Pleural Effusion Classification on Chest X-Ray Images with Contrastive Learning

Felipe André Zeiser, Ismael Santos, Henrique Bohn, Cristiano André da Costa, Gabriel de Oliveira Ramos, Rodrigo da Rosa Righi, Andreas Maier, José Rodrigo M. Andrade, Alexandre Bacelar

Abstract: Diagnosing pleural effusion is important to recognize the disease’s etiology and reduce the length of hospital stay for patients after fluid content analysis. In this context, machine learning techniques have been increasingly used to help physicians identify radiological findings. In this work, we propose using contrastive learning to classify chest X-rays with and without pleural effusion. A model based on contrastive learning is trained to extract discriminative features from the images and reports to maximize the similarity between the correct image and text pairs. Preliminary results show that the proposed approach is promising, achieving an AUC of 0.900, an accuracy of 86.28%, and a sensitivity of 88.54% for classifying pleural effusion on chest X-rays. These results demonstrate that the proposed method achieves comparable or superior to state of the art results. Using contrastive learning can be a promising alternative to improve the accuracy of medical image classification mo