Application of deep learning in radiology has the potential to automate workflows, support radiologists with decision support, and provide patients a logic-based algorithmic assessment. Unfortunately, medical datasets are often not uniformly distributed due to a naturally occurring imbalance. For this research, a multi-classification of liver MRI sequences for imaging of hepatocellular carcinoma (HCC) was conducted on a highly imbalanced clinical dataset using deep convolutional neural network. We have compared four multi classification classifiers which were Model A and Model B (both trained using imbalanced training data), Model C (trained using augmented training images) and Model D (trained using under sampled training images). Data augmentation such as 45-degree rotation, horizontal and vertical flip and random under sampling were performed to tackle class imbalance. HCC, the third most common cause of cancer-related mortality [1], can be diagnosed with high specificity using Magnetic Resonance Imaging (MRI) with the Liver Imaging Reporting and Data System (LI-RADS). Each individual MRI sequence reveals different characteristics that are useful to determine likelihood of HCC. We developed a deep convolutional neural network for the multi-classification of imbalanced MRI sequences that will aid when building a model to apply LI-RADS to diagnose HCC. Radiologists use these MRI sequences to help them identify specific LI-RADS features, it helps automate some of the LIRADS process, and further applications of machine learning to LI-RADS will likely depend on automatic sequence classification as a first step. Our study included an imbalanced dataset of 193,868 images containing 10 MRI sequences: in- phase (IP) chemical shift imaging, out-phase (OOP) chemical shift imaging, T1-weighted post contrast imaging (C+, C-, C-C+), fat suppressed T2 weighted imaging (T2FS), T2 weighted imaging, Diffusion Weighted Imaging (DWI), Apparent Diffusion Coefficient map (ADC) and In phase/Out of phase (IPOOP) imaging. Model performance for Models A, B, C and D provided a macro average F1 score of 0.97, 0.96, 0.95 and 0.93 respectively. Model A showed higher classification scores than models trained using data augmentation and under sampling. / Thesis / Master of Science (MSc)

Medical records are rapidly being digitized to electronic medical records. Although Electronic Medical Records (EMRs) improve administration, billing, and logistics, an open research problem remains as to how doctors can leverage EMRs to enhance patient care. This thesis describes a system that analyzes a patient's evolving EMR in context with available biomedical knowledge and the accumulated experience recorded in various text sources including the EMRs of other patients. The aim of the Clinician Decision Support (CDS) Dashboard is to provide interactive, automated, actionable EMR text-mining tools that help improve both the patient and clinical care staff experience. The CDS Dashboard, in a secure network, helps physicians find de-identified electronic medical records similar to their patient's medical record thereby aiding them in diagnosis, treatment, prognosis and outcomes. It is of particular value in cases involving complex disorders, and also allows physicians to explore relevant medical literature, recent research findings, clinical trials and medical cases. A pilot study done with medical students at the Virginia Tech Carilion School of Medicine and Research Institute (VTC) showed that 89% of them found the CDS Dashboard to be useful in aiding patient care for doctors and 81% of them found it useful for aiding medical students pedagogically. Additionally, over 81% of the medical students found the tool user friendly. The CDS Dashboard is constructed using a multidisciplinary approach including: computer science, medicine, biomedical research, and human-machine interfacing. Our multidisciplinary approach combined with the high usability scores obtained from VTC indicated the CDS Dashboard has a high potential value to clinicians and medical students. / Master of Science

