Machine Learning for Health Informatics

Machine Learning (ML) studies algorithms which can learn from data to gain knowledge from experience and to make decisions and predictions. Health Informatics (HI) studies the effective use of probabilistic information for decision making. The combination of both has greatest potential to rise quality, efficacy and efficiency of treatment and care. Health systems worldwide are confronted with “big data” in high dimensions, where the inclusion of a human is impossible and automatic ML (aML) show impressive results. However, sometimes we are confronted with complex data, “little data”, or rare events, where aML-approaches suffer of insufficient training samples. Here interactive ML (iML) may be of help, particularly with a doctor-in-the-loop, e.g. in subspace clustering, k-Anonymization, protein folding and protein design. However, successful application of ML for HI needs an integrated approach, fostering a concerted effort of four areas: (1) data science, (2) algorithms (with focus on networks and topology (structure), and entropy (time), (3) data visualization, and last but not least (4) privacy, data protection, safety & security.

The decision tree is one of the earliest predictive models in machine learning. In the soft decision tree, based on the hierarchical mixture of experts model, internal binary nodes take soft decisions and choose both children with probabilities given by a sigmoid gating function. Hence for an input, all the paths to all the leaves are traversed and all those leaves contribute to the final decision but with different probabilities, as given by the gating values on the path. Tree induction is incremental and the tree grows when needed by replacing leaves with subtrees and the parameters of the newly-added nodes are learned using gradient-descent. We have previously shown that such soft trees generalize better than hard trees; here, we propose to bag such soft decision trees for higher accuracy. On 27 two-class classification data sets (ten of which are from the medical domain), and 26 regression data sets, we show that the bagged soft trees generalize better than single soft trees and bagged hard trees. This contribution falls in the scope of research track 2 listed in the editorial, namely, machine learning algorithms.

The paper reviews a new perspective to discover and compute discrete dynamics, which is based on MP grammars. They are a particular type of multiset rewriting grammars, introduced in 2004 for modeling metabolic systems, which express dynamics in terms of finite difference equations. MP regression algorithms, providing the best MP grammar reproducing a given time series of observed states, were introduced since 2008. Applications of these grammars to the analysis of biological dynamics were developed, and their flexibility to model complex and uncertain phenomena was apparent in the last years. In this paper we recall the main features of this modeling framework, by stressing their peculiarity to afford complex situations, where classical continuous methods cannot be applied or are computationally prohibitive. Moreover, the computational universality of MP grammars of a very simple type is shown, and one of the most relevant cases of MP biological models is shortly presented.

Given its immense growth, scientific literature can be explored to reveal new discoveries, based on yet uncovered relations between knowledge from different, relatively isolated fields of research specialization. This chapter proposes a bisociation-based text mining approach, which shows to be effective for cross-domain knowledge discovery. The proposed cross-domain literature mining functionality, including text acquisition, text preprocessing, and bisociative cross-domain literature mining facilities, is made publicly available within a new browser-based workflow execution engine TextFlows, which supports visual construction and execution of text mining and natural language processing (NLP) workflows. To support bisociative cross-domain literature mining, the TextFlows platform includes implementations of several elementary and ensemble heuristics that guide the expert in the process of exploring new cross-context bridging terms. We have extended the TextFlows platform with several components, which—together with document exploration and visualization features of the CrossBee human-computer interface—make it a powerful, user-friendly text analysis tool for exploratory cross-domain knowledge discovery. Another novelty of the developed technology is the enabled use of controlled vocabularies to improve bridging term extraction. The potential of the developed functionality was showcased in two medical benchmark domains.

Routinely collected data in hospital Electronic Medical Records (EMR) is rich and abundant but often not linked or analysed for purposes other than direct patient care. We have created a methodology to integrate patient-centric data from different EMR systems into clinical pathways that represent the history of all patient interactions with the hospital during the course of a disease and beyond. In this paper, the literature in the area of data visualisation in healthcare is reviewed and a method for visualising the journeys that patients take through care is discussed. Examples of the hidden knowledge that could be discovered using this approach are explored and the main application areas of visualisation tools are identified. This paper also highlights the challenges of collecting and analysing such data and making the visualisations extensively used in the medical domain.

This paper starts by presenting the state-of-the-art in visualisation of clinical and other health related data. Then, it describes an example clinical problem and discusses the visualisation tools and techniques created for the utilisation of these data by clinicians and researchers. Finally, we look at the open problems in this area of research and discuss future challenges.

