Advancing Medical Imaging Informatics by Deep Learning-Based Domain Adaptation

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.

Download Full-text

The applications of deep neural networks to sdBV classification

Open Astronomy ◽

10.1515/astro-d-17-0450 ◽

2017 ◽

Vol 26 (1) ◽

Author(s):

Thomas M. Boudreaux

Keyword(s):

Neural Network ◽

Deep Learning ◽

Large Scale ◽

Feature Detection ◽

Synthetic Data ◽

Acoustic Mode ◽

Training Data ◽

Accurate Analysis ◽

Pulsating Stars ◽

High Speeds

AbstractWith several new large-scale surveys on the horizon, including LSST, TESS, ZTF, and Evryscope, faster and more accurate analysis methods will be required to adequately process the enormous amount of data produced. Deep learning, used in industry for years now, allows for advanced feature detection in minimally prepared datasets at very high speeds; however, despite the advantages of this method, its application to astrophysics has not yet been extensively explored. This dearth may be due to a lack of training data available to researchers. Here we generate synthetic data loosely mimicking the properties of acoustic mode pulsating stars and we show that two separate paradigms of deep learning - the Artificial Neural Network And the Convolutional Neural Network - can both be used to classify this synthetic data effectively. And that additionally this classification can be performed at relatively high levels of accuracy with minimal time spent adjusting network hyperparameters.

Download Full-text

Learning from Synthetic Dataset for Crop Seed Instance Segmentation

10.1101/866921 ◽

2019 ◽

Author(s):

Yosuke Toda ◽

Fumio Okura ◽

Jun Ito ◽

Satoshi Okada ◽

Toshinori Kinoshita ◽

...

Keyword(s):

Neural Network ◽

Deep Learning ◽

Large Scale ◽

Synthetic Data ◽

Seed Morphology ◽

Training Data ◽

Training Dataset ◽

Labor Costs ◽

Data Annotation ◽

Instance Segmentation

Incorporating deep learning in the image analysis pipeline has opened the possibility of introducing precision phenotyping in the field of agriculture. However, to train the neural network, a sufficient amount of training data must be prepared, which requires a time-consuming manual data annotation process that often becomes the limiting step. Here, we show that an instance segmentation neural network (Mask R-CNN) aimed to phenotype the barley seed morphology of various cultivars, can be sufficiently trained purely by a synthetically generated dataset. Our attempt is based on the concept of domain randomization, where a large amount of image is generated by randomly orienting the seed object to a virtual canvas. After training with such a dataset, performance based on recall and the average Precision of the real-world test dataset achieved 96% and 95%, respectively. Applying our pipeline enables extraction of morphological parameters at a large scale, enabling precise characterization of the natural variation of barley from a multivariate perspective. Importantly, we show that our approach is effective not only for barley seeds but also for various crops including rice, lettuce, oat, and wheat, and thus supporting the fact that the performance benefits of this technique is generic. We propose that constructing and utilizing such synthetic data can be a powerful method to alleviate human labor costs needed to prepare the training dataset for deep learning in the agricultural domain.

Download Full-text

Siamese Reconstruction Network: Accurate Image Reconstruction from Human Brain Activity by Learning to Compare

Applied Sciences ◽

10.3390/app9224749 ◽

2019 ◽

Vol 9 (22) ◽

pp. 4749

Author(s):

Lingyun Jiang ◽

Kai Qiao ◽

Linyuan Wang ◽

Chi Zhang ◽

Jian Chen ◽

...

Keyword(s):

Deep Learning ◽

Human Brain ◽

Brain Activity ◽

Feature Space ◽

Training Data ◽

Reconstruction Method ◽

Learning Method ◽

Training Samples ◽

Visual Reconstruction ◽

Relationship Of

Decoding human brain activities, especially reconstructing human visual stimuli via functional magnetic resonance imaging (fMRI), has gained increasing attention in recent years. However, the high dimensionality and small quantity of fMRI data impose restrictions on satisfactory reconstruction, especially for the reconstruction method with deep learning requiring huge amounts of labelled samples. When compared with the deep learning method, humans can recognize a new image because our human visual system is naturally capable of extracting features from any object and comparing them. Inspired by this visual mechanism, we introduced the mechanism of comparison into deep learning method to realize better visual reconstruction by making full use of each sample and the relationship of the sample pair by learning to compare. In this way, we proposed a Siamese reconstruction network (SRN) method. By using the SRN, we improved upon the satisfying results on two fMRI recording datasets, providing 72.5% accuracy on the digit dataset and 44.6% accuracy on the character dataset. Essentially, this manner can increase the training data about from n samples to 2n sample pairs, which takes full advantage of the limited quantity of training samples. The SRN learns to converge sample pairs of the same class or disperse sample pairs of different class in feature space.