Neuromuscular disorders change the underlying structure and function of motor units within a muscle, and are detected using needle electromyography. Currently, inferences about the presence or absence of disease are made subjectively and are largely impression-based. Quantitative electromyography (QEMG) attempts to improve upon the status quo by providing greater levels of precision, objectivity and reproducibility through numeric analysis, however, their results must be transparently presented and explained to be clinically viable. The probabilistic muscle characterization (PMC) model is ideally suited for a clinical decision support system (CDSS) and has many analogues to the subjective analysis currently used. To improve disease characterization performance globally, a hierarchical classification strategy is developed that accounts for the wide range of MUP feature values present at different levels of involvement (LOI) of a disorder. To improve utility, methods for detecting LOI are considered that balance the accuracy in reporting LOI with its clinical utility. Finally, several aggregation methods that represent commonly used human decision-making strategies are considered and evaluated for their suitability in a CDSS. Four aggregation measures (Average, Bayes, Adjusted Bayes, and WMLO) are evaluated, that offer a compromise between two common decision making paradigms: conservativeness (average) and extremeness (Bayes). Standard classification methods have high specificity at a cost of poor sensitivity at low levels of disease involvement, but tend to improve with disease progression. The hierarchical model is able to provide a better balance between low-LOI sensitivity and specificity by providing the classifier with more concise definitions of abnormality due to LOI. Furthermore, a method for detecting two discrete levels of disease involvement (low and high) is accomplished with reasonable accuracy. The average aggregation method offers a conservative decision that is preferred when the quality of the evidence is poor or not known, while the more extreme aggregators such as Bayes rule perform optimally when the evidence is accurate, but underperform otherwise due to outlier values that are incorrect. The methods developed offer several improvements to PMC, by providing a better balance between sensitivity and specificity, through the definition of a clinically useful and accurate measure of LOI, and by understanding conditions for which each of the aggregation measures is better suited. These developments will enhance the quality of decision support offered by QEMG techniques, thus improving the diagnosis, treatment and management of neuromuscular disorders.

Serum PSA (Prostate Specific Antigen) level is used for prediction of prostatic carcinoma, but it suffers from weak sensitivity and specificity. We applied logistic regression, artificial neural networks, decision tree, and genetic algorithm to prostate cancer prediction problem to design a model for Turkish population. A hybrid model of logistic regression and decision tree has been designed. The model could prevent 33 biopsies (4.4% of our patients who have PSA level between 0 and 10) from our data set without a loss from sensitivity. The prepared online decision support tool and a questionnaire were published on a website. Fifty urologists have completed the questionnaire. Cronbach&rsquo / s alpha was 0.770. On a five graded Likert scale, the mean score of &ldquo / attitude to computer use in healthcare&rdquo / (ACH) was 4.2. The mean of eight responses related to the online tool (Attitude to Decision Support Tool / ADST), was 3.7. ADST was correlated with ACH (r=0.351, p=0.013). Physicians who have positive attitude to computer use in healthcare tend to use the tool (r=0.459, p=0.001). The first factor influencing the opinions of the urologists was the attitude of the user to computer use in healthcare, the other factor was the attitude of the user to the decision support tool itself. To increase the acceptance, education and training of physicians in the use of information technologies in healthcare, informing users about the logic of the decision support tool, and redesigning the system according to user feedback may be helpful.

