Parasitic Egg Detection with a Deep Learning Ensemble

Intestinal parasitic infections are a healthcare problem with a high impact in some areas. While nowadays the assessment performed by experts is mostly manual, it is possible to introduce machine learning techniques to help automating this task, or at least reduce the workload. This can lead to shorter detection times and faster treatment application. In the context of deep learning, several object detection techniques have been proposed and validated on general purpose datasets such as ImageNet or COCO. In this work, an ensemble of these is proposed for this particular task. The merged detections of FasterRCNN, TOOD, YOLOX and Cascade with Swin-Transformers applied to this problem achieved 0.915 IoU, larger than the result that each method obtained independently.