Ensemble convolutional neural networks for cell classification in microscopic images

Abstract

In this challenge, we address the cell classification problem using deep convolutional neural networks (CNNs). For a better generalization of the CNN classifier, various data augmentation and preprocessing were tested and an ensemble of state-of-the-art CNNs was adopted. In addition, to check the stability of the CNN model, the Grad-CAM technique was used to visualize the most discriminative part of each cell when predicting the category of the cell image. Our model achieves an accuracy of 86.9% in the preliminary testing and 87.9% in the final testing.