Digital Health

Statement on Digital Health

"Machine intelligence is a game changer in the health sector that will bring improvements in precision and efficiency of medicine and increase the wellbeing of individuals. Digitalization already today innovates the health sector to organize data, to diagnose, to plan and to execute interventions. We must now explore way of unleashing the potential of data science. Innovation here requires the cooperation of many disciplines in medicine and technology and must be complimented by an understanding of the patient needs. It is key that machine decisions are explainable and controllable by the medical personal to improve health care provision while keeping it safe and humane to the individual. We invite you to propose concepts, discuss viability and spark ideas to innovate health with data science."

---

 

Statement on Digital Health

"In the healthcare community, there are currently few topics that are as frequently discussed as digital health. At the same time, there is probably no topic with a greater degree of uncertainty in terms of its real impact on our healthcare system.  Therefore, every healthcare player needs to systematically develop a road map for how to deal with the endless medical advances and business models that digital health solutions will enable."

Abstract

Talk title: Digital Health – The Future of Swiss Healthcare
Authors: Alfred Angerer
Affiliations: ZHAW – Winterthur, Institute of Healthcare Economics

The purpose of this presentation is to provide an overview of the digital health market in Switzerland. Three basic questions will be answered:

1. What should I know about digital health?
2. Why should I care about this trend?
3. How can I react to this trend?

The first part of the presentation (What?) provides a basic overview of the topic. The author will then present a logical categorization of digital health solutions that he has developed. A current estimation for the size of the digital health market will also be shown.

The second part (Why?) focuses on the current and future impact of digital health. At the moment in Switzerland, digital health is still in a very early phase. Discussions among practitioners are focused on relatively simple technological solutions, such as electronic health records. A survey asking Swiss practitioners about the future development of the Swiss market will be presented. However, the results are mostly ambiguous, reflecting the current, extremely dynamic character of the sector.

The presentation will close with a call to action for the players in the market (How?). Since companies in the health sector will only be able to benefit from this economic opportunity if they develop the right business strategy, the presentation focuses on how digital health can provide a competitive advantage. It is strongly recommended that practitioners develop a sound, systematic approach to the topic and start quantifying precisely the economic and medical impact of digital health solutions.

---

 

Statement on Digital Health

"In the 19th century medicine was going through transformations from a butchering art to a practice based on natural science. The first stethoscope was introduced, general anesthetics became available, and the germ theory of disease and antiseptics began overtaking miasma theory in a Kuhnian paradigm shift. Medicine transformed into a profession requiring specific expert knowledge. The 20th century was characterised by developments in chemistry, genetics, surgery, radiography and biological treatments, further strengthening the professionalisation.

According to Thomas Kuhn, science develops through paradigm shifts. New paradigms contradict old models, create tensions, and eventually become the new models. Digital Health is arguably such a shift too. The transformation is extremely exciting from a technological perspective, and it is not less challenging from a cultural one. The computational technologies will have implications in research and development, the operative processes in health and care institutions, and the management and care of patients.  In particular, digitalisation will almost surely shift the patient-caregiver relationship further away from a paternalistic model and empower patients in decision making. Thus, stakeholders of health-care face the challenge of safely integrating digital solutions, taking into account, for instance, the patient's health literacy and the risk of dehumanisation of care."

---

 

Abstract

Talk title:  Computational methods for the analysis of intra-tumour heterogeneity in single-cell dnaSeq data
Authors: Katharina Jahn, Jack Kuipers, Jochen Singer, Niko Beerenwinkel
Affiliations: ETH Zurich, SIB

The mutational heterogeneity observed within tumours is a key obstacle to the development of effective cancer therapies. A thorough understanding of subclonal tumour composition and the underlying mutational history is essential to open up the design of treatments tailored to individual patients. Present studies on tumour evolution are primarily based on sequencing data obtained from bulk tumour tissue that is comprised of hundred thousands or millions of cells. The admixed mutation profiles obtained in such bulk studies often underestimate the mutational heterogeneity of a tumour.

Through recent technological advances it is now possible to sequence the DNA of individual cells. This opens up not only the possibility to analyse the evolutionary history of tumours at an unprecedented resolution but also to leverage the potential of circulating tumour cells whose mutational profiles are of particular interest to the analysis of metastatic seeding patterns. The transition from bulk to single-cell sequencing data poses a number of statistical challenges such as elevated noise rates due to allelic drop out, missing data and contamination with doublet samples.

