Subdomain Adaptation With Manifolds Discrepancy Alignment.

Reducing domain divergence is a key step in transfer learning. Existing works focus on the minimization of global domain divergence. However, two domains may consist of several shared subdomains, and differ from each other in each subdomain. In this article, we take the local divergence of subdomains into account in transfer. Specifically, we propose to use the low-dimensional manifold to represent the subdomain, and align the local data distribution discrepancy in each manifold across domains. A manifold maximum mean discrepancy (M3D) is developed to measure the local distribution discrepancy in each manifold. We then propose a general framework, called transfer with manifolds discrepancy alignment (TMDA), to couple the discovery of data manifolds with the minimization of M3D. We instantiate TMDA in the subspace learning case considering both the linear and nonlinear mappings. We also instantiate TMDA in the deep learning framework. Experimental studies show that TMDA is a promising method for various transfer learning tasks.

Digital Health during COVID-19: Informatics Dialogue with the World Health Organization.

BACKGROUND: On December 16, 2020 representatives of the International Medical Informatics Association (IMIA), a Non-Governmental Organization in official relations with the World Health Organization (WHO), along with its International Academy for Health Sciences Informatics (IAHSI), held an open dialogue with WHO Director General (WHO DG) Tedros Adhanom Ghebreyesus about the opportunities and challenges of digital health during the COVID-19 global pandemic. OBJECTIVES: The aim of this paper is to report the outcomes of the dialogue and discussions with more than 200 participants representing different civil society organizations (CSOs). METHODS: The dialogue was held in form of a webinar. After an initial address of the WHO DG, short presentations by the panelists, and live discussions between panelists, the WHO DG and WHO representatives took place. The audience was able to post questions in written. These written discussions were saved with participants' consent and summarized in this paper. RESULTS: The main themes that were brought up by the audience for discussion were: (a) opportunities and challenges in general; (b) ethics and artificial intelligence; (c) digital divide; (d) education. Proposed actions included the development of a roadmap based on the lessons learned from the COVID-19 pandemic. CONCLUSIONS: Decision making by policy makers needs to be evidence-based and health informatics research should be used to support decisions surrounding digital health, and we further propose next steps in the collaboration between IMIA and WHO such as future engagement in the World Health Assembly.

Modeling Multi-View Dependence in Bayesian Networks for Alzheimer's Disease Detection.

Early detection of Alzheimer's disease is important for deploying interventions to prevent or slow disease progression. We propose a multi-view dependence modeling framework that integrates multiple data sources to distinguish patients at different stages of the disease. We design interpretable models that can handle heterogeneous data types including neuro-images, bio- and clinical markers, and historical and genotypical characteristics of the subjects. We learn the dependence structure from data with guidance from domain knowledge in Bayesian Networks, visualizing and quantifying the conditional probabilistic dependence among the variables. Our results indicate that the hybrid dependence models also improve prediction performance.

Knowledge-Driven Interpretation of Multi-View Data in Medicine.

We propose a new approach to clinical decision support with interpretable recommendations from multi-view data. We introduce a Bayesian network structure learning method to help identify the relevant factors and their relationships. Guided by minimal domain knowledge, this method highlights the significant associations among the demography, medical and family history, lifestyle, and biomarker data to facilitate informed clinical decisions. We demonstrate the effectiveness of the method for detecting mild neurocognitive disorder in the elderly from a real-life dataset in Singapore. The empirical results show that our method achieves better interpretability in addition to comparable accuracy with respect to the benchmark studies.

A Deep Learning Approach to Neuroanatomical Characterisation of Alzheimer's Disease.

Alzheimer's disease (AD) is a neurological degenerative disorder that leads to progressive mental deterioration. This work introduces a computational approach to improve our understanding of the progression of AD. We use ensemble learning methods and deep neural networks to identify salient structural correlations among brain regions that degenerate together in AD; this provides an understanding of how AD progresses in the brain. The proposed technique has a classification accuracy of 81.79% for AD against healthy subjects using a single modality imaging dataset.

Fusing Heterogeneous Data for Alzheimer's Disease Classification.

