Automatic Quantification of COVID-19 Pulmonary Edema by Self-supervised Contrastive Learning

Zhaohui Liang, Zhiyun Xue, Sivaramakrishnan Rajaraman, Feng Yang, Sameer K. Antani

Abstract: We proposed a self-supervised machine learning method to automatically rate the severity of pulmonary edema in the frontal chest X-ray radiographs (CXR) which could be potentially related to COVID-19 viral pneumonia. For this we use the modified radiographic assessment of lung edema (mRALE) scoring system. The new model was first optimized with the simple Siamese network (SimSiam) architecture where a ResNet-50 pretrained by ImageNet database was used as the backbone. The encoder projected a 2048-dimension embedding as representation features to a downstream fully connected deep neural network for mRALE score prediction. A 5-fold cross-validation with 2,599 frontal CXRs was used to examine the new model’s performance with comparison to a non-pretrained SimSiam encoder and a ResNet-50 trained from scratch. The mean absolute error (MAE) of the new model is 5.05 (95%CI 5.03–5.08), the mean squared error (MSE) is 66.67 (95%CI 66.29–67.06), and the Spearman's correlation coefficient (Spearman ρ) to the expert-annotated scores is 0.77 (95%CI 0.75–0.79). All the performance metrics of the new model are superior to the two comparators (P < 0.01), and the scores of MSE and Spearman ρ of the two comparators have no statistical difference (P > 0.05). The model also achieved a prediction probability concordance of 0.811 and a quadratic weighted kappa of 0.739 with the medical expert annotations in external validation. We conclude that the self-supervised contrastive learning method is an effective strategy for mRALE automated scoring. It provides a new approach to improve machine learning performance and minimize the expert knowledge involvement in quantitative medical image pattern learning.

External IDs:dblp:conf/miccai/LiangXRFA23