Continual Learning for Domain Adaptation in Chest X-ray Classification

Abstract

Over the last years, Deep Learning has been successfully applied to a broad range of medical applications. Especially in the context of chest X-ray classification, results have been reported which are on par, or even superior to experienced radiologists. Despite this success in controlled experimental environments, it has been noted that the ability of Deep Learning models to generalize to data from a new domain (with potentially different tasks) is often limited. In order to address this challenge, we investigate techniques from the field of Continual Learning (CL) including Joint Training (JT), Elastic Weight Consolidation (EWC) and Learning Without Forgetting (LWF). Using the ChestX-ray14 and the MIMIC-CXR datasets, we demonstrate empirically that these methods provide promising options to improve the performance of Deep Learning models on a target domain and to mitigate effectively catastrophic forgetting for the source domain. To this end, the best overall performance was obtained using JT, while for LWF competitive results could be achieved -even without accessing data from the source domain.

Deep Learning for Chest X-ray Analysis: A Survey

“…The authors investigated both unsupervised and semi-supervised approaches in this work, where some labels from the target domain were available. Another work by Lenga et al (2020) studied several recently proposed continual learning approaches, namely joint training, elastic weight consolidation and learning without forgetting, to improve the performance on a target domain and to mitigate effectively catastrophic forgetting for the source domain. The authors evaluated these methods for 2 publicly available datasets, ChestX-ray14 and MIMIC-CXR, for a multi-class abnormality classification task and demonstrated that joint training achieved the best performance.…”

Chest X-Ray Investigation: A Convolutional Neural Network Approach

“…Dunnmon et al (Dunnmon et al,20) have proposed an automated a binary classification for enabling a high-performance of chest radiographs with CNNs machine learning approach. Authors have trained CNNs for classifying chest radiographs (amid 1998 and 2012) of the dataset before evaluation with a set of 533 images.…”

Continual Class Incremental Learning for CT Thoracic Segmentation

“…For domain incremental learning, Ozgung et al [20] proposed learning rate regularisation to Memory Aware Synapses [1] to perform MRI brain segmentation. Lenga et al have shown that LwF outperforms elastic weight consolidation (EwC) [11], when applied to incremental X-ray domain learning [15].…”

“…Learning without Forgetting (LwF) [17], which combines both regularisation-based and pseudo-rehearsal techniques, has been the stateof-the-art medical imaging continual learning method for situations in which access to previous training data is not possible. It has shown promising results in medical imaging for both incremental domain learning [15] and incremental class learning [19]. However, it has not been previously evaluated on the incremental class learning problem with a task sequence exceeding two tasks.…”