Download Full-text

Data augmentation for computed tomography angiography via synthetic image generation and neural domain adaptation

Current Directions in Biomedical Engineering ◽

10.1515/cdbme-2020-0015 ◽

2020 ◽

Vol 6 (1) ◽

Author(s):

Malte Seemann ◽

Lennart Bargsten ◽

Alexander Schlaefer

Keyword(s):

Computed Tomography ◽

Neural Networks ◽

Deep Learning ◽

Medical Imaging ◽

Computed Tomography Angiography ◽

Data Augmentation ◽

Domain Adaptation ◽

Synthetic Image ◽

Wide Range ◽

The Impact

AbstractDeep learning methods produce promising results when applied to a wide range of medical imaging tasks, including segmentation of artery lumen in computed tomography angiography (CTA) data. However, to perform sufficiently, neural networks have to be trained on large amounts of high quality annotated data. In the realm of medical imaging, annotations are not only quite scarce but also often not entirely reliable. To tackle both challenges, we developed a two-step approach for generating realistic synthetic CTA data for the purpose of data augmentation. In the first step moderately realistic images are generated in a purely numerical fashion. In the second step these images are improved by applying neural domain adaptation. We evaluated the impact of synthetic data on lumen segmentation via convolutional neural networks (CNNs) by comparing resulting performances. Improvements of up to 5% in terms of Dice coefficient and 20% for Hausdorff distance represent a proof of concept that the proposed augmentation procedure can be used to enhance deep learning-based segmentation for artery lumen in CTA images.

Download Full-text

Classification of Very-High-Spatial-Resolution Aerial Images Based on Multiscale Features with Limited Semantic Information

Remote Sensing ◽

10.3390/rs13030364 ◽

2021 ◽

Vol 13 (3) ◽

pp. 364

Author(s):

Han Gao ◽

Jinhui Guo ◽

Peng Guo ◽

Xiuwan Chen

Keyword(s):

Deep Learning ◽

Land Cover ◽

Spatial Resolution ◽

Large Scale ◽

High Spatial Resolution ◽

Training Data ◽

Aerial Images ◽

Rural Landscapes ◽

Feature Representations ◽

Object Based

Recently, deep learning has become the most innovative trend for a variety of high-spatial-resolution remote sensing imaging applications. However, large-scale land cover classification via traditional convolutional neural networks (CNNs) with sliding windows is computationally expensive and produces coarse results. Additionally, although such supervised learning approaches have performed well, collecting and annotating datasets for every task are extremely laborious, especially for those fully supervised cases where the pixel-level ground-truth labels are dense. In this work, we propose a new object-oriented deep learning framework that leverages residual networks with different depths to learn adjacent feature representations by embedding a multibranch architecture in the deep learning pipeline. The idea is to exploit limited training data at different neighboring scales to make a tradeoff between weak semantics and strong feature representations for operational land cover mapping tasks. We draw from established geographic object-based image analysis (GEOBIA) as an auxiliary module to reduce the computational burden of spatial reasoning and optimize the classification boundaries. We evaluated the proposed approach on two subdecimeter-resolution datasets involving both urban and rural landscapes. It presented better classification accuracy (88.9%) compared to traditional object-based deep learning methods and achieves an excellent inference time (11.3 s/ha).