We developed a Bayesian inference scheme for tumour mutation histories based on single-cell sequencing data [Jahn et al., 2016]. In this talk I will focus on two recent extensions of this work, a novel single-cell mutation caller that takes the underlying cell phylogeny into account [Singer et al., 2018] and a rigorous statistical test to identify the presence of parallel mutations and mutational loss [Kuipers et al., 2017]. Our results on simulated and real tumour data show that a thorough modelling of the noise inherent to single-cell data allows for an accurate reconstruction of tumour mutation histories.

[Jahn et al., 2016] Jahn, K., Kuipers, J., and Beerenwinkel, N., 2016. Tree inference for single-cell data. Genome Biology, 17:86.
[Kuipers et al., 2017] Kuipers, J., Jahn, K., Raphael, B. & Beerenwinkel, N., Genome Research 27 (11), 1885-1894.
[Singer et al., 2018] Singer, J. , Kuipers, J, Jahn, K. & Beerenwinkel, N., In revision.

---

 

Statement on Digital Health

"AI has the potential to bring much deeper understanding of disease biology by combining data from very different modalities – for instance, digital pathology images and transcriptomics – into coherent models. Diagnostic and prognostic models have been successfully applied to various diseases, and in 2017, an AI-based diagnostic model was FDA approved. The success of such approaches relies heavily on digitizing and organizing data in a machine-learnable way."

Abstract

Talk title: A tour of Novartis Oncology Bioinformatics
Authors: Eric Y. Durand
Affiliations: Novartis Institute of Biomedical Research (NIBR)

In this presentation, I will give an overview of the bioinformatics activities in the Oncology Disease Area at NIBR. I will start by presenting the NIBR organization. I will then illustrate our research with two vignettes from recent publications. With the first vignette, I will give an overview of the DRIVE project, in which we performed a large scale genomic functional screen. In DRIVE, we targeted 7500 genes in 300 cell lines with shRNAs, uncovering many cancer dependencies. In the second vignette, I will present a forward genetic screen using the PiggyBac transposon system. In this project, we used insertional mutagenesis to study the genetic dependencies of a large variety of cancers. We also used this system to interrogate resistance mechanisms to our TP53-MDM2 inhibitor, HDM201.

---

 

Statement on Digital Health

"The wealth of data available today for health and science provides great opportunity for improving our knowledge and predictions, from basic science to personalized medicine. It also poses unique challenges due to the complexity of the data, the weak signal, and the sheer quantity of data to be processed. The development of personalized medicine has led to a few exciting breakthroughs, for instance in cancer treatment, but we also see major struggles in translating discoveries to clinical settings and in applying methods to complex diseases with many confounding factors. Thus, additional methods are needed to sift through digital health data to extract interpretable insights and actionable predictions."

Abstract

Talk title: Computational prediction and phylogenetics for designing peptide binders
Authors: Spencer Bliven and Maria Anisimova
Affiliations: ZHAW Institute of Applied Simulation, Swiss Institute of Bioinformatics

The ability to recognize specific proteins using antibodies was a revolution in biotechnology, and protein binders have become a major product category in the pharmaceutical industry. Protein binders play a fundamental role in biochemistry techniques and have led to breakthroughs in clinical, therapeutic and diagnostic settings. Several techniques are available for creating protein binders for new targets, but at present they require experimental selection that is expensive and time consuming. For this reason, it would be desirable to computationally design protein binders for novel targets with high sensitivity and specificity. We present a machine learning model to predict the affinity of designed armadillo repeat proteins (dArmRP) to arbitrary polypeptide targets, providing the first step towards computational design.

The Armadillo repeat protein family (ArmRP) bind extended peptides. They consist of approximately 42-residue repeats, each of which interact with two amino acids from the target. A thermostable designed ArmRP with a consensus repeat has been previously designed (1) and demonstrated to have modular binding affinity.(2)

A model to predict the effect of mutations on the binding affinity to various targets was constructed. Training data was provided by the Plückthun lab (Biochemistry, University of Zürich). Features describing the mutations were extracted as per Atchley (3) and used to train a logistic regressor using a linear kernel.

ArmRPs offer an attractive modular system as each ArmRP unit binds a consecutive dipeptide of the target.(1) This simplifies the design problem to the task of identifying dipeptide-binding modules which can be concatenated to bind targets with high affinity. Using the regression model of binding affinity of individual models to dipeptides, we hope to be able to design ArmRP mutations to engineer a wide range of specific target binding.

Acknowledgements

This research was supported by COST Action BM1405 and COST Switzerland SEFRI project IZCNZ0-174836. Training data was provided by Andreas Plückthun, University of Zurich.

