Objective: To explore potential risk factors for the emergency caesarean section in term, singleton pregnancies. Methods: A retrospective population based case-control study in term deliveries from the University Hospital in Brno, Czech Republic collected between 2014 and 2016. Cases were deliveries by emergency caesarean section; controls were all others modes of delivery. We excluded elective caesarean from the populations. Results: In the database of 13769 deliveries, we identified 2178 cases. Univariate and multivariable analysis of clinical features were performed. The following risk factors were associated with emergency caesarean section: Breech presentation OR 21.6 (14.6–30.5), obstructed labor OR 42.7 (28.5–63.9), scar on uterus OR 18 (14.3–22.5), fetal distress OR 6.1 (5.3–7.2) and primipara OR 3.7 (3.1–4.4). Conclusion: Univariate and multivariable analysis of the data from 13769 deliveries were performed, and significant risk factors were identified increasing the chance of undergoing caesarean section.
The study is devoted to the sleep stage identification problem. Proposed method is based on calculation of covariance matrices from segments of multi-modal recordings. Mathematical properties of the extracted covariance matrices allow to define a distance between two segments - a distance in a Riemannian manifold. In the paper we tested minimum distance to a class center and k-nearest-neighbours classifiers with the Riemannian metric as a distance between two objects, and classification in a tangent space to a Riemannian manifold. Methods were tested on the data of patients suffering from sleep disorders. The maximum obtained accuracy for KNN is 0.94, for minimum distance to a class center it is only 0.816 and for classification in a tangent space is 0.941.
Recently, various studies have shown that meaningful knowledge can be discovered by applying data mining techniques in medical applications, i.e., decision support systems for disease diagnosis. However, there are still several computational challenges due to the high-dimensionality of medical data. Feature selection is an essential pre-processing procedure in data mining to identify relevant feature subset for classification. In this study, we proposed a hybrid feature selection mechanism by combining symmetrical uncertainty and Bayesian network. As a case study, we applied our proposed method to the hypertension diagnosis problem. The results showed that our method can improve the classification performance and outperformed existing feature selection techniques.
In this work, we study the use of convolutional neural networks for biomedical signal processing. Convolutional neural networks show promising results for classifying images when compared to traditional multilayer perceptron, as the latter do not take spatial structure of the data into an account.Cardiotocography (CTG) is a monitoring of fetal heart rate (FHR) and uterine contractions (UC) used by obstetricians to assess fetal wellbeing. Because of the complexity of FHR dynamics, regulated by several neurological feedback loops, the visual inspection of FHR remains a difficult task. The application of most guidelines often result in significant inter-and intra-observer variability.Convolutional neural network (CNN, or ConvNet) is inspired by the organization of the animal visual cortex. In the paper we are applying continuous wavelet transform (CWT) to the UC and FHR signals with different levels of time/frequency detail parameter and in two different resolutions. The output 2D structures are fed to convolutional neural network (we are using Tensorflow framework [ 1]) and we are minimizing the cross entropy function.On the testing dataset (with pH threshold at 7.15) we have achieved the accuracy of 94.1% which is a promising result that needs to be further studied.
Data intensive disciplines, such as life sciences and medicine, are promoting vivid research activities in the area of data science. Modern technologies, such as high-throughput mass-spectrometry and sequencing, micro-arrays, high-resolution imaging, etc., produce enormous and continuously increasing amounts of data. Huge public databases provide access to aggregated and consolidated data on genome and protein sequences, biological pathways, diseases, anatomy atlases, and scientific literature. There has never been before more potentially available data to study biomedical systems, ranging from single cells to complete organisms. However, it is a non-trivial task to transform the vast amount of biomedical data into actionable, useful and usable information, triggering scientific progress and supporting patient management.
Pathologist needs to routinely make management decisions about patients who are at risk for a disease such as cancer. Although, making a decision for cancer diagnosis is a dificult task since it context dependent. The term context contains a large number of elements that limits strongly any possibility to automatize this. The decision-making the process being highly contextual, the decision support system must benefit from its interaction with the expert to learn new practices by acquiring missing knowledge incrementally and learning new practices, it is called in deferent research human/doctor in the loop, and thus enriching its experience base.
The analysis of organizational and medical treatment processes is crucial for the future development of the healthcare domain. Recent approaches to enable process mining on healthcare data make use of the hospital information systems' Audit Trails. In this work, methods are proposed to integrate Audit Trail data into the generic OpenSLEX meta model to allow for an analysis of healthcare data from different perspectives (e.g. patients, doctors, resources). Instead of flattening the event data in a single log file the proposed methodology preserves as much information as possible in the first stages of data extraction and preparation. By building on established standardized data and message specifications for auditing in healthcare, we increase the range of analysis opportunities in the healthcare domain.