Download Full-text

A Physics-Infused Deep Learning Model for the Prediction of Refractive Indices and Its Use for the Large-Scale Screening of Organic Compound Space

10.26434/chemrxiv.8796950 ◽

2019 ◽

Author(s):

Mojtaba Haghighatlari ◽

Gaurav Vishwakarma ◽

Mohammad Atif Faiz Afzal ◽

Johannes Hachmann

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Large Scale ◽

Organic Molecules ◽

Learning Model ◽

Training Data ◽

Refractive Indices ◽

Learning Models ◽

Deep Learning Model ◽

Machine Learning Models

<div><div><div><p>We present a multitask, physics-infused deep learning model to accurately and efficiently predict refractive indices (RIs) of organic molecules, and we apply it to a library of 1.5 million compounds. We show that it outperforms earlier machine learning models by a significant margin, and that incorporating known physics into data-derived models provides valuable guardrails. Using a transfer learning approach, we augment the model to reproduce results consistent with higher-level computational chemistry training data, but with a considerably reduced number of corresponding calculations. Prediction errors of machine learning models are typically smallest for commonly observed target property values, consistent with the distribution of the training data. However, since our goal is to identify candidates with unusually large RI values, we propose a strategy to boost the performance of our model in the remoter areas of the RI distribution: We bias the model with respect to the under-represented classes of molecules that have values in the high-RI regime. By adopting a metric popular in web search engines, we evaluate our effectiveness in ranking top candidates. We confirm that the models developed in this study can reliably predict the RIs of the top 1,000 compounds, and are thus able to capture their ranking. We believe that this is the first study to develop a data-derived model that ensures the reliability of RI predictions by model augmentation in the extrapolation region on such a large scale. These results underscore the tremendous potential of machine learning in facilitating molecular (hyper)screening approaches on a massive scale and in accelerating the discovery of new compounds and materials, such as organic molecules with high-RI for applications in opto-electronics.</p></div></div></div>

Download Full-text

Super-Resolution Based on Generative Adversarial Network for HRTEM Images

International Journal of Pattern Recognition and Artificial Intelligence ◽

10.1142/s0218001421540276 ◽

2021 ◽

pp. 2154027

Author(s):

Fuqi Mao ◽

Xiaohan Guan ◽

Ruoyu Wang ◽

Wen Yue

Keyword(s):

Deep Learning ◽

High Resolution ◽

Information Structure ◽

Rapid Development ◽

Super Resolution ◽

Training Data ◽

Learning Technology ◽

Hrtem Image ◽

Generative Adversarial Network ◽

Dual Regression

As an important tool to study the microstructure and properties of materials, High Resolution Transmission Electron Microscope (HRTEM) images can obtain the lattice fringe image (reflecting the crystal plane spacing information), structure image and individual atom image (which reflects the configuration of atoms or atomic groups in crystal structure). Despite the rapid development of HTTEM devices, HRTEM images still have limited achievable resolution for human visual system. With the rapid development of deep learning technology in recent years, researchers are actively exploring the Super-resolution (SR) model based on deep learning, and the model has reached the current best level in various SR benchmarks. Using SR to reconstruct high-resolution HRTEM image is helpful to the material science research. However, there is one core issue that has not been resolved: most of these super-resolution methods require the training data to exist in pairs. In actual scenarios, especially for HRTEM images, there are no corresponding HR images. To reconstruct high quality HRTEM image, a novel Super-Resolution architecture for HRTEM images is proposed in this paper. Borrowing the idea from Dual Regression Networks (DRN), we introduce an additional dual regression structure to ESRGAN, by training the model with unpaired HRTEM images and paired nature images. Results of extensive benchmark experiments demonstrate that the proposed method achieves better performance than the most resent SISR methods with both quantitative and visual results.

Download Full-text

Using Greedy Random Adaptive Procedure to Solve the User Selection Problem in Mobile Crowdsourcing

Sensors ◽

10.3390/s19143158 ◽

2019 ◽

Vol 19 (14) ◽

pp. 3158

Author(s):

Jian Yang ◽

Xiaojuan Ban ◽

Chunxiao Xing

Keyword(s):

