Generative Adversarial Network with Masking Bits Based Image Augmentation Technique for Nuclei Image Classification

In the domain of medical image analytics, especially in nuclei cell image analysis, the major challenges associated with it is the availability of smaller amount of interpreted samples of images and the inability to construct a machine learning model using deep neural network. The classification performance of the deep neural network model trained with smaller number of images is very low. To improve the classification performance of the model, the dataset size is increased by using the Generative Adversarial Network (GAN) with the integration of masking bits. It generates the images that remain more similar to the original nuclei cell images present in the dataset with various logical operations along with masking bits. Moreover, GAN contains two layers such as generator layer and the discriminator layer. The generator layer takes the image as an input and produces the images that remain most comparable to the unique Nuclei images. The discriminator layer takes input as original data and the generated Nuclei cell images from the generator to predict whether the generator nuclei cell images belong to original dataset or not. Finally, the classification performance of the machine learning model is measured by using the convolution neural network (CNN) architecture.