Parasitic Egg Detection In Microscopic Images: ICIP 2022 Challenge.pdf

The dataset, Chula-ParasiteEgg-11, contains 11 types of parasitic eggs from faecal smear samples, with a total of 13,750 microscopic images, making it the largest dataset of its kind. The training and testing datasets are randomly divided, with 1,000 images/class for training and approximately 250 images/class for testing. Some testing images contain two or more types of parasitic eggs, and some testing images do not have any eggs. The labels take the form of bounding boxes. The visual quality of some images is degraded to replicate the real scenario, acquired with speed or automated system in different lighting conditions. The average size of the parasitic eggs ranges approximately in the interval of 15-140 µm, with details of size and diameter listed in Table 1. The training and testing datasets are available on the IEEE Dataport. The stool samples of parasitic eggs used for the experiments were prepared in laboratory conditions at the Department of Clinical Microscopy, Faculty of Allied Health Sciences, Chulalongkorn University. The specimens were fixed with 10% formalin prior to storing at 4°C. Stool samples were examined microscopically by experienced laboratory technologists. The dataset exhibits real and synthetic degradation due to out-of-focus, noise, motion blur, colour, and light imbalance, aiming to enhance the robustness of the detection models. The challenge has gathered experts in the fields of image processing, medical imaging, and computer vision, which should not be limited to the domain of microscopic imaging as knowledge transfer can benefit the growth of the community. The dataset has been downloaded from more than 20 countries in 4 continents. The outcome of the challenge could further improve and assist diagnosis in real clinical use, and even automate detection and identification of intestinal parasite eggs, which can be used by non-experts.