Presentations and Publications

Selected Oral Presentations:

Bayesian Fusion Framework for Semantic SLAM in Endoscopy (Michel Hayoz¹, Christopher Hahne¹, Mathias Gallardo¹, Thomas Kurmann² ,
Maximilian Allan², Raphael Sznitman¹)

[1] Artificial Intelligence for Medical Imaging - ARTORG- Universität Bern

[2] Intuitive Surgical, Inc.

Presented by Michel Hayoz

Abstract: Surgical scene understanding plays a critical role in tomorrow’s intervention-assisting systems. Important applications comprise guidance systems in robotic systems and virtual overlays of pre-operative information both during surgery and for training purposes. A key task to enable scene understanding is semantic labeling of the anatomy and surgical instruments in 2D and 3D. To this end, we propose a novel semantic segmentation fusion framework that jointly employs 3D scene reconstruction and camera pose estimation from Simultanous Localization and Mapping SLAM to improve semantic segmentation of surgical scenes. Using these various estimates coherently, our framework is able to modulate the fusion of semantic labels by explicitly modeling two common challenges in MIS: (1) unreliable scene reconstructions and (2) static scenes. We show in our experiments that our method performs better than state-of-the-art methods and demonstrates practical improvements through qualitative evaluation.

Aiming for more objectivity in creativity assessment – Applying word vectors on creativity data (Magdalena L.O. Camenzind¹, Michael Single^2,3, Stephan M. Gerber^2,3, Tobias Nef^2,3, Claudio L. Bassetti³, René M. Müri^1,2,3, and Aleksandra K. Eberhard-Moscicka^1,3)

[1] Perception and Eye Movement Laboratory, Departments of Neurology and BioMedical Research, University Hospital Inselspital, University of Bern, Bern, Switzerland

[2] Gerontechnology and Rehabilitation Group, ARTORG Center for Biomedical Engineering Research, University of Bern, Bern, Switzerland

[3] Department of Neurology, University Hospital Bern and University of Bern, Switzerland

Presented by Magdalena Camenzind

Abstract: Divergent thinking (DT), as a compound of creativity, refers to an ability to produce multiple solutions to a given problem. The output of divergent thinking tasks can be assessed based on semantic similarity (i.e., flexibility), which is often subjectively rated, hence influenced by inter-individual variations in the perception of semantic similarities. In this study, the assessment of flexibility was objectified by applying word vectors of the German language to quantify the semantic similarity of words. To this end, a word vector model was trained on a diverse 0.67-billion-words German text corpus. The validity of the self-trained model was demonstrated on two datasets of human word similarity judgments whereby the performance was in line with the performance of publicly available pre-trained word vectors. Additionally, synonymous words were linked and assigned to the same vector in order to reduce an overestimation of flexibility amongst synonymous ideas or answers produced in divergent thinking tasks. Three different metrics to calculate a final flexibility value were proposed, allowing an individually tailored assessment of participants’ flexibility performances. Overall, this novel, non-confounded by fluency, and objective flexibility assessment tool has a potential to enable a more time- and cost-effective assessment of creativity performance.

A unified toolchain for the living systematic review of mental health during the COVID-19 pandemic (Alexander Holloway¹, Georgia Salanti¹)

[1] Institute for Social and Preventive Medicine, University of Bern

Presented by Alexander Holloway

Abstract: It is not yet clear what effect the COVID-19 pandemic and its containment measures have had on the mental health of the general public. To address this, we have created a living systematic review that uses Bayesian dose-response modelling to meta-analyse studies that report longitudinal data on the general population’s mental health. We make use of crowdsourcing to identify and subsequently extract data from relevant papers, in a growing corpus of over 100,000 studies. Our crowd consists of over one hundred medical and research professionals from all over the world, who have undergone training to identify data that are relevant to our overarching study aims. To our knowledge, this is the largest crowdsourced living systematic review that has been performed to date, and as such, no pre-existing tools are available to facilitate our desired workflow. Here, we present the toolchain we developed to facilitate study screening and data extraction, data cleaning and feature enrichment, and data analysis. The results of these analyses are subsequently published in the form of interactive visualisations on our project’s website (https://mhcovid.ispm.unibe.ch). This work forms part of an ongoing study, so only preliminary results of some project aims will be presented.

The Bern Sleep Database: Clustering of Patients with Sleep Disorders (Florence Aellen¹, Julia Van der Meer², Anelia Dietmann², Markus Schmidt², Claudio L. A. Bassetti², Athina Tzovara^1,2)

[1] Institute of Computer Science, University of Bern

[2] Department of Neurology, Inselspital, Bern University Hospital and University of Bern

Presented by Julia Van der Meer

Abstract: Not publicly available.

Poster Publications

Poster presenters are invited to voluntary upload their poster submission to the Bern Data Science Day 2022 Project on the BORIS Publications University of Bern repository with DOI for reference.