Segmentation of focal (localized) brain pathologies such as brain tumors and brain lesions caused by multiple sclerosis and ischemic strokes are necessary for medical diagnosis, surgical planning and disease development as well as other applications such as tractography. Over the years, attempts have been made to automate this process for both clinical and research reasons. In this regard, machine learning methods have long been a focus of attention. Over the past two years, the medical imaging field has seen a rise in the use of a particular branch of machine learning commonly known as deep learning. In the non-medical computer vision world, deep learning based methods have obtained state-of-the-art results on many datasets. Recent studies in computer aided diagnostics have shown deep learning methods (and especially convolutional neural networks - CNN) to yield promising results. In this chapter, we provide a survey of CNN methods applied to medical imaging with a focus on brain pathology segmentation. In particular, we discuss their characteristic peculiarities and their specific configuration and adjustments that are best suited to segment medical images. We also underline the intrinsic differences deep learning methods have with other machine learning methods.

Epilepsy is one of the most common neurological disorder. This disorder can be diagnosed by non-invasive examinations, such as electroencephalography, whose records are called electroencephalograms (EEG). The EEG can be stored in medical databases for reusing in future. In these data, one can apply data mining process supported by machine learning techniques in order to find patterns that can be used for building predictive models. This paper presents an application of the cross-correlation technique and the kNN algorithm for classification in a set with 200 EEG segments in order to differentiate normal and epileptic (abnormal) signals. The results were evaluated using 10-fold cross-validation and contingency table methods. With the evaluation using cross validation, it was not found statistically significant difference for classification using kNN. The contingency table results found that the kNN with k = 1 and k = 7 performed better for classifying abnormal and normal EEG, respectively. Also, the kNN with k = 1 and k = 7 were more likely to correctly classify normal and abnormal EEG, respectively.

Biosignals have become an important indicator not only for medical diagnosis and subsequent therapy, but also passive health monitoring. Extracting meaningful features from biosignals can help people understand the human functional state, so that upcoming harmful symptoms or diseases can be alleviated or avoided. There are two main approaches commonly used to derive useful features from biosignals, which are hand-engineering and deep learning. The majority of the research in this field focuses on hand-engineering features, which require domain-specific experts to design algorithms to extract meaningful features. In the last years, several studies have employed deep learning to automatically learn features from raw biosignals to make feature extraction algorithms less dependent on humans. These studies have also demonstrated promising results in a variety of biosignal applications. In this survey, we review different types of biosignals and the main approaches to extract features from the signal in the context of biomedical applications. We also discuss challenges and limitations of the existing approaches, and possible future research.

Modern machine Learning is devoted to the construction of algorithms and computational procedures that can automatically improve with experience and learn from data. Defeasible argumentation has emerged as sub-topic of artificial intelligence aimed at formalising common-sense qualitative reasoning. The former is an inductive approach for inference while the latter is deductive, each one having advantages and limitations. A great challenge for theoretical and applied research in AI is their integration. The first aim of this chapter is to provide readers informally with the basic notions of defeasible and non-monotonic reasoning. It then describes argumentation theory, a paradigm for implementing defeasible reasoning in practice as well as the common multi-layer schema upon which argument-based systems are usually built. The second aim is to describe a selection of argument-based applications in the medical and health-care sectors, informed by the multi-layer schema. A summary of the features that emerge from the applications under review is aimed at showing why defeasible argumentation is attractive for knowledge-representation, conflict resolution and inference under uncertainty. Open problems and challenges in the field of argumentation are subsequently described followed by a future outlook in which three points of integration with machine learning are proposed.

Parkinson’s disease (PD) results primarily from dying of dopaminergic neurons in the Substantia Nigra, a part of the Mesencephalon (midbrain), which is not curable to date. PD medications treat symptoms only, none halt or retard dopaminergic neuron degeneration. Here machine learning methods can be of help since one of the crucial roles in the management and treatment of PD patients is detection and classification of tremors. In the clinical practice, this is one of the most common movement disorders and is typically classified using behavioral or etiological factors. Another important issue is to detect and evaluate PD related gait patterns, gait initiation and freezing of gait, which are typical symptoms of PD. Medical studies have shown that 90% of people with PD suffer from vocal impairment, consequently the analysis of voice data to discriminate healthy people from PD is relevant. This paper provides a quick overview of the state-of-the-art and some directions for future research, motivated by the ongoing PD_manager project.

The generalized adoption of Electronic Medical Records (EMR) together with the need to give the patient the appropriate treatment at the appropriate moment at the appropriate cost is demanding solutions to analyze the information on the EMR automatically. However most of the information on the EMR is non-structured: texts and images. Extracting knowledge from this data requires methods for structuring this information. Despite the efforts made in Natural Language Processing (NLP) even in the biomedical domain and in image processing, medical big data has still to undertake several challenges. The ungrammatical structure of clinical notes, abbreviations used and evolving terms have to be tackled in any Name Entity Recognition process. Moreover abbreviations, acronyms and terms are very much dependant on the language and the specific service. On the other hand, in the area of medical images, one of the main challenges is the development of new algorithms and methodologies that can help the physician take full advantage of the information contained in all these images. However, the large number of imaging modalities used today for diagnosis hinders the availability of general procedures as machine learning is, once again, a good approach for addressing this challenge. In this chapter, which concentrates on the problem of name entity recognition, we review previous approaches and look at future works. We also review the machine leaning approaches for image segmentation and annotation.