Mobile Networks ◽

Large Scale ◽

Rapid Development ◽

Synthetic Data ◽

Data Sets ◽

User Selection ◽

Adaptive Procedure ◽

Mobile Crowdsourcing ◽

Marginal Gain ◽

The Cost

With the rapid development of mobile networks and smart terminals, mobile crowdsourcing has aroused the interest of relevant scholars and industries. In this paper, we propose a new solution to the problem of user selection in mobile crowdsourcing system. The existing user selection schemes mainly include: (1) find a subset of users to maximize crowdsourcing quality under a given budget constraint; (2) find a subset of users to minimize cost while meeting minimum crowdsourcing quality requirement. However, these solutions have deficiencies in selecting users to maximize the quality of service of the task and minimize costs. Inspired by the marginalism principle in economics, we wish to select a new user only when the marginal gain of the newly joined user is higher than the cost of payment and the marginal cost associated with integration. We modeled the scheme as a marginalism problem of mobile crowdsourcing user selection (MCUS-marginalism). We rigorously prove the MCUS-marginalism problem to be NP-hard, and propose a greedy random adaptive procedure with annealing randomness (GRASP-AR) to achieve maximize the gain and minimize the cost of the task. The effectiveness and efficiency of our proposed approaches are clearly verified by a large scale of experimental evaluations on both real-world and synthetic data sets.

Download Full-text

A general approach for improving deep learning-based medical relation extraction using a pre-trained model and fine-tuning

Database ◽

10.1093/database/baz116 ◽

2019 ◽

Vol 2019 ◽

Cited By ~ 2

Author(s):

Tao Chen ◽

Mingfen Wu ◽

Hexi Li

Keyword(s):

Deep Learning ◽

Large Scale ◽

Relation Extraction ◽

Training Model ◽

Biomedical Literature ◽

Training Data ◽

Fine Tuning ◽

Learning Approaches ◽

Additional Time ◽

Clinical Records

Abstract The automatic extraction of meaningful relations from biomedical literature or clinical records is crucial in various biomedical applications. Most of the current deep learning approaches for medical relation extraction require large-scale training data to prevent overfitting of the training model. We propose using a pre-trained model and a fine-tuning technique to improve these approaches without additional time-consuming human labeling. Firstly, we show the architecture of Bidirectional Encoder Representations from Transformers (BERT), an approach for pre-training a model on large-scale unstructured text. We then combine BERT with a one-dimensional convolutional neural network (1d-CNN) to fine-tune the pre-trained model for relation extraction. Extensive experiments on three datasets, namely the BioCreative V chemical disease relation corpus, traditional Chinese medicine literature corpus and i2b2 2012 temporal relation challenge corpus, show that the proposed approach achieves state-of-the-art results (giving a relative improvement of 22.2, 7.77, and 38.5% in F1 score, respectively, compared with a traditional 1d-CNN classifier). The source code is available at https://github.com/chentao1999/MedicalRelationExtraction.

Download Full-text

Multi-Source Distilling Domain Adaptation

Proceedings of the AAAI Conference on Artificial Intelligence ◽

10.1609/aaai.v34i07.6997 ◽

2020 ◽

Vol 34 (07) ◽

pp. 12975-12983

Author(s):

Sicheng Zhao ◽

Guangzhi Wang ◽

Shanghang Zhang ◽

Yang Gu ◽

Yaxian Li ◽

...

Keyword(s):

Domain Adaptation ◽

Feature Space ◽

Wasserstein Distance ◽

Training Data ◽

Source Distribution ◽

Single Source ◽

Multiple Sources ◽

Target Domain ◽

Target Feature ◽

Training Samples

Deep neural networks suffer from performance decay when there is domain shift between the labeled source domain and unlabeled target domain, which motivates the research on domain adaptation (DA). Conventional DA methods usually assume that the labeled data is sampled from a single source distribution. However, in practice, labeled data may be collected from multiple sources, while naive application of the single-source DA algorithms may lead to suboptimal solutions. In this paper, we propose a novel multi-source distilling domain adaptation (MDDA) network, which not only considers the different distances among multiple sources and the target, but also investigates the different similarities of the source samples to the target ones. Specifically, the proposed MDDA includes four stages: (1) pre-train the source classifiers separately using the training data from each source; (2) adversarially map the target into the feature space of each source respectively by minimizing the empirical Wasserstein distance between source and target; (3) select the source training samples that are closer to the target to fine-tune the source classifiers; and (4) classify each encoded target feature by corresponding source classifier, and aggregate different predictions using respective domain weight, which corresponds to the discrepancy between each source and target. Extensive experiments are conducted on public DA benchmarks, and the results demonstrate that the proposed MDDA significantly outperforms the state-of-the-art approaches. Our source code is released at: https://github.com/daoyuan98/MDDA.

Download Full-text