Differential White Blood Cell Counting

This thesis was completed with the goal of developing a model for the detection and differentiation of 8 different classes of white blood cells. Manual differential counting can be a time consuming and laborious task that would greatly be improved by an automated system. The model created in this project was to be used to automatically determine the number of each cell type in a sample, greatly reducing the need for manual counting. The model was produced using a convolutional neural network that utilized transfer learning from MobileNet_V2, an image recognition network produced by Google. After 20 epochs of training both the classification head and the feature acquisition sections of the model, an accuracy of 94 percent was achieved. This project made use of a dataset published in the journal Computer Methods and Programs in Biomedicine for training