The tendency for visual data mining applications in the medical field is increasing, because it is rich with temporal information, furthermore visual data mining is becoming a necessity for intelligent analysis and graphical interpretation. The use of interactive machine learning allows to improve the quality of medical decision-making processes by effectively integrating and visualizing discovered important patterns and/or rules. This chapter provides a survey of visual intelligent decision support systems in the medical field. First, we highlight the benefits of combining potential computational capabilities of data mining with human judgment of visualization techniques for medical decision-making. Second, we introduce the principal challenges of such decision systems, including the design, development and evaluation. In addition, we study how these methods were applied in the medical domain. Finally, we discuss some open questions and future challenges.

A major challenge in precision medicine is the development of biomarkers which can effectively guide patient treatment in a manner which benefits both the individual and the population. Much of the difficulty is the poor reproducibility of existing approaches as well as the complexity of the problem. Machine learning tools with rigorous statistical inference properties have great potential to move this area forward. In this chapter, we review existing pipelines for biomarker discovery and validation from a statistical perspective and identify a number of key areas where improvements are needed. We then proceed to outline a framework for developing a master pipeline firmly grounded in statistical principles which can yield better reproducibility, leading to improved biomarker development and increasing success in precision medicine.

The explosive growth of medical databases and the widespread development of high performance machine learning (ML) algorithms led to the search for efficient computer-aided medical diagnosis (CAMD) techniques. Automated medical diagnosis can be achieved by building a model of a certain disease under surveillance and comparing it with the real time physiological measurements taken from the patient. If this practice is carried out on a regular basis, potential risky medical conditions can be detected at an early stage, thus making the process of fighting the disease much easier. With CAMD, physicians can trustfully use the “second opinion” of the ‘digital assistant’ and make the final optimum decision. The recent development of intelligent technologies, designed to enhance the process of differential diagnosis by using medical databases, significantly enables the decision-making process of health professionals. Up-to-date online medical databases can now be used to support clinical decision-making, offering direct access to medical evidence. In this paper, we provide an overview on selected ML algorithms that can be applied in CAMD, focusing on the enhancement of neural networks (NNs) by hybridization, partially connectivity, and alternative learning paradigms. Particularly, we emphasize the benefits of using such effective algorithms in breast cancer detection and recurrence, colon cancer, lung cancer, liver fibrosis stadialization, heart attack, and diabetes. Generally, the aim is to provide a theme for discussions on ML-based methods applied to medicine.

The rapidly increasing capabilities of neurotechnologies are generating massive volumes of complex multi-modal data at a rapid pace. This neurological big data can be leveraged to provide new insights into complex neurological disorders using data mining and knowledge discovery techniques. For example, electrophysiological signal data consisting of electroencephalogram (EEG) and electrocardiogram (ECG) can be analyzed for brain connectivity research, physiological associations to neural activity, diagnosis, and care of patients with epilepsy. However, existing approaches to store and model electrophysiological signal data has several limitations, which make it difficult for signal data to be used directly in data analysis, signal visualization tools, and knowledge discovery applications. Therefore, use of neurological big data for secondary analysis and potential development of personalized treatment strategies requires scalable data processing platforms. In this chapter, we describe the development of a high performance data flow system called Signal Data Cloud (SDC) to pre-process large-scale electrophysiological signal data using open source Apache Pig. The features of this neurological big data processing system are: (a) efficient partitioningof signal data into fixed size segments for easier storage in high performance distributed file system, (b) integration and semantic annotation of clinical metadata using an epilepsy domain ontology, and (c) transformation of raw signal data into an appropriate format for use in signal analysis platforms. In this chapter, we also discuss the various challenges being faced by the biomedical informatics community in the context of Big Data, especially the increasing need to ensure data quality and scientific reproducibility.

Identifying measurable genetic indicators (or biomarkers) of a specific condition of a biological system is a key element of precision medicine. Indeed it allows to tailor diagnostic, prognostic and treatment choice to individual characteristics of a patient. In machine learning terms, biomarker discovery can be framed as a feature selection problem on whole-genome data sets. However, classical feature selection methods are usually underpowered to process these data sets, which contain orders of magnitude more features than samples. This can be addressed by making the assumption that genetic features that are linked on a biological network are more likely to work jointly towards explaining the phenotype of interest. We review here three families of methods for feature selection that integrate prior knowledge in the form of networks.