In multi-view learning, multimodal representations of a real world object or situation are integrated to learn its overall picture. Feature sets from distinct data sources carry different, yet complementary, information which, if analysed together, usually yield better insights and more accurate results. Neuro-degenerative disorders such as dementia are characterized by changes in multiple biomarkers. This work combines the features from neuroimaging and cerebrospinal fluid studies to distinguish Alzheimer's disease patients from healthy subjects. We apply statistical data fusion techniques on 101 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. We examine whether fusion of biomarkers helps to improve diagnostic accuracy and how the methods compare against each other for this problem. Our results indicate that multimodal data fusion improves classification accuracy.

Automatic detection and quantification of brain midline shift using anatomical marker model.

Brain midline shift (MLS) is a significant factor in brain CT diagnosis. In this paper, we present a new method of automatically detecting and quantifying brain midline shift in traumatic injury brain CT images. The proposed method automatically picks out the CT slice on which midline shift can be observed most clearly and uses automatically detected anatomical markers to delineate the deformed midline and quantify the shift. For each anatomical marker, the detector generates five candidate points. Then the best candidate for each marker is selected based on the statistical distribution of features characterizing the spatial relationships among the markers. Experiments show that the proposed method outperforms previous methods, especially in the cases of large intra-cerebral hemorrhage and missing ventricles. A brain CT retrieval system is also developed based on the brain midline shift quantification results.

Unsupervised medical image classification by combining case-based classifiers.

We introduce an automated pathology classification system for medical volumetric brain image slices. Existing work often relies on handcrafted features extracted from automatic image segmentation. This is not only a challenging and time-consuming process, but it may also limit the adaptability and robustness of the system. We propose a novel approach to combine sparse Gabor-feature based classifiers in an ensemble classification framework. The unsupervised nature of this non-parametric technique can significantly reduce the time and effort for system calibration. In particular, classification of medical images in this framework does not rely on segmentation, nor semantic-based or annotation-based feature selection. Our experiments show very promising results in classifying computer tomography image slices into pathological classes for traumatic brain injury patients.

An automated pathological class level annotation system for volumetric brain images.

We introduce an automated, pathological class level annotation system for medical volumetric brain images. While much of the earlier work has mainly focused on annotating regions of interest in medical images, our system does not require annotated region level training data nor assumes perfect segmentation results for the regions of interest; the time and effort needed for acquiring training data are hence significantly reduced. This capability of handling high-dimensional noisy data, however, poses additional technical challenges, since statistical estimation of models for such data is prone to over-fitting. We propose a framework that combines a regularized logistic regression method and a kernel-based discriminative method to address these problems. Regularized methods provide a flexible selection mechanism that is well-suited for high dimensional noisy data. Our experiments show promising results in classifying computer tomography images of traumatic brain injury patients into pathological classes.

A generative model based approach to retrieving ischemic stroke images.

This paper proposes a generative model approach to automatically annotate medical images to improve the efficiency and effectiveness of image retrieval systems for teaching, research, and diagnosis. The generative model captures the probabilistic relationships among relevant classification tags, tentative lesion patterns, and selected input features. Operating on the imperfect segmentation results of input images, the probabilistic framework can effectively handle the inherent uncertainties in the images and insufficient information in the training data. Preliminary assessment in the ischemic stroke subtype classification shows that the proposed system is capable of generating the relevant tags for ischemic stroke brain images. The main benefit of this approach is its scalability; the method can be applied in large image databases as it requires only minimal manual labeling of the training data and does not demand high-precision segmentation of the images.

Efficient mining of haplotype patterns for linkage disequilibrium mapping.

Effective identification of disease-causing gene locations can have significant impact on patient management decisions that will ultimately increase survival rates and improve the overall quality of health care. Linkage disequilibrium mapping is the process of finding disease gene locations through comparisons of haplotype frequencies between disease chromosomes and normal chromosomes. This work presents a new method for linkage disequilibrium mapping. The main advantage of the proposed algorithm, called LinkageTracker, is its consistency in producing good predictive accuracy under different conditions, including extreme conditions where the occurrence of disease samples with the mutation of interest is very low and there is presence of error or noise. We compared our method with some leading methods in linkage disequilibrium mapping such as HapMiner, Blade, GeneRecon, and Haplotype Pattern Mining (HPM). Experimental results show that for a substantial class of problems, our method has good predictive accuracy while taking reasonably short processing time. Furthermore, LinkageTracker does not require any population ancestry information about the disease and the genealogy of the haplotypes. Therefore, it is useful for linkage disequilibrium mapping when the users do not have such information about their datasets.