References

1. Reichen C, Hansen S, Plückthun A. Modular peptide binding: from a comparison of natural binders to designed armadillo repeat proteins. J Struct Biol. 2014 Feb;185(2):147–62.
2. Hansen S, Tremmel D, Madhurantakam C, Reichen C, Mittl PRE, Plückthun A. Structure and Energetic Contributions of a Designed Modular Peptide-Binding Protein with Picomolar Affinity. J Am Chem Soc. 2016 Mar 16;138(10):3526–32.
3. Atchley WR, Zhao J, Fernandes AD, Drüke T. Solving the protein sequence metric problem. Proc Natl Acad Sci USA. 2005 May 3;102(18):6395–400. 

---

 

Statement on Digital Health

Digital health opens up incredible opportunities in terms of better and more efficient health care but the fast development must be accompanied by solid validation guidelines and relevant regulations to assure high quality for the patients and end users.

Abstract

Talk title: Wearables and machine learning for personalized fertility predictions
Authors: Lisa Falco       
Affiliations: Ava AG

How data science can empower women with powerful health insights throughout their entire reproductive lives. Ava is currently using a combination of machine learning and wearable technologies to help people around the globe become parents.

By measuring nine physiological parameters while sleeping, the Ava bracelet tracks the effects of the hormones changing throughout the menstrual cycle. This enables women to better understand their bodies and the Ava algorithms are helping them to pinpoint their fertile days to speed up the time to conception. 

Ava’s algorithms have stepwise developed from expert algorithms to AI as the user database has gone from non-existent to big data. The use of deep learning now allows to capture more individual variations and speed up the time development pace.

---

 

Statement on Digital Health

"The past few years has seen a tremendous increase in the use of machine learning methods in the medical domain, with some impressive results in medical imaging. These methods provide automated tools to physicians that help diagnosis, prognosis, and treatment planning at the individual patient level.  At the same time, the availability of open radiology data, with some datasets now exceeding 100,000 images, has allowed a vast increase in the complexity of these methods. A recent trend in this regard is the open availability of genomic data together with imaging data, with promising results suggesting that precision medicine should rely on methods that can integrate imaging and clinical data with genomic and other -omic data. These are exciting times, where algorithms, computing power, and data availability combined with engineering and medical expertise will come together to greatly improve patient outcomes for many diseases."

Abstract

Talk title: Predictive radiology for precision medicine: medical imaging meets -omics
Authors: Jonas Richiardi
Affiliations: Lausanne University Hospital, Department of Radiology, and Siemens Healthcare, Advanced Clinical Imaging Technology

High-resolution medical imaging data, together with large-scale genotype data and post-mortem gene expression data, are being generated and shared openly at an increasing pace. This opens tremendous opportunities to uncover new relationship between imaging, which can be performed non-invasively in-vivo and has very good spatial resolution, and abnormal molecular mechanisms potentially underlying abnormal findings such as brain atrophy.

For some diseases, such as multiple sclerosis or dementia, proteomic data is also collected in clinical practice, and can lead to much improved forecasts of clinical outcomes by combining imaging markers and liquid biomarkers.

Even without -omic data, predictive methods are increasingly gaining ground in medical radiology. In all cases, the key is the development of novel computational techniques that can link biological levels and enable individualised prediction and treatment.

In this talk, we will give an overview of how machine learning techniques are used in radiology, and present some of our ongoing work that seeks to combine proteomic data or genomic data with imaging in order to improve differential diagnosis, personal prognosis, and treatment planning.

---

 

Statement on Digital Health

"Continuous monitoring of vitals by wearable devices will be an important part of preventive healthcare. Heat-flux based wearables for core-body temperature monitoring have the potential to deliver early warnings for vascular or neurodegenerative diseases, to detect sleep disorders, stress levels and burn-out, and to optimize timing of exercise or drug administration."