There has been a recent surge in the implementation of electronic health care records. These patient records contain valuable medical information including patient demographic data, diagnosis, therapeutic approach, and patient outcomes. It is important to analyze patterns within these records in order to more effectively treat individuals. In this paper, a method is presented for identifying these themes and patterns within patient data. This methodology includes extraction of the main themes or patterns in the data and linking those themes back to the corpus from which they were generated. In our research, we partitioned graphs from terms gathered from electronic medical records. We used two sets of data including eight charts and ten case studies for this study from primary disease categories. The Electronic Medical Records (EMRs) and case studies were modeled as networks of interacting terms where the interactions were captured by their co-occurrences in the documents. A greedy algorithm was used to find communities with high modularity. Finally, we compared our method with probabilistic topic modeling algorithms and evaluated the efficacy of our method by using recall and precision measures.

In this paper, we present the current state-of-the-art of decision making (DM) and machine learning (ML) and bridge the two research domains to create an integrated approach of complex problem solving based on human and computational agents. We present a novel classification of ML, emphasizing the human-in-the-loop in interactive ML (iML) and more specific on collaborative interactive ML (ciML), which we understand as a deep integrated version of iML, where humans and algorithms work hand in hand to solve complex problems. Both humans and computers have specific strengths and weaknesses and integrating humans into machine learning processes might be a very efficient way for tackling problems. This approach bears immense research potential for various domains, e.g., in health informatics or in industrial applications. We outline open questions and name future challenges that have to be addressed by the research community to enable the use of collaborative interactive machine learning for problem solving in a large scale.

Parkinson’s Disease (PD) is a chronic and progressive illness that affects hundreds of thousands of people worldwide. Although it is quite easy to identify someone affected by PD when the illness shows itself (e.g. tremors, slowness of movement and freezing-of-gait), most works have focused on studying the working mechanism of the disease in its very early stages. In such cases, drugs can be administered in order to increase the quality of life of the patients. Since the beginning, it is well-known that PD patients feature the micrography, which is related to muscle rigidity and tremors. As such, most exams to detect Parkinson’s Disease make use of handwritten assessment tools, where the individual is asked to perform some predefined tasks, such as drawing spirals and meanders on a template paper. Later, an expert analyses the drawings in order to classify the progressive of the disease. In this work, we are interested into aiding physicians in such task by means of machine learning techniques, which can learn proper information from digitized versions of the exams, and them recommending a probability of a given individual being affected by PD depending on its handwritten skills. Particularly, we are interested in deep learning techniques (i.e. Convolutional Neural Networks) due to their ability into learning features without human interaction. Additionally, we propose to fine-tune hyper-arameters of such techniques by means of meta-heuristic-based techniques, such as Bat Algorithm, Firefly Algorithm and Particle Swarm Optimization.

Recommender systems are a classical example for machine learning applications, however, they have not yet been used extensively in health informatics and medical scenarios. We argue that this is due to the specifics of benchmarking criteria in medical scenarios and the multitude of drastically differing end-user groups and the enormous context-complexity of the medical domain. Here both risk perceptions towards data security and privacy as well as trust in safe technical systems play a central and specific role, particularly in the clinical context. These aspects dominate acceptance of such systems. By using a Doctor-in-the-Loop approach some of these difficulties could be mitigated by combining both human expertise with computer efficiency. We provide a three-part research framework to access health recommender systems, suggesting to incorporate domain understanding, evaluation and specific methodology into the development process.

The various interplaying variables of tumor growth remain key questions in cancer research, in particular what makes such a growth malignant and what are possible therapies to stop the growth and prevent re-growth. Given the complexity and heterogeneity of the disease, as well as the steadily growing set of publicly available big data sets, there is an urgent need for approaches to make sense out of these open data sets. Machine learning methods for tumor growth profiles and model validation can be of great help here, particularly, discrete multi-agent approaches.

In this tutorial, we will provide an introduction to the main Python software tools used for applying machine learning techniques to medical data. The focus will be on open-source software that is freely available and is cross platform. To aid the learning experience, a companion GitHub repository is available so that you can follow the examples contained in this paper interactively using Jupyter notebooks. The notebooks will be more exhaustive than what is contained in this chapter, and will focus on medical datasets and healthcare problems. Briefly, this tutorial will first introduce Python as a language, and then describe some of the lower level, general matrix and data structure packages that are popular in the machine learning and data science communities, such as NumPy and Pandas. From there, we will move to dedicated machine learning software, such as SciKit-Learn. Finally we will introduce the Keras deep learning and neural networks library. The emphasis of this paper is readability, with as little jargon used as possible. No previous experience with machine learning is assumed. We will use openly available medical datasets throughout.