Introduction: Rheumatoid arthritis (RA) is a chronic inflammatory disease. Treatment strategies emphasize early multi-professional interventions to reduce disease activity and to prevent disability, but there is a lack of knowledge on how optimal treatment can be provided to each individual patient. Aim: To elucidate how clinical manifestations of early RA are associated to disease and disability outcomes, to strive for greater potential to establish prognosis in early RA, and to facilitate implementation of decision support through analyses of the decision-making environment in chronic care. Methods: Multivariate statistics and mathematical modelling, as well as field observations and focus group interviews. Results: Decision support: A prognostic tree that predicted patients with a poor prognosis (moderate or high levels of DAS-28) at one year after diagnosis had a performance of 25% sensitivity, 90% specificity and a positive predictive value of 76%. Implementation of a decision support application at a rheumatology unit should include taking into account incentive structures, workflow and awareness, as well as informal communication structures. Prognosis: A considerable part of the variance in disease activity at one year after diagnosis could be explained by disease progression during the first three months after diagnosis. Using different types of knowledge – different expertise – prior to standardized data mining methods was found to be a promising when mining (clinical) data for new patterns that elicit new knowledge. Disease and disability: Women report more fatigue than men in early RA, although the difference is not consistently significant. Fatigue in early RA is closely and rather consistently related to disease activity, pain and activity limitation, as well as to mental health and sleep disturbance. Conclusion: A decision tree was designed to identify patients at risk of poor prognosis at one year after the diagnosis of RA. When constructing prediction rules for good or poor prognosis, including more measures of disease and disability progressions showed promise. Using different types of knowledge – different lenses of expertise – prior to standardized data mining methods was also a promising method when mining (clinical) data for new patterns that elicit new knowledge. / Introduktion: Reumatoid artrit (RA) är en kronisk inflammatorisk sjukdom. Dagens behandlingsstrategi bygger på tidiga multiprofessionella insatser för att reducera sjukdomsaktivitet och minska risken för framtida funktionshinder. Idag finns stora datamängder tillgängliga gällande medicinering och utfall vid RA. Dessa data erbjuder möjligheter att generera ny kunskap som kan användas för att forma beslutsstöd. Syfte: Att undersöka hur olika kliniska manifestationer vid tidig RA samvarierar med funktionshinder och sjukdomsaktivitet, att pröva metoder att ställa prognos vid tidig RA, och att analysera en kontext för beslutsfattande inom vård av kroniskt sjuka. Metod: Multivariat statistik och matematisk modellering, samt observationsstudier och fokusgruppsintervjuer. Resultat: Beslutsstöd: Ett beslutsträd utformades för att bestämma vilka patienter som har dålig prognos (måttlig eller hög DAS-28) ett år efter diagnos. Beslutsträdet hade 25 % sensitivitet, 90 % specificitet och ett positivt prediktivt värde på 76 %. Vid införande av beslutsstöd på en reumatologisk klinik befanns det nödvändigt att hänsyn tas till incitamentsstrukturer, arbetsflöde och samarbetsformer. Informella kommunikationsstrukturer kan också ha stort inflytande på klinisk praxis. Prognos: En betydande del av variansen i sjukdomsaktivitet ett år efter diagnos kan förklaras av sjukdomsprogression första tre månaderna efter diagnos. Att formalisera olika experters erfarenheter före standardiserade ”data mining” metoder är en lovande ansats när man letar efter mönster i (kliniska) databaser. Funktionshinder och sjukdomsaktivitet: Kvinnor rapporterar mer trötthet än män vid tidig RA, men skillnaden är inte konsistent över tid. Trötthet vid tidig RA är nära relaterat till sjukdomsaktivitet, smärta och aktivitets begränsningar, men också till mental hälsa och sömnstörningar. Slutsats: Ett beslutsträd har utformats för att predicera patienter med dålig prognos inom tidig RA. Studier av fler mått på sjukdoms- och funktionshindersprogression behövs vid konstruktion av prediktionsregler för god eller dålig prognos framledes. Att använda sig av kunskap från olika experter – olika experters glasögon – vid sökandet efter mönster i stora datamängder för att generera ny kunskap är en lovande metodik. Implementering av beslutsstöd bör göras under övervägande av incitamentsstrukturer, arbetsflöde och samarbetsformer.

