Advancing Medical Imaging Informatics by Deep Learning-Based Domain Adaptation

Abstract

Introduction: There has been a rapid development of deep learning (DL) models for medical imaging. However, DL requires a large labeled dataset for training the models. Getting large-scale labeled data remains a challenge, and multi-center datasets suffer from heterogeneity due to patient diversity and varying imaging protocols. Domain adaptation (DA) has been developed to transfer the knowledge from a labeled data domain to a related but unlabeled domain in either image space or feature space. DA is a type of transfer learning (TL) that can improve the performance of models when applied to multiple different datasets. Objective: In this survey, we review the state-of-the-art DL-based DA methods for medical imaging. We aim to summarize recent advances, highlighting the motivation, challenges, and opportunities, and to discuss promising directions for future work in DA for medical imaging. Methods: We surveyed peer-reviewed publications from leading biomedical journals and conferences between 2017-2020, that reported the use of DA in medical imaging applications, grouping them by methodology, image modality, and learning scenarios. Results: We mainly focused on pathology and radiology as application areas. Among various DA approaches, we discussed domain transformation (DT) and latent feature-space transformation (LFST). We highlighted the role of unsupervised DA in image segmentation and described opportunities for future development. Conclusion: DA has emerged as a promising solution to deal with the lack of annotated training data. Using adversarial techniques, unsupervised DA has achieved good performance, especially for segmentation tasks. Opportunities include domain transferability, multi-modal DA, and applications that benefit from synthetic data.

Correcting data imbalance for semi-supervised COVID-19 detection using X-ray chest images

“…As stated in [24] , using the target dataset is vital for training a model, as using a different source dataset from other hospitals/clinics to train the model might yield poor test performance in the target dataset. Such distribution mismatch among different data sources is a frequent short-coming of deep learning solutions in the context of medical imaging.…”

“…Furthermore, given the different patient ethnicity’s and characteristics, along with varying imaging protocols, using a model trained with data from another set of hospitals or clinics (from possibly different countries) might yield a distribution mismatch between the training and test datasets. This possibly would yield a very low performance [23] , [24] . Therefore, training the model with data from the specific clinic/hospital where the model is intended to be used (target data), is an urgent task, which faces the challenge of dealing with very limited labelled datasets [23] , [24] , [25] .…”

“…This possibly would yield a very low performance [23] , [24] . Therefore, training the model with data from the specific clinic/hospital where the model is intended to be used (target data), is an urgent task, which faces the challenge of dealing with very limited labelled datasets [23] , [24] , [25] .…”

Application of Artificial Intelligence Technology in Oncology: Towards the Establishment of Precision Medicine

“…Domain adaptation has been proposed as one of the methods to solve the domain shift problem and is being actively studied [172,[176][177][178][179][180][181][182][183][184][185][186][187]. Domain adaptation is a type of transfer learning in which knowledge obtained from a domain with sufficient labeled training data (source domain) is applied to a target domain that lacks sufficient information (target domain) to learn things such as discriminators that work with high accuracy in the target domain (here, a domain is a collection of data).…”

Deep Learning Based Cardiac MRI Segmentation: Do We Need Experts?

“…For these reasons, the medical imaging literature has had an increasing number of publications, whose goal is to compensate for the lack of expert annotations [8]. While some methods leverage partly-annotated data sets [9], others use domain adaptation strategies to compensate for small training data sets [10]. Some other approaches artificially increase the number of annotated data with generative adversarial networks (GANs) [11,12], while others use third-party neural networks to help experts annotate images more rapidly [13].…”