Unmasking Interstitial Lung Diseases: Leveraging Masked Autoencoders for Diagnosis

TL;DR

This study demonstrates that masked autoencoders trained on unlabelled chest CT scans can learn meaningful features and enhance diagnosis of diffused lung diseases, addressing data scarcity issues.

Contribution

The paper introduces a novel application of masked autoencoders for lung disease diagnosis, leveraging unlabelled data to improve feature extraction and classification accuracy.

Findings

MAEs effectively extract clinically relevant features

Improved diagnostic performance with MAE pretraining

Utilization of publicly available scans enhances model robustness

Abstract

Masked autoencoders (MAEs) have emerged as a powerful approach for pre-training on unlabelled data, capable of learning robust and informative feature representations. This is particularly advantageous in diffused lung disease research, where annotated imaging datasets are scarce. To leverage this, we train an MAE on a curated collection of over 5,000 chest computed tomography (CT) scans, combining in-house data with publicly available scans from related conditions that exhibit similar radiological patterns, such as COVID-19 and bacterial pneumonia. The pretrained MAE is then fine-tuned on a downstream classification task for diffused lung disease diagnosis. Our findings demonstrate that MAEs can effectively extract clinically meaningful features and improve diagnostic performance, even in the absence of large-scale labelled datasets. The code and the models are available here: https://github.com/eedack01/lung_masked_autoencoder.