Using GAN Augment CNN performance in Acute Leukemia Classification

Generative Adversarial Networks (GAN) is a type of artificial neural network and deep learning used to synthesize images. In recent years, many studies have applied GAN in the field of medical imaging. This study uses Convolutional Neural Networks (CNN) to predict the classification of acute leukemia and compares the accuracy of CNN model classification prediction before and after the use of GAN data proliferation. After using the original data to generate images with 8 different parameter combinations, twodimensional images using CD3 and CD7 were used to obtain a classification accuracy of 77.4% without GAN data proliferation. Following GAN data proliferation to add images for which the original data were insufficient, CNN classification prediction accuracy increased to 84.7%. This indicates that using GAN to fill gaps for medical diagnostic data can effectively provide the large volume of data required for CNN model training.