Neuromuscular disorders change the underlying structure and function of motor units within a muscle, and are detected using needle electromyography. Currently, inferences about the presence or absence of disease are made subjectively and are largely impression-based. Quantitative electromyography (QEMG) attempts to improve upon the status quo by providing greater levels of precision, objectivity and reproducibility through numeric analysis, however, their results must be transparently presented and explained to be clinically viable. The probabilistic muscle characterization (PMC) model is ideally suited for a clinical decision support system (CDSS) and has many analogues to the subjective analysis currently used. To improve disease characterization performance globally, a hierarchical classification strategy is developed that accounts for the wide range of MUP feature values present at different levels of involvement (LOI) of a disorder. To improve utility, methods for detecting LOI are considered that balance the accuracy in reporting LOI with its clinical utility. Finally, several aggregation methods that represent commonly used human decision-making strategies are considered and evaluated for their suitability in a CDSS. Four aggregation measures (Average, Bayes, Adjusted Bayes, and WMLO) are evaluated, that offer a compromise between two common decision making paradigms: conservativeness (average) and extremeness (Bayes). Standard classification methods have high specificity at a cost of poor sensitivity at low levels of disease involvement, but tend to improve with disease progression. The hierarchical model is able to provide a better balance between low-LOI sensitivity and specificity by providing the classifier with more concise definitions of abnormality due to LOI. Furthermore, a method for detecting two discrete levels of disease involvement (low and high) is accomplished with reasonable accuracy. The average aggregation method offers a conservative decision that is preferred when the quality of the evidence is poor or not known, while the more extreme aggregators such as Bayes rule perform optimally when the evidence is accurate, but underperform otherwise due to outlier values that are incorrect. The methods developed offer several improvements to PMC, by providing a better balance between sensitivity and specificity, through the definition of a clinically useful and accurate measure of LOI, and by understanding conditions for which each of the aggregation measures is better suited. These developments will enhance the quality of decision support offered by QEMG techniques, thus improving the diagnosis, treatment and management of neuromuscular disorders.

Main reason of this study is to set forth the internal paradox of the basic approach of the artificial intelligence in the medical field to by discussing on the theoretical and application levels and to suggest solutions in theory and practice against that. In order to rule out the internal paradox in the medical decision support systematic, a new medical model is suggested and based on this, concepts such as disease, health, etiology, diagnosis and treatment are questioned. Meanwhile, with the current scientific data, a simple application sample based on how a decision making system which was set up by fuzzy logic and which is based on the perception of human as a complex adaptive system has been explained. Finally, results of the research about accuracy and validity of this application, current improvements based on the current model and the location on the artificial intelligence theory is discussed.

Background: UK primary care accounts for 90% of patient contacts in the NHS, and over 300 million consultations every year. Consequently, when primary is suboptimal it has important impacts on population health. At the same time, virtually all general practices use electronic health records (EHR) to capture patient data. Clinical Decision Support (CDS) systems use it to highlight when individual patients do not receive care consistent with clinical guidelines, though ignore the wider population. Electronic Audit and Feedback (e-A&F) systems address the wider population, but their results are difficult to interpret. EHR data has the richness to suggest ways in which care quality could be improved, though this is currently not exploited. The aim of this thesis was to make progress towards better use of primary care EHR data for the purposes of quality improvement (QI) by focusing on e-A&F as a vehicle. Research Objectives were: 1) Develop a model and recommendations to guide EHR data analysis and its communication to health professionals; 2) Use these models and recommendations to develop a system for UK primary care; 3) Implement and evaluate the system to test the models and recommendations, and derive generalisable knowledge. Methods: The overall approach of this thesis was informed by guidance from the Medical Research Council on the development of complex interventions, and Boyrcki et al.’s evidence-based framework for the development of health information technologies (Chapter 2). Theory was first identified through a critical examination of the empirical and theoretical literature regarding CDS and e-A&F systems (Chapter 3), then built upon in a systematic