Cycle-GAN based Data Augmentation to improve Faster-RCNN Generalizability to Detect Intestinal Parasites from Microscopy images

Abstract

Intestinal parasites are responsible for affecting millions of people in developing and underdeveloped countries, primarily diagnosed using traditional manual light microscopes but suffer from drawbacks such as highly expensive, time-consuming, and requiring specialized expertise. Recent advances in deep learning have shown potential for addressing these challenges. For that, labeled medical imaging data is required which is scarce and expensive to generate, posing a major challenge in developing generalized deep learning models that require substantial amounts of data. Here, we utilized the generative adversarial network to generate synthetic dataset and improved the performance of deep learning models. Our framework exploits the potential of Generative Adversarial Networks (CycleGANs) and Faster RCNN to generate new datasets and detect intestinal parasites, respectively, on images of varying quality, leading to improved model generalizability and diversity. In this experiment, we evaluated the effectiveness of Cycle Generative Adversarial Network (CycleGAN) + Faster RCNN, we employed widely-used evaluation metrics such as precision, recall, and F1-score. We demonstrated that the proposed framework effectively augmented the images dataset and improved the detection performance, with F1-Score of 0.98% and mIoU of 0.97% are achieved which is better than without data augmentation. We show that this state-of-the-art approach sets the stage for further advancements in the field of medical image analysis. Additionally, we have built a new dataset, which is now publicly accessible, offering a broader range of classes and variability for future research and development.