Keynote Lectures

Abstract
Machine learning is a key technology in medical image analysis. From unsupervised clustering of data to supervised learning of categories, statistical modeling tools are used across all modalities, from the molecular to MRI brain imagery. A leading trend in the machine learning community is Deep Learning which was termed one of the 10 breakthrough technologies of 2013 (MIT Technology Review, 2013). Within the Deep learning schemes, convolutional neural networks (CNNs) have become the most powerful technique for a range of different tasks in imaging and computer vision domains. CNNs are machine learning models that represent mid-level and high-level abstractions obtained from images.  An overview of the leading machine learning tools will be presented in the talk for various medical imaging tasks. I will also present the initial work emerging with Deep learning in this domain, reviewing the challenges facing this field and its initial promising results.  

Abstract
Advances in computer-graphic technologies in realistically simulating aspects of human appearance, motion and (interactive) behaviour have lead to the question: How does engaging and interacting with ever more realistic virtual humans actually influence human subjective experience and behaviour? One general and seemingly counter-intuitive answer, formulated in the Uncanny Valley Hypothesis, is that relatively high levels of anthropomorphic realism are likely to evoke an unpleasant affective state characterised by feelings of strangeness and the uncanny. Consideration of the inconsistent findings in this new field of research suggests that this general answer is likely to be superseded by a more differentiated perspective that better reflects the psychological complexities of human nature. This perspective suggests that there will be great variability across individuals, stimuli, situations, tasks and over time in the relationship between anthropomorphic realism of virtual humans (or particular aspects of virtual humans) and human experience and behaviour. One of the many challenges that arises from this perspective lies in understanding how realistic a virtual human (or particular aspects of the virtual human) needs to be – in view of differences between individuals, stimuli, situations, tasks and time – to affect the outcomes for which a virtual human is designed. Psychology offers insight into this challenge and the opportunities that follow from this perspective.

Abstract
Over the past decade researchers have been developing micro- and nanorobots for use as biomedical platforms with applications such as chemical sensing and drug delivery. Understanding and controlling the physical and chemical interactions at the micro- and nanoscale is crucial for the realization of small biomedical robots. One of the main aspects investigated has been the fabrication and optimization of the motility component of these small agents, and one of the most promising approaches is to use electromagnetic systems to wirelessly control and actuate magnetic micro and nano structures. A goal of the research at the Multi-Scale Robotics Lab consists of creating untethered magnetically controlled micro and nanorobots to make current medical procedures safer and less invasive, and to create entirely new procedures that were never before possible. To increase their performance and to provide additional biofunctionalities (biocompatibility, drug delivery, sensing), other materials must be incorporated. In this work, we will present several magnetic micro- and nanoagents that have been produced in our laboratory with a focus on biomedical and environmental applications.

Abstract
The energies involved in weak chiral interactions occurring between odorant binding proteins (OBPs) and carvone enantiomers are evaluated, down to a few KJ/mol, by means of a water-gated organic field-effect transistor (WGOFET) whose Au-gate is modified with a porcine-OBP (pOBP) self-assembled monolayer. The output current measured is dependent on the concentration of the analyte and pM concentrations can be detected. The binding curves also are significantly different between the two enantiomers. The modelling of the two curves allows the energies associated with the OBP-carvone complexes formation to be independently extracted, from the very same set of data. From the dissociation constants the standard free-energy the complex formation at the electrode is derived, while the threshold voltage shifts gives information on the electrostatic component. This approach, representing a unique tool to quantitatively investigate low-energy bio-chemical interactions, is rather general as it relies on the relative dielectric constants of the protein-SAMs and of the organic semiconductors being much lower than that of water. The role of the OBPs in the olfaction system is still under debate and the detection of neutral odorant species at the pM level by means of a WGOFET adds relevant pieces of information to the understanding of the odor perception mechanism at the molecular level.