Abstract

Talk title:

Core-body temperature measurement solution for wrist wearables based on heat flux sensors

Authors:

Krzysztof Kryszczuk 1
Lukas Durrer 2
Michele Zahner 2
Ilia Britvich 2
Philipp Eib 2
Anneke Godeschalk 3
Sandra Röthlisberger 3
David Schreier 3
Johannes Mathis 3

Affiliations:

1 Predictive Analytics Group, Institute of Applied Simulation (IAS), School of LSFM, ZHAW
2 GreenTEG AG, Zurich
3 Inselspital, Bern

With increasing population and changing demography a paradigm change from reactive to preventive healthcare is required. Wearable companies, e.g. Fitbit, Apple, Huawei and Samsung, see the opportunity to push wearables into volumes comparable to smart phones. For early detection of diseases like heart failure, Parkinson or Alzheimer’s, a continuous monitoring of vitals is required with users only accepting wrist devices. The main challenge is to monitor vitals with high accuracy at distal body positions. greenTEG developed a core-body temperature sensor for wearables enabling measurements at the wrist. A proof-of-concept algorithm was developed within a CTI-founded clinical trial in collaboration with the Insel hospital in Bern and the ZHAW Wädenswil.

 

---

 

Statement on Digital Health

"Digital technologies are rapidly expanding into healthcare with anticipated exponential growth in the coming years. Their potential impact is broad and extends far beyond an app-store gimmick. For example, new motion-capture systems and sophisticated biosensors in common commercially available devices are now used by both researchers and clinicians. Still, a number of important challenges remain, such as integration of digital technologies into the emergency room and physician workflow, and validation of novel measures intended to augment or replace patient-reported outcomes."

---

 

Statement on Digital Health

"We have the opportunity to shape and target the use and application of digital health to achieve solutions that are sustainable, efficient and have a real benefit for the human."

Abstract

Talk title: Solving real life problems – bridging user needs and engineering for digital health solutions
Authors: Dr. Eveline Graf, Prof. Dr. Markus Wirz
Affiliations: ZHAW, School of Health Professions, Institute of Physiotherapy

A user-centered design approach for new technologies focuses the development process on the needs and requirements of the future users of the technology and does not limit research and development actions on functionality. The aim of this approach is to maximize acceptance and implementation success. There is always a variety of users and stakeholders. In the health care sector, this can be patients, relatives, therapists, physicians but also insurances, politicians as well as companies with commercial interests. Considering the visions of each of the users and stakeholders in a project and translating this into a technical solution requires the involvement of a variety of professionals with different expertise. This can result in a stimulating and productive work environment but also poses a variety of challenges that need to be addressed proactively.

---

 

Statement on Digital Health

"Exploring big data by machine learning helps to know. Model-based data analysis helps to understand. Those are two different things. To make progress in medicine, understanding will be pivotal."

Abstract

Talk title: Model-based analysis of time-resolved Comet assay – A prototype for evaluating dynamics in scattered data?
Authors: Stephan Scheidegger ^1,2, Mathias S. Weyland ^1,3, Katarzyna J. Nytko⁴, Pauline Thumser-Henner⁴, Carla Rohrer-Bley⁴, Rudolf M. Füchslin ^1,5
Affiliations: ¹ ZHAW School of Engineering, Winterthur; ² Kantonsspital Aarau; ³AMI, Université de Fribourg; ⁴Radio-Onkologie, Vetsuisse-Fakultät Universität Zürich; ⁵ECLT European Centre for Living Technology, Venice, Italy

Purpose: Clinical trials as well as biological experiments in vitro often deliver “real-world” – data. The variability in the investigated population (individuals such as volunteers, patients, animals, microorganism or cells) can lead to remarkable scattering of measurements. The resulting large uncertainty (error bars) may hinder a model-based data analysis of time-resolved data with dynamic models, when using averaged values. In addition, the (possibly temporally changing) distributions of measured values may contain important information about the investigated system. An illustrative example is the analysis of time-resolved Comet assay data, where DNA damage is quantified on a per-cell granularity. Even in a genetically well-defined cell line, the reaction of the cells upon radiation can vary remarkably. The cause for this variation lies in the combination of the individual response (repair pathway activation) and the stochastic nature of damage induction by ionizing radiation (induction of a certain number of double strand breaks at various potential target sites on the chromosome). Thus, irradiating cells and quantifying the damage with the Comet assay yields a distribution of varying amounts of DNA damage per cell. This distribution encodes information about repair - and cell death processes triggered by the treatment. The parameter search by fitting a radio-biological model (Multi-Hit-Repair – MHR - model; Scheidegger et al. 2013) using median values of time-resolved Comet of soft tissue sarcoma data resulted in a parameter set which predicts the survival of a sarcoma cell line in vitro (Chaachouay et al., 2015). However, it was also possible to find different sets of parameters describing the same survival curve (Weyland et al. 2018). Survival curves at fixed dose rates / conditions are therefore not suitable for determining parameters of the of the MHR model. Better parameter estimation can be achieved by varying the dose rate and experimental conditions. This results in a strong increase of experimental work. Therefore, we investigated the possibilities of fitting the MHR model to the experimentally observed distribution of Comet assay read out for extracting information about the underlying dynamics of the cellular response onto combined hyperthermia-radiotherapy (HT-RT).