The present work presents a comparative assessment of glucose prediction models for diabetic patients using data from sensors monitoring blood glucose concentration as well as data from in silico simulations. The models are based on neural networks and linear and nonlinear mathematical models evaluated for prediction horizons ranging from 5 to 120 min. Furthermore, the implementation of compartment models for simulation of absorption and elimination of insulin, caloric intake and information about physical activity is examined in combination with neural networks and mathematical models, respectively. This assessment also addresses the recent progress and challenges in designing glucose regulators based on model predictive control used as part of artificial pancreas devices for type 1 diabetic patients. The assessments include 24 papers in total, from 2006 to 2016, in order to investigate progress in blood glucose concentration prediction and in Artificial Pancreas devices for type 1 diabetic patients.
The health care industry in the United States is undergoing a paradigm shift from the traditional fee-for-service model to various payment and incentive models based on quality of care rather than quantity of services. One specific scenario where more treatment does not equate to better care is red blood cell transfusions. While blood transfusions often save lives, there are numerous complications which can result and blood should be transfused only if medically necessary. Several studies have indicated that a very high percentage of units transfused are not clinically appropriate. These transfusions increase cost and negatively impact patient outcomes. In this paper, we will present an analytics project to identify and track the transfusions which are performed without clear necessity. Furthermore, we will describe how we utilized data discovery and supervised learning to improve our classification algorithm and the accuracy of our results. We will demonstrate that our project is effectively reducing red blood cell transfusions.
This research presents a methodology for health data analytics through a case study for modelling cancer patient records. Timeline-structured clinical data systems represent a new approach to the understanding of the relationship between clinical activity, disease pathologies and health outcomes. The novel Southampton Breast Cancer Data System contains episode and timeline-structured records on > 17,000 patients who have been treated in University Hospital Southampton and affiliated hospitals since the late 1970s. The system is under continuous development and validation. Modern data mining software and visual analytics tools permit new insights into temporally-structured clinical data. The challenges and outcomes of the application of such software-based systems to this complex data environment are reported here. The core data was anonymised and put through a series of pre-processing exercises to identify and exclude anomalous and erroneous data, before restructuring within a remote data warehouse. A range of approaches was tested on the resulting dataset including multidimensional modelling, sequential patterns mining and classification. Visual analytics software has enabled the comparison of survival times and surgical treatments. The systems tested proved to be powerful in identifying episode sequencing patterns which were consistent with real-world clinical outcomes. It is concluded that, subject to further refinement and selection, modern data mining techniques can be applied to large and heterogeneous clinical datasets to inform decision making.
Multidisciplinary Team Meetings (MDTM) are conducted to discuss the treatment of one or more patients. This paper discusses MDTM with a focus on tumor treatment and shows workflows in different settings, identifies organizational and technical problems in the MDTM and solutions thereof. It aims to answer the following research questions: (RQ1) What is the current state of the art in MDTM?(RQ2) How are they conducted and what is the variation in different hospital settings? (RQ3) What technical problems and possible solutions thereof exist? This is done by conducting a literature review entailing a forward search of 837 papers and a backward search. The results show that a unified workflow model for MDTM can't be found since they are highly dependent on institutional and tumor dependent specifics. The identified problems and solutions show a lack of research towards technical solutions and process interoperability. An outlook on extending research in these areas is given.
Data are the crucial component of most computer based clinical decision support systems. This review focuses on data for a system which should improve everyday life of diabetics. The aim is to identify issues arising during the process of acquisition of photos of dishes obtained by diabetic patients. Solutions are proposed that will improve the quality of subsequent processing and final conclusions. This research will lead to a proposal of some guidelines that patients should follow when taking the pictures of dishes. For this purpose, a sample of 906 photos from 6 patients including meals and text records of activities was examined carefully in order to extract useful information about how do diabetics chose to record the details, how much and how long do they follow the suggestions. Based on the analysis, representative examples are presented with corresponding suggestions for each case.
In this study, a new spatio-spectral filtering method for motor imagery signal analysis is introduced. Motor imagery is an important research area in brain computer interfacing. EEG signals related with motor imagery have characteristic frequencies originating from sensorimotor cortex. Common spatial patterns (CSP) method is a very popular and successful spatial filtering algorithm in motor imagery classification. However, CSP only optimizes spatial filters, subject specific frequency selection should be done manually, which is a meticulous process. Therefore, an automatic method for spectral filter optimization is needed. Proposed filter bank common spatio-spectral patterns (FBCSSP) algorithm optimizes spatial and spectral filters. FBCSSP method uses a network of a filter bank and two consecutive CSP layers so that proposed structure has a subject specific response in both spatial and spectral domains. We inspected the proposed method in terms of classification accuracy and physiological consistence of the created filters using publicly available data set. FBCSSP method gave higher classification accuracy than other spatio-spectral pattern methods in the literature. Also, obtained spatial and spectral filters were consistent with the spatial and spectral properties of motor imagery signals.
