CHEST X-RAY CLASSIFICATION USING SELF- SUPERVISED LEARNING

Abstract

Deep learning models have created a tremendous impact in a medical image classification problems, especially in the case of
chest radiographs. So far, the pathologies in Chest X-Ray images were classified largely using the supervised methodology in
which the model learns from both the data and the corresponding labels available. These models require large amount of data
in order to perform significantly well and it is very difficult to create large datasets of images with labels particularly medical
images like Chest X-Rays. But in recent times, models trained using an unsupervised learning mechanism called self-supervised
learning has been performing on par with models trained with the help of supervised learning, where you don’t need a large
labeled dataset to train a good model. In this paper, shows how models that are trained with self-supervised mechanism on large
unlabeled Chest X-Ray images outperform models which are trained with the help of supervised transfer learning in the
classification task