Materials & Methods: We used the alkaline Comet assay for evaluating the effect of HT-RT. For this, cancer cells were exposed to heat (42°C) and ionizing radiation (6 Gy). For a Comet assay, cells selected for analysis undergo lysis to release chromatin from the nucleus. DNA fragments produced by the interaction of DNA with radiation are dragged away from the nucleus by an electrophoretic process. These (labeled) DNA fragments will become visible in fluorescent microscopic images as a “comet tail” attached to the remaining bulk DNA of the nucleus. For every time point after irradiation, approximately 100-150 cells per tumor cell line and experimental condition have been analyzed.

The Multi-Hit repair (MHR) – model describes a chain of (tumor) cell populations. Each population is characterized by a defined number of potentially lethal DNA damages (hits). The hit induction rate is assumed to be linear proportional to the radiation dose rate. Damages can be repaired or eliminated (cell death mechanisms such as apoptosis). The repair probability is governed by a transient biological dose equivalent. The direct model outputs are time-resolved population sizes. The surviving fraction can be directly evaluated from the number of vital tumor cells after completion of repair processes. From the population size, the number of damages and subsequently the upper and lower limits of DNA fragments as function of time can be estimated. In addition, a histogram of the population sizes can be compared to the experimentally observes distributions.

Results:

The comparison of experimentally observed and calculated distributions (Fig.1, Weyland et al., 2018b) exhibits clear differences: The calculated results in the bottom half of the figure were generated from MHR model parameters found when analyzing survival curves of a different cell type (Scheidegger et al. 2013). Thus, quantitative differences are to be expected. Yet the MHR model is able to predict distributions qualitatively similar to those observed experimentally.

Fig.1. Fig.1.Distributions of DNA fragments in tail (Weyland et al., 2018b), experimentally observed (top) and calculated by the MHR-model (bottom): The histograms (distributions) are plotted symmetrically around the time point selected for analysis (time after irradiation).

Discussion and Conclusions: Fitting time-resolved Comet data with the MHR-model is a very specific application of a method which potentially has a large field of applications. The essential ingredients for this are (time-resolved) distributions of measurements (data), a dynamic model delivering distributions of simulation results as function of time which can be compared to the experimental data and an algorithm allowing fitting of the simulated distribution to the experimentally observed distribution.

References:

Chaachouay H, Scheidegger S, Schulz N, Grosse N, Füchslin RM, van Loon B, Rohrer Bley C (2015): Monitoring of DNA damage and repair kinetics during radiotherapy in vivo: a minimally invasive approach using the dog as a model. Molecular Radiation Biology / Oncology, 11, 22

Scheidegger S, Fuchs HU, Zaugg K, Bodis S, Füchslin RM (2013): Using State Variables to Model the Response of Tumour Cells to Radiation and Heat: A Novel Multi-Hit-Repair (MHR-) Approach. Computational and Mathematical Methods in Medicine, 2013, http://dx.doi.org/10.1155/2013/587543

Weyland MS, Thumser-Henner P, Nytko KJ, Rohrer Bley C, Füchslin RM, Scheidegger S (2018a): Extracting information about cellular repair processes after hyperthermia – radiotherapy by model-based data analysis – ambiguities in survival prediction as a challenge? Strahlenther Onkol (2018) 194, 503–504, doi: 10.1007/ s00066- 018- 1295-1

Weyland MS, Thumser-Henner P, Rohrer Bley C, Scheidegger S, Füchslin RM (2018b): Dynamic DNA damage and repair modeling: bridging the gap between damage readout and model state. XIII Italian Workshop on Artificial Life and Evolutionary Computation

---

 

Statement on Digital Health

"Digital health will power the individualization and dematerialization of health care. Patients will be more empowered to manage prevention, diagnosis and treatment. They can start earlier and access digital help more frequently. However the interface from the analog health care to the digital health services is messy and full of misunderstandings. The future of health requires easy moving between digital services and conventional care. I am working at Thryve Our software enables health services to access health data from smartphones, smart watches, fitness trackers and other networked devices such as scales or blood pressure monitors via a single API."