Objective: To identify potential risk factors for low umbilical cord artery pH in term, singleton pregnancies. Methods: Retrospective case-control study. Cases were deliveries characterized by umbilical cord artery \(pH \le 7.05\). Controls were with no sign of hypoxia. Results: In the database of 10637 deliveries, collected between 2014 and 2015 at the University Hospital in Brno delivery ward, we identified 99 cases. Univariate analysis of clinical features was performed. The following risk-factors were associated with low pH: the length of the first stage (odds ratio (OR) 1.40 (95 % CI 1.04–1.89)) and the length of the second stage of labor (OR 2.86 (95 % CI 1.70–4.81)), primipara (OR 2.99 (1.90–4.71)) and meconium stained fluid (OR 1.60 (1.07–2.38)). Conclusion: Among the risk factors that increase the chance of low umbilical cord artery pH at term, we identified: excessive length of the first and second stage of labour, parity, and meconium stained fluid.
In the today’s world we witness an impact of the ‘Big data’ phenomenon. Although there are many definitions and different scientists view the problem from their perspectives (web, IoT, smartphones, security, GIS, HIS, cloud systems, networks, ...), there is still need for efficient, robust and scalable algorithms that ease processing of such data.
The era of “big data” arose the need to have computational tools in support of biological tasks. Many types of bioinformatics tools have been developed for different biological tasks as target, pathway and gene set analysis, but integrated resources able to incorporate a unique web interface, and to manage a biological scenario involving many different data sources are still lacking. In many bioinformatics approaches several data processing and evaluation steps are required to reach the final results. In this work, we face a biological case study by exploiting the capabilities of an integrated multi-component resources database that is able to deal with complex biological scenarios. As example of our problem-solving approach we provide a case study on the analysis of functional effect of miRNA single nucleotide polymorphisms (SNPs) in cancer disease.
With the technology emerging more and more possible applications of process mining in healthcare become apparent. In most cases the goal of applying process mining to the healthcare domain is to find out what actually happened and to deliver a concise assessment of the organizational reality by mining the event logs of health information systems. To develop medical guidelines or patient pathways considering economic aspects and quality of care, a comparative analysis of different existing approaches is useful (e.g. how different hospitals execute the same process in different ways). This work discusses how to use existing process mining techniques for comparative analysis of healthcare processes and presents an approach based on the L* life-cycle model.
Cancer constitutes a condition and is referred to a group of numerous different diseases, that are characterized by uncontrolled cell growth. Tumors, in the broader sense, are described by abnormal cell growth and are not exclusively cancerous. The molecular basis involves a process of multiple steps and underlying signaling pathways, building up a complex biological framework. Cancer research is based on both disciplines of quantitative and life sciences which can be connected through Bioinformatics and Systems Biology. Our study aims to provide an enhanced computational model on tumor growth towards a comprehensive simulation of miscellaneous types of neoplasms. We create model profiles by considering data from selected types of tumors. Growth parameters are evaluated for integration and compared to the different disease examples. Herein, we describe an extension to the recently presented visualization tool for tumor growth. The integration of profiles offers exemplary simulations on different types of tumors. The enhanced bio-computational simulation provides an approach to predicting tumor growth towards personalized medicine.
Biomedical research requires deep domain expertise to perform analyses of complex data sets, assisted by mathematical expertise provided by data scientists who design and develop sophisticated methods and tools. Such methods and tools not only require preprocessing of the data, but most of all a meaningful input selection. Usually, data scientists do not have sufficient background knowledge about the origin of the data and the biomedical problems to be solved, consequently a doctor-in-the-loop can be of great help here. In this paper we revise the viability of integrating an analysis guided visualization component in an ontology-guided data infrastructure, exemplified by the principal component analysis. We evaluated this approach by examining the potential for intelligent support of medical experts on the case of cerebral aneurysms research.
Information management in healthcare is nowadays experiencing a great revolution. After the impressive progress in digitizing medical data by private organizations, also the federal government and other public stakeholders have also started to make use of healthcare data for data analysis purposes in order to extract actionable knowledge. In this paper, we propose an architecture for supporting interoperability in healthcare systems by exploiting Big Data techniques. In particular, we describe a proposal based on big data techniques to implement a nationwide system able to improve EHR data access efficiency and reduce costs.
