Fighting Covid-19 Challenge

A platform for open research on large Covid-19 imaging datasets

7

Cooperating Clinics

71

Processed Covid-19 Datasets

140

Number of Pre-Registrations

Medical imaging is potentially well suited for Covid-19 diagnosis. This challenge is about connecting the best brains to support doctors with artificial intelligence systems.

Our Motivation

Medical Imaging

Medical imaging, in particular CT, is potentially well suited for Covid-19 detection and diagnosis. It could be faster than “traditional” testing.

Deep Learning

The first applications of deep learning on CT data show a promising result. Studies, however, published no benchmarks and only a few models.

Data Labeling

High quality labeled multicenter training data is not accessible to developers. Regulations and computing resources are a challenge!

Artifical Intelligence

AI on images is useful beyond diagnosis. Prediction of patient permanence in the ICU, disease progression and survival analysis are key to save lives.

Our Plan

Build a dataset

Multi-center imaging dataset obtained from various centers globally

Allow access

Full access for data contributors (hospitals) early on. Supervised access for third parties.

Provide compute

People submit their model on shared computational resources provided by us.

The Challenges

Four challenges tackling clinical problems and hypothesizing that those can be solved by smart image and data processing methods. The four tasks will be defined by the clinical and research partners at the end of stage 1.

One challenge is open, i.e. participants can solve a problem they defined themselves.

Each team has to commit to solving one or two of those challenges at the time of registration.

The Challenge Comes In Two Stages

Phase 1 Dataset Aggregation

Dataset aggregation from various global clinical partners. Our contributing partners (academic institutions providing patient data) have the right to access the entire database and can begin their research. Clinicians and scientists will develop goals and hypotheses that will define the challenges.

Phase 2 Coding

Third parties can contribute and run their code on the datasets without downloading them. Thus, patient rights are protected while maintaining maximum data access.

Timeline 2020

March 30

Data collection starts: Data collection in biobank. Data curation and standardization.

April 6

Stage 1: First batch of data has been provided. Collaborative definition of the challenges. Data labeling.

End of May

Stage 2: The challenges are disclosed. Teams can register, download sample data and prepare.

June

Start of Challenge: Challenge goes public. Large scale compute.

Join the fight against Covid-19

Who Can Participate?

Initially, access will be reserved to research groups that are able to contribute data (at least a total of 200 cases split into COVID19 and control cases → see data). Later, access will be extended. This will allow ramping up infrastructure.

Academic Radiology Departments

Contribute data + define challenges, full access

Radiology departments and private practices

Donate datasets of your cases to help!

Deep learning experts from any background

Use your skills on the entire dataset!

Join the fight against Covid-19

We Are Looking For High Quality Data

Collection and Curation

We accept anonymized DICOM datasets stripped of any patient information + clinical metadata (we provide a form) containing the parameters on the right:
One dataset can contain one or several chest CT scans of the same patient (indicate on what day since hospital admission it was taken).

Clinical metadata (example)

  • sex (m/f/d)
  • age (in years at the time of admission)
  • PCR confirmed positive / negative / not available
  • clinical condition (outpatient, inpatient, intermediate care, intensive care)
  • days since admission to hospital
  • (pre-existing relevant health conditions)

Curation

  • Standardize data format
  • Quality control
  • Standardize metadata

Data levels

Increasing capabilities

With the progress of the project, we aim to instruct and help our clinical partners to extend their datasets by additional clinical variables and metrics acquired throughout the treatment process.

Level I data

  • Sex, age
  • PCR result (positive, negative, n/a)
  • clinical condition (ER, inpatient, intermediate care, intensive care)
  • days since admission to hospital

Level II data

  • Follow-up scans (CT and Rx)
  • (pre-existing relevant health conditions)
  • Lab results, pO2

Level III data

  • Outcomes at discharge (healed, impaired, fatal)
  • Events during hospitalization (IMC, ICU, ventilator, interventions) with exact dates
  • Drug treatment

Join the fight against Covid-19

Patient Data Protection and Ethics

1

No storage of personal data

Our challenge does not require any personal patient data as only records that have been fully anonymized enter our data pool.
2

Removal of patient data before donating

We ask all participants from the medical community to remove any personal patient information before uploading the images to our local storage.
3

Data Quarantine

Whenever we receive medical image material, we conduct a check to strip the images of any personal information prior to the upload in our data repository.
4

Ethical approval

For this challenge, we obtained ethical approval by the Ethics committee of the medical faculty of LMU Munich who supervises M3i’s Digital Biobank.
5

Private data repositories

The images will be saved in private data repositories on Amazon S3 servers. Only the byproducts of training, validation, testing will be allowed access by the user.
6

Data sharing

The resulting models of the challenges will be published as open-source and be freely available.

Dissemination

Reproducible science

It is important that the challenge creates value for patients and caregivers. And provides opportunity for future research. That’s why the winning algorithms will be made open-source and available for public research use.

Models/Networks

  • The resulting models will be made open-source and publicly available on GitHub

Benchmark datasets

  • Science never stops. Our tested COVID-19 benchmark will be publicly available after challenge conclusion. Thus, we create and maintain a gold-standard performance test for future research.

Publication

  • We encourage and support the teams to publish their results

Data Collection Process

Join the fight against Covid-19

Team

We are a global, interdisciplinary team driven by one goal

Linda bij de Leij

My name is Linda bij de Leij and I work at the University Medical Center Groningen (UMCG), in the Netherlands. There, I'm a communication officer for the Data Science Center in Health (DASH) and the Research IT-programme. For the Covid19Challenge I will, among other things, keep you updated on the progress of the challenge through the website, social media and email updates. I am happy to join the team!