Abstract

Talk title: Tracker data in mental health symptom tracking
Authors: Paul Burggraf
Affiliations: mHealth Pioneers GmbH

Major depressive disorder is the biggest cause of disability throughout the world, 25% of all sick leave days are due to depression. However, 50% of major depression cases are not detected in primary care settings and thus cannot be treated. Screening for depression is still predominantly dependent on symptom-savvy family members and physicians - a help that many cannot access. Research has recently shown that symptoms of mental health issues can be found early on in sleep and activity data, as well as in communication meta data. This talk will discuss the difference between symptoms trackable in one-on-one assessments and the symptoms trackable in automated data analysis. Furthermore, I will present a setup for a hybrid screening solution that combines the two paradigms for symptom tracking for a novel depression screening.

---

 

Statement on Digital Health

"Digital Health is the key to personalized health care. Compiling individualized treatments for each patient bears the potential to boost our health care system to the next level. The best clinical practice will be patient-centered and data-driven. The basis for such a system relies on efficient and secure data management and data sharing."

---

 

Statement on Digital Health

"Digital health needs an ecosystem where health data can be published, shared and reused for better medical services and scientific research. Thus regulating data access and use rights respecting personal health data protection regulations as well as public and scientific interest in medical progress and private business interests in health products and services are fundamental for the development of an open and trustworthy health data ecosystem. This data ecosystem should be based on three principles: open data, my data and shared data."

Abstract

Talk title: Swiss Data Space – Perspectives for Health Data Politics in Switzerland
Authors: André Golliez
Affiliations: Swiss Data Alliance

Digital health needs an ecosystem where health data can be published, shared and reused for improved medical services and scientific research. Thus regulating data access and use rights respecting personal health data protection regulations as well as public and scientific interest in medical progress and private business interests in health products and services are fundamental for the development of an open and trustworthy health data ecosystem. This data ecosystem should be based on three principles: open data, my data and shared data.

• "Open Data" are data that are not related to a person. That can be used without restriction and are produced generally by public administrations, government-related enterprises or publicly funded research institutions.
• “My Data» are personal data provided by a company or an administration of the data subject to enable them to switch easier to another service provider as provided for in Article 20 of the European Data Protection Regulation (data portability).
• "Shared Data" are sensitive data that can only be used under restricted conditions by a selected group of users. "Data Sharing" is an important option for data reuse beyond organizational borders between private companies or between private companies and public administrations.

The task of health data politics is to set frame work conditions for an open and trustworthy health data ecosystem in Switzerland: regulating data access and reuse rights on the legal level, financing the necessary data publication and sharing infrastructures, defining technical standards enabling open exchange of health data, and last but not least establishing a data sharing culture based on strong ethical principles.

---

 

Abstract

Talk title: Clinical data sharing: a data scientist's perspective
Authors: Norman Juchler^1,2, Sabine Schilling¹, Kazuhiro Watanabe^1,5, Daniel Rüfenacht⁴, Philippe Bijlenga³, Vartan Kurtcuoglu², Sven Hirsch¹
Affiliations:

1. Zurich University of Applied Sciences (ZHAW), Wädenswil, Switzerland
2. University of Zurich (UZH), Zürich, Switzerland
3. University Hospital Geneva (HUG), Geneva, Switzerland
4. CABMM & Klinik Hirslanden, Zürich, Switzerland
5. Tohoku University, Sendai, Japan

An ever increasing amount of medical data is collected and used for scientific and clinical purposes. To benefit from the abundance of data, however, one has to deal with several challenges. The diversity of data sources, the variability seen in the biological systems and the biases and distortions inherent in the acquired data request for robust and flexible data processing pipelines.

Here, we illustrate some of these challenges at the hand of our research on intracranial aneurysms and share insights how to deal with these challenges. We present the AneuX AneurysmDataBase. It stores data acquired at multiple clinical centers, supports heterogeneous data (clinical data, imaging data, genetic data, morphological and histological data, etc.) and is aimed for use in both scientific and industrial contexts. We further present five scientific studies that demonstrate the usage of the AneurysmDataBase. In the first application, we evaluated the PHASES score, a recent scoring scheme to guide the clinicians whether to treat an unruptured intracranial aneurysm. We further examined and improved existing morphological descriptors with the goal to associate aneurysm shape with its disease status. In a third study, we quantified the qualitative rating of aneurysm shape by humans. A fourth study aims at inferring information about the disease directly from imaging data by means of convolutional neural nets. Finally, we sketch how to query aneurysms with similar anatomical and morphological properties from a database.

With our work, we demonstrate how clinical data sharing can be used for quantitative analyses of aneurysm properties and for the development of diagnostic and prognostic tools.