¿µ¹®³»¿ë (English Abstract) |
Semi-supervised learning has become a popular topic in machine learning, since it only requires a few amounts of labelled data. Recently, it has been widely applied to medical imaging to overcome the shortage of labelled data in this task. To compensate this limitation of medical images, semi-supervised learning offers the use of unlabeled data and small amount of labelled data to learn the representations of the input. This paper studies the performance of semi-supervised learning on Ultra-wide-field Images. These images are the scans of the eyes that capture a wide range of human retinal including macula, fovea and optic disc. Ophthalmologists use these scans to detect diseases within the eyes and give diagnosis for patients. We summaries and carry out a popular semi-supervised learning for the well-known chest X-ray dataset and promising result of 0.648 mean AUROC score even though we use a much smaller amount of unlabeled dataset(2000 unlabeled images). Through this work, we show that semi-supervised learning could be one of the go-to method when it comes to the task of analyzing Ultra-wide-filed Images. |