Title Computational Intelligence in (e)Healthcare: Challenges and Opportunities
Speaker Prof. Amir Hussain
Chair Henrique Martins

Abstract
A major discrepancy between optimal patient care and clinical care actually delivered has long been ascertained: this leads to an inefficient allocation of resources, ultimately impacting the quality of healthcare provision. Huge amounts of information, as part of the so called Big Data, is increasingly being collected in the healthcare domain both as structured and unstructured form: in this scenario, a first key challenge is how to effectively mine this information in order to build more intelligent models which can effectively assist clinicians to make optimal decisions.

Alternative models for traditional primary care have been actively explored, with many pilot predictive models successfully developed over the last few decades. Clinical decision support systems (CDSS) are currently aimed at supporting clinical processes and appropriate use of healthcare knowledge (e.g., diagnosis, investigation, treatment, short-and long-term care, etc.). However, the next generation of intelligent CDSS will be required to take account of many variables modeling habits and behavioral aspects (for example by exploring integration with wireless sensor networks): this will add more depth to the analysis of complex diseases (such as brain disorders), e.g., by capturing anomalies in behavioral patterns. Key features required to ensure successful deployment of intelligent CDSS into clinical practice will need to look beyond their predictive accuracy – desirable features are likely to include usability, transparency, integration with relevant clinical workflows, capability of providing real-time recommendations, etc. Computational intelligence techniques, many of which can be considered black-box models, can help improve the accuracy when compared to traditional predictive models which are generally based on linear or logistic regression techniques, yet are more transparent in nature.

A newly proposed hybrid computational intelligence framework tries to combine the benefits of these two approaches by maximizing the accuracy of the predictive model, whilst keeping it as transparent as possible. The initial statistical based features selection process is followed by the application of conventional linear or logistic regression (or other type of transparent classifiers if needed); which leads to a detailed analysis of the distribution of misclassifications across the input space. This, in turn, results in the development of a policy for intelligent selection of the best predictive strategy, making use of any appropriate machine learning technique for further reducing the misclassifications in required input spaces. In this stage of the framework, the transparency constraint is relaxed in spaces where the original model is not accurate enough, with the result that the overall ‘hybrid’ intelligent predictive CDSS model is still reasonably transparent but with further improved performance.

In addition to variables modeling habits and behavioral aspects, next-generation intelligent CDSS will also be required to take into account the so called ‘wisdom of the patient’. The resulting (e)healthcare system design and delivery of the future, in fact, will be based on community, collaboration, self-caring, co-creation and co-production using computationally intelligent technologies delivered via the Web. Engaging patients in their healthcare and encouraging people to take responsibility for protecting their health are seen as the best way to ensure the sustainability of healthcare systems. Patients can play a distinct role in their own care by diagnosing and treating minor, self-limiting conditions and by preventing occurrences or recurrences of disease or harm, by intelligently selecting the most appropriate form of treatment for acute conditions in partnership with healthcare professionals, and by actively managing chronic diseases. In this context, we explore the use of novel computational intelligence methods for physiological signals acquired in an unsupervised manner, to perform intelligent artefact rejection, automatic and real-time signal quality assessment, partial restoration of distorted signals, and most importantly classifier design for feature extraction. Emphasis is given on creating a universal, robust and next-generation computational intelligence framework for bio-signal analysis, for addressing a variety of chronic conditions such as neurological brain disorders and cardiovascular diseases. Further, for future intelligent CDSS to diagnose and provide support to patients more effectively, the acquired physiological signals (e.g. EEG, ECG etc.) will have to be properly interpreted. Any types of errors or artefacts present in the remotely acquired and transmitted signals may cause an inaccurate disease diagnosis which may lead to improper treatment.

In contrast to the above ‘patient-centred collaborative’ approach to healthcare, traditional paternalistic practice styles undermine people’s confidence in their ability to look after themselves, so replacing paternalism with the proposed ‘partnership’ approach could help enhance a sense of self-efficacy. A growing body of evidence demonstrates that patient engagement in treatment decisions and in managing their own healthcare, can lead to more appropriate and cost-effective utilization of healthcare services and better health outcomes. This shift in emphasis to e-health does not replace traditional healthcare models but rather complements them and will ideally become the prevailing ‘hybrid’ intelligent healthcare delivery model of the future.

An example of how such a process can take place is the newly proposed Sentic PROMs: a novel computational intelligence framework for measuring healthcare quality that exploits the ensemble application of standard PROMs and sentic computing to overcome common barriers to the use of health related quality of life (HRQoL) measurement systems, such as the respondent burden (time needed to complete the forms) and the need for staff to be trained to understand the results. Sentic PROMs, in fact, aim to be clinically useful and timely, sensitive to change, culturally sensitive, low burden, low cost, involve the patient and built-into standard procedures and needs to meet the requirements of regulators, payers and continuous quality improvement. In particular, to bridge the gap between the structuredness of questionnaire data and the unstructuredness of natural language data, both which are different at structural-level yet startlingly similar at concept-level, Sentic PROMs exploit both the semantics and sentics associated with patient opinions to accordingly aggregate such data and, hence, more accurately evaluate patients’ health status and experience in a semi-structured way, whilst tracking their physio-emotional sensitivity. We conclude that the sentic PROM framework has the potential to help improve the design and delivery of both existing as well as future hybrid (e)healthcare systems and services.

Biography
Professor Hussain obtained his BEng (with the highest 1st Class Honours) and PhD (with a resulting international patent on novel neural network architectures and algorithms) from the University of Strathclyde in Glasgow, in 1992 and 1997 respectively. Following a post-doctoral Research Fellowship at the University of Paisley (1996-98) and a research Lectureship at the University of Dundee in Scotland (1998-2000), he joined the University of Stirling in 2000, where he is currently Professor of Computing Science and founding Director of the Cognitive Signal-Image and Control Processing Research (COSIPRA) Laboratory. He has (co)authored/edited more than a dozen Books and over 200 papers to-date in leading international journals and refereed Conference proceedings. Since 2003, he has generated over €2m in research income (as principal investigator), including from UK research councils, EU FP6/7, international charities and industry. He is founding Editor-in-Chief of both Springer’s Cognitive Computation journal and Springer Briefs in Cognitive Computation, Associate Editor for the IEEE Transactions on Neural Networks & Learning Systems and serves on the Editorial Board of a number of other journals. He has served as invited speaker, general/program (co)chair and organizing/programme committee member for over 50 leading international conferences to-date. He is founding General co-Chair of the annual International Conference on Brain Inspired Cognitive Systems (BICS’2004-2013) and the IEEE ICEIS’2006. He is Chair of the IEEE UK & Republic of Ireland (RI) Industry Applications Society Chapter and Fellow of the UK Higher Education Academy.