Fausto Milletarì, PhD

Thomas Heiliger

Thomas Heiliger works at Brainlab for several years and has deep inside in digital surgery and medicine from an industry point of view.
In addition, he is working at the Ludwig-Maximilians Universität München on his doctoral thesis on a scientific project related to research in the field of augmented reality in minimally invasive abdominal surgery. The preparation and autonomous enrichment of medical image data with 3D objects are one of his fields of research.
Regarding the challenge, he will be taking part in the labeling process of the data and push collaboration with external industry partners.

Dr. Szilard Szabo

Szilard Szabo is a doctor of medical informatics and a physician in training at Ludwig-Maximilians Universität München, as well as team leader digital biobank at M3i Industry-in-Clinic Platform. For the Covid19Challenge, he is responsible for the data teams and helping the clinics to transfer their data.

Annika Reinke

Felix Swamy v. Zastrow

Felix Swamy v. Zastrow is a fifth-year medical student at Ludwig-Maximilians Universität München and joined the M3i Industry-in-Clinic Platform a year ago as part of the segmentation team. He takes part in the specification of data and is responsible for the training of labeling specialists and quality control. For the Covid19Challenge, he is involved in the segmentation and labeling of the data as well.

Seyed-Ahmad Ahmadi, PhD

Seyed-Ahmad Ahmadi is an AI scientist from Ludwig-Maximilians Universität München. Within the Covid19Challenge, he advised on structuring the process and he is, together with the Technical University of Munich, in charge of building the reference model.

Dr. Stefan Taing

Stefan Taing is a partner at Munich Innovation Group GmbH and CEO & co-founder of M3i Industry-in-Clinic Platform. For the Covid19Challenge, he takes care of many administrative tasks and provides some of his employees to work for the challenge.

Prof. Nassir Navab

Nassir Navab is chairholder, computer aided medical procedure & augmented reality at Technical University of Munich. For the Covid19Challenge, he contributes to building the reference model in cooperation with Ludwig-Maximilians Universität München.

Gergely Dietz, MD

Gergely Dietz is a Medical Doctor and CEO at Neumann Medical, which we proudly present as one of our new partners. Neumann Medical is a health IT company, developing a disruptive new method for the generation and collection of medical data. By using structured reporting and standardized image annotation, the company supports the Covid19Challenge by collecting and digitizing Covid-data from medical institutes in the CEE region.

Dr. Andreas Liebl

Andreas Liebl is the managing director at UnternehmerTUM and head of the appliedAI Initiative - UnternehmerTUM. For the Covid19Challenge, he and the appliedAI team can provide support on the data collection and marketing of the challenge. We are happy to have them on board!

Dr. Simon Weidert (M.D.)

Simon Weidert works at the University Clinic of the Ludwig-Maximilians Universität München as an orthopedic trauma and spine specialist and he is also founder and co-CEO of the M3i Industry-in-Clinic Platform. He is part of the core team that had the idea to use an existing digital biobank solution in a way that creates a challenge that allows teams from all over the world to participate and crowd-source solutions for clinical problems. Simon firmly believes that using artificial intelligence can contribute to solving specific problems that arise in the current pandemic, which is his main motivation for the COVID-19 Challenge The method used in this challenge, could be applied to other future problems as well, because a crowd will always be smarter than an individual.

Dr. Jens Elsner

Jens Elsner is a partner engineering services and industry expert telecommunications at Munich Innovation Group GmbH and CEO of Munich Innovation Labs. For the Covid19Challenge, he does the project management and operations of the challenges. He leads his own AI team at Munich Innovation Labs, which helps us with IT tasks.

Wolfgang Männel

Wolfgang Männel is the managing director of Tathros Innovation and Tathros Digital, as well as senior partner at Munich Innovation Group GmbH. For the Covid19Challenge, he is an advisor and he helped to create and set up the website for the challenge.

Dr. Pál Maurovich Horvat

Pál Maurovich Horvat is the director of the Medical Imaging Centre and chairman of Radiology at the Semmelweis University, Budapest, Hungary. Dr. Maurovich Horvat is the elected vice president for nuclear cardiology & cardiac CT at the European Association of Cardiovascular Imaging (EACVI). He has graduated from Semmelweis University and from Harvard University. He is the author of more than 140 papers with over 5000 citations. Dr. Maurovich Horvat helps to coordinate the Covid19Challenge network in Central and Eastern Europe.

The Medical Imaging Centre of the Semmelweis University is a high volume institution with three departments: Radiology, Nuclear Medicine and Neuroradiology. The Medical Imaging Centre is the diagnostic imaging centre for COVID-19 patients at the Semmelweis University.

Istvan Köveshazi

Istvan Köveshazi is a fifth-year medical student and doctoral candidate at the Ludwig-Maximilians Universität München. His doctoral thesis is about pedicle screw planning in the spine surgery using artificial intelligence. He also works on several AI projects at M3i Industry-in-Clinic Platform which require the implementation of medically correct segmentation. For the Covid19Challenge, he is responsible for data sorting, anonymization and categorization, and he is the contact person for questions regarding our anonymization tool. He also played a part in defining the segmentation procedure, for example the selection of the most suitable medical imaging software for our project.

Scientific Advisory Committee

Our team is advised by

Dr. Franz MJ Pfister

Dr. méd. Amine Korchi

Prof. Dr. Lena Maier-Hein

Dr. Daniel Kondermann

Debdoot Sheet, PHD