DCBIOSTEC 2018 Abstracts


Full Papers
Paper Nr: 8
Title:

Dimension Reduction and Automated Evaluation in Retinal OCT Volumes

Authors:

Anna Breger, Bianca S. Gerendas, Ursula Schmidt-Erfurth and Martin Ehler

Abstract: Spectral-domain optical coherence tomography (OCT) is nowadays the most frequently used imaging technology in ophthalmology. It provides good cross-sectional visualization of retinal morphology, such as retinal fluid and disruptions in the photoreceptor region, which can be observed in various retinal disorders associated with vision loss. We aim for developing learning algorithms that yield automated quantification of retinal fluid and automated identification of disruptions in the photoreceptor regions. This is beneficial for daily clinical routine, because manual annotations are very time-consuming, may lead to inconsistent results and those tasks go beyond evaluation algorithms provided by manufacturers. Computation cost of data analysis methods, such as machine learning tools, depend highly on the dimension of the input data. To enable working with high-dimensional data (e.g. choosing a high-dimensional feature space) dimension reduction is required for preprocessing. My PhD project deals with derivation of numerical strategies for dimension reducing preprocessing steps in learning tools and its application to clinical OCT data of the human retina.

Paper Nr: 9
Title:

Novel Machine Learning Approaches in Multi-site Analysis for Autism Spectrum Disorders

Authors:

Elisa Ferrari, Maria Evelina Fantacci and Alessandra Retico

Abstract: Applying Machine Learning (ML) techniques on neuroanatomical Magnetic Resonance (MR) data, is becoming widespread for studying psychiatric disorders. However, such instruments require some precautions that, if not applied, may lead to inconsistent results that depend on the procedural choices made in the analysis, especially when the data under examination are extremely heterogeneous and many sources of bias are present. This is the case of studies on Autism Spectrum Disorder, in which the scarcity of data and the variability of this disease impose to examine data of subjects that differ both in the medical conditions and in the phenotypical characteristics. In this project two techniques that may be able to deal with these difficulties are proposed.

Paper Nr: 10
Title:

How Design can Improve the Quality of Internet based Treatment by Enhancing the Engagement of the Patient

Authors:

Rosaline Danielle Erica Barendregt

Abstract: Internet based treatment is a promising technique that is gaining interest in the fields of psychology and psychiatry. There already exist several internet based systems that are built on evidence from cognitive behavioural therapy. Randomised clinical trials show that use of such systems has a similar clinical effect as traditional face to face treatment, which is very promising. However, there is no evidence on how to optimise the user interaction design for such systems. This PhD project aims to prove the importance of user interaction design for internet based treatment, to find methods to measure the quality of the design, to find new design guidelines to use in internet based treatment programs, and to apply the findings to established treatment programs for people with mental problems.

Paper Nr: 12
Title:

Model Driven Engineering for Interactive Clinical Practice Guidelines Development - Healthinf 2018: Doctoral Consortium Submission

Authors:

Job N. Nyameino, Yngve Lamo, Khalid A. Mughal and Martin C. Were

Abstract: Clinical guidelines are systematically developed statements that assist practitioners and patients make decisions about appropriate health care for specific circumstances (Field and Lohr, 1992). They contain consensus based recommendations for clinical practice or public health policy. There is, however, a disconnect between published guidelines and their application in clinical practice. This is due to several factors including: lack of awareness and familiarity by providers, guideline complexity, poor layouts, poor accessibility and lack of clear intervention goals (Fischer et al., 2016). Our project aims to introduce a mobile-based guideline dissemination channel that is interactive and that is aimed at promoting active learning of the guideline content.

Area 1 - Software Agents and Internet Computing

Full Papers
Paper Nr: 11
Title:

Closing the Gap: Structuring the Unstructured FDA Adverse Event Report Narratives

Authors:

Susmitha Wunnava

Abstract: Detecting drug-related Adverse Drug Reactions mentioned in the Adverse Event narratives is imperative towards extracting valuable patient information from unstructured text into a structured thus actionable format. This then unlocks advanced data analytics towards intelligent pharmacovigilance. Existing approaches to ADR detection are either too coarse-grained such as document-level classification systems that detect all documents having a mention of the ADRs or, too fine-grained such as word-level classification systems also referred as Named Entity Recognition systems that detect only the word or phrases belonging to the ADR. A major limitation with these approaches is that, the information extracted does not capture the full context in which the ADR is mentioned and hence gives only partial information. To address these challenges, we are working on developing a middle-ground in between the above approaches, a sentence-level classification system that will detect the entire sentence which contains a mention of the drug-related ADR and therefore provide a bigger context in which the ADR was mentioned. In particular, this research focuses on deep learning based approaches ideal for sequence to sequence learning and prediction labeling tasks from medical narratives.