Automated CTC Identification Using improved GAN Model

This work proposed a novel Generative Adversarial Network (GAN) methodology to identify the Circulating Tumor Cells (CTCs) using a medical image. The proposed GAN model integrates a modified Pyramid Scene Parsing Network (PSPNet) as a generator and an EfficientNet as a discriminator. The modified PSPNet has an attention mechanism to process an efficient feature extraction that is used to capture intricate spatial data from CTC images. Also on the discriminator side, it has a regulator unit in EfficientNet named RegEfficientNet that is introduced for effective CTCs detection. This discriminator is processed to distinguish between genuine CTCs and false positives effectively to attain a valuable component in the GAN architecture. To achieve an optimal solution in result, Modified Siberian Tiger Optimization (MSTO) is used to optimize and fine-tune the GAN model's parameters. The Experimental results show that the proposed GAN model achieves better in terms of all metrics namely accuracy, sensitivity, F1 score and specificity than conventional methods respectively.