literature search and metasynthesis of qualitative studies of A&F (and e-A&F) interventions (Chapter 4). This resulted in the development a new theory of A&F (Clinical Performance Feedback Intervention Theory; CP-FIT), which was used to inform the development of an e-A&F system for UK primary care – the Performance Improvement plaN GeneratoR (PINGR; version 1). PINGR was then iteratively optimised through a series of three empirical studies. First, its usability was evaluated by software experts using Heuristic Evaluation and Cognitive Walkthrough methodologies (Chapter 5). GPs then performed structured tasks using the system in a laboratory whilst their on-screen interactions and eye movements were recorded (Chapter 6). Finally, PINGR was implemented in 15 GP practices, and CP-FIT used to guide the mixed methods evaluation including examinations of usage records, and interviews with 38 health professionals. Results: There are both empirical and theoretical arguments for combining features from CDS and e-A&F systems to increase their effectiveness; a key recommendation is that e-A&F systems should suggest clinical actions to health professionals (Chapter 3). This is supported by CP-FIT, which has three core propositions: 1) A&F interventions exert their effects through health professionals taking action; 2) Health care organisations have limited capacity to engage with A&F; and 3) Health care professionals and organisations have a strong set of beliefs and behaviours regarding how they provide patient care (Chapter 4). Based on these findings, the unique feature of PINGR is that it suggests improvement actions to users based on EHR data analysis (‘decision-supported feedback’). Key findings from PINGR’s usability evaluation with software experts translated into a set of design guidelines for e-A&F interfaces regarding: summarising clinical performance, patient lists, patient-level information, and suggested actions (Chapter 5). When tested with GPs, these guidelines were found to impact: user engagement; actionability; and information prioritisation (Chapter 6). Following its implementation in practice, PINGR was used on 227 occasions to facilitate the care of 725 patients. These patients were 1.8 (95% CI 1.6-1.9) times more likely to receive improved care according to at least one clinical guideline. Barriers and facilitators to its success included: the resources available to use it; its perceived relative advantages; how compatible it was with pre-existing beliefs and ways of working; the credibility of its data; the complexity of the clinical problems it highlighted; and the ability to act on its recommendations (Chapter 7). Conclusion: It is both feasible and acceptable to health professionals to make better use of EHR data for QI by enabling e-A&F systems to suggest actions for them to take. When designing e-A&F interfaces, attention should be paid to how they summarise clinical performance, and present patient lists and detailed patient-level information. Implementation of e-A&F interventions is influenced by availability of resources, compatibility with existing workflows, and ability to take action based on their feedback results. Unresolved tensions exist regarding how they may deal with patient complexity. Policymakers should consider the relevance of these findings for National Clinical Audits and pay-for-performance initiatives.

abstract: Type 1 diabetes (T1D) is a chronic disease that affects 1.25 million people in the United States. There is no known cure and patients must self-manage the disease to avoid complications resulting from blood glucose (BG) excursions. Patients are more likely to adhere to treatments when they incorporate lifestyle preferences. Current technologies that assist patients fail to consider two factors that are known to affect BG: exercise and alcohol. The hypothesis is postprandial blood glucose levels of adult patients with T1D can be improved by providing insulin bolus or carbohydrate recommendations that account for meal and alcohol carbohydrates, glycemic excursion, and planned exercise. I propose an evidence-based decision support tool, iDECIDE, to make recommendations to improve glucose control by taking into account meal and alcohol carbohydrates, glycemic excursion and planned exercise. iDECIDE is deployed as a low-cost and easy to disseminate smartphone application. A literature review was conducted on T1D and the state-of-the-art in diabetes technology. To better understand self-management behaviors and guide the development of iDECIDE, several data sources were collected and analyzed: surveys, insulin pump paired with glucose monitoring, and self-tracking of exercise and alcohol. The analysis showed variability in compensation techniques for exercise and alcohol and that patients made unaided decisions, suggesting a need for better decision support. The iDECIDE algorithm can make insulin and carbohydrate recommendations. Since there were no existing in-silico methods for assessing bolus calculators, like iDECIDE, I proposed a novel methodology to retrospectively compare insulin pump bolus calculators. Application of the methodology shows that iDECIDE outperformed the Medtronic insulin pump bolus calculator and could have improved glucose control. This work makes contributions to diabetes technology researchers, clinicians and patients. The iDECIDE app provides patients easy access to a decision support tool that can improve glucose control. The study of behaviors from diabetes technology and self-report patient data can inform clinicians and the design of future technologies and bedside tools that integrate patient’s behaviors and perceptions. The comparison methodology provides a means for clinical informatics researchers to identify and retrospectively test promising insulin blousing algorithms using real-life data. / Dissertation/Thesis / Doctoral Dissertation Biomedical Informatics 2017