A model driven approach to imbalanced data sampling in medical decision making.

Classification is an important medical decision support function that can be seriously affected by disproportionate class distribution in the training data. In medical decision making, the rate of misclassification and the cost of misclassifying a minority (positive) class as a majority (negative) class are especially high. In this paper, we propose a new model-driven sampling approach to balancing data samples. Most existing data sampling methods produce new data points based on local, deterministic information. Our approach extends the idea of generative sampling to produce new data points based on an induced probabilistic graphical model. We present the motivation and the design of the proposed algorithm, and compare it with two representative imbalanced data sampling approaches on four medical data sets varying in size, imbalance ratio, and dimension. The empirical study helped identify the challenges in imbalanced data problems in medicine, and highlighted the strengths and limitations of the relevant sampling approaches. Performance of the model driven approach is shown to be comparable with existing approaches; potential improvements could be achieved by incorporating domain knowledge.

Towards a multi-level game model for influenza epidemics.

Although game theory has been first invented to reason with economic scenarios with rational agents, it has since been extended into many other fields including biological and medical sciences. In this paper we propose to model the interactions between virus and human in an influenza epidemic in a two player, adversarial game scenario with multiple levels of abstraction. As conventional game representations are inadequate in this complex problem domain, we propose Object Oriented Multi-Agent Influence Diagrams (OO-MAID), a novel graphical representation for multi-level games, which takes advantage of both organizational information and probabilistic independence in the problem domain. The OO-MAID representation can be readily applied in similar medical independent characteristics. We demonstrate the feasibility of this novel approach with sample models in the domain.

Writing for publication in biomedical informatics.

Writing for publication can be a rewarding activity for researchers at all levels of experience. However, many students and researchers are less familiar with the various aspects of the publication process. The purpose of this workshop is to provide participants with the knowledge, skills, and practical advice that can lead to successful scientific publications.

Biomedical knowledge discovery with topological constraints modeling in Bayesian networks: a preliminary report.

Serving as exploratory data analysis tools, Bayesian networks (BNs) can be automatically learned from data to compactly model direct dependency relationships among the variables in a domain. A major challenge in BN learning is to effectively represent and incorporate domain knowledge in the learning process to improve its efficiency and accuracy. In this paper, we examine two types of domain knowledge representation in BNs: matrix and rule. We develop a set of consistency checking mechanisms for the representations and describe their applications in BN learning. Empirical results from the canonical Asia network example show that topological constraints, especially those imposed on the undirected links in the corresponding completed partially directed acyclic graph (CPDAG) of the learned BN, are particularly useful. Preliminary experiments on a real-life coronary artery disease dataset show that both efficiency and accuracy can be improved with the proposed methodology. The bootstrap approach adopted in the BN learning process with topological constraints also highlights the set of the learned links with high significance, which can in turn prompt further exploration of the actual relationships involved.

Predicting coronary artery disease with medical profile and gene polymorphisms data.

Coronary artery disease (CAD) is a main cause of death in the world. Finding cost-effective methods to predict CAD is a major challenge in public health. In this paper, we investigate the combined effects of genetic polymorphisms and non-genetic factors on predicting the risk of CAD by applying well known classification methods, such as Bayesian networks, naïve Bayes, support vector machine, k-nearest neighbor, neural networks and decision trees. Our experiments show that all these classifiers are comparable in terms of accuracy, while Bayesian networks have the additional advantage of being able to provide insights into the relationships among the variables. We observe that the learned Bayesian Networks identify many important dependency relationships among genetic variables, which can be verified with domain knowledge. Conforming to current domain understanding, our results indicate that related diseases (e.g., diabetes and hypertension), age and smoking status are the most important factors for CAD prediction, while the genetic polymorphisms entail more complicated influences.