting mitotic activity in animal cells. This information will prove to be useful for correct diagnosis of cancer and reduce potential errors of detecting mitotic cells. Instead of acquiring the highly meticulous task of the labelled data, transfer learning can produce promising results by reusing the learned features from human datasets and applying the learned knowledge to the datasets of the animal cells. This project proposes to develop a higher detection accuracy with our framework of transfer learning and optimize the deep learning model to produce the desired output of detecting mitotic cells in animals with the limited amount of training data. Transfer learning techniques will be aimed to improve the performance of learning in other domains and reduce the cost of the expensive labeled data.