Made available in DSpace on 2015-05-14T12:36:28Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 2509666 bytes, checksum: 468b10281902841768d6df68528b3e41 (MD5) Previous issue date: 2011-09-09 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Over and over health organizations are attracted by the benefits from the process of automation in health care through information technology, such as speed, accuracy, security, control and economy. Evidenced in recent years by the effort increas-ing of several countries for the development of projects that aim to leverage the use of information technology in health. Nevertheless, there is no standard, nor a set of stan-dards, to solve guaranteed all the problems of health organizations, due to the complexi-ty of the health field and the inherent discrepancies in each one of them. In addition, the use of electronic health record alone ensures only improvement on accessibility and rea-dability of information. To achieve satisfactory levels of patient safety, quality in health care, increasing efficiency and reducing the cost, it is necessary to use a clinical decision support system. Although it is notorious its relevance and there are several solutions us-ing clinical decision support, only few implementations for the clinic routine have achieved significant results in relation to improving results and procedures in health. The MultiPersOn-CDS aims to solve this problem through an extendible and customiz-able framework that enables, through the use of ontologies, the development of specia-lized contextual agents to support clinical decision making with a focus on clinical care and in the workflow of professional health, adressed mainly at improving the health care. For the construction and validation of MultiPersOn allows the creation of multi-purpose agents of clinical decision support, able to adapt to the inherent changes in the health field, making possible the dissemination of these agents by health organizations due to greater simplicity of make changes and customizations in the framework. As a result, efforts can then be concentrated on building clinical decision support agents, without worrying about the adequacy of them with applications, thus trying to reduce the technological barriers to the use and dissemination of clinical decision support tools in electronic health record. / Cada vez mais organizações de saúde são atraídas pelas vantagens oriundas do processo de automatização do atendimento em saúde, por meio da tecnologia da infor-mação, como agilidade, precisão, segurança, controle e economia. Fato evidenciado pelo aumento nos últimos anos dos esforços de diversos países para a elaboração de pro-jetos que visem alavancar o uso da tecnologia da informação em saúde. Apesar disso, não existe um padrão, nem tampouco um conjunto de padrões, que resolvam garanti-damente todos os problemas das organizações de saúde, devido à complexidade do do-mínio de saúde e as discrepâncias inerentes a cada uma delas. Além disso, o uso do re-gistro eletrônico em saúde isoladamente garante melhoria apenas na acessibilidade e legibilidade das informações. Para alcançar níveis satisfatórios de segurança ao paciente, de qualidade no atendimento em saúde, aumentando a eficiência e diminuindo o custo, faz-se necessário a utilização de um sistema de apoio à decisão clínica. Embora seja no-tória a sua relevância e haja várias soluções, utilizando o apoio à decisão clínica, atual-mente, há poucas implementações de apoio à decisão para a rotina clínica que tenham alcançado resultados significativos em relação à melhoria dos resultados e dos processos em saúde. O MultiPersOn-CDS objetiva solucionar esse problema, através de um fra-mework estendível e personalizável que possibilite, através do uso de ontologias, a ela-boração de agentes contextuais especializados para o apoio a decisão clínica com foco tanto na assistência clínica quanto no fluxo de trabalho dos profissionais de saúde, vi-sando sobretudo a melhoria da prestação de saúde. Com a construção e validação do MultiPersOn, permite-se a criação de agentes multipropósitos de apoio à decisão clíni-ca, capazes de se adaptar às mudanças inerentes à área de saúde, tornando possível a disseminação desses agentes pelas organizações de saúde, devido a maior simplicidade de realizar alterações e customizações no framework. Como resultado, os esforços podem então ser concentrados na construção de agentes de apoio à decisão clínica, sem se pre-ocupar na adequação deles com as aplicações, tentando-se assim diminuir as barreiras tecnológicas para a utilização e disseminação das ferramentas de apoio à decisão clínica nos registros eletrônicos em saúde.