The Intelligent Imaging Laboratory at the Section for Biomedical Imaging
Establishment and future
Research within the field of artificial intelligence is a focus of the Section for Biomedical Imaging (SBMI). The work is carried out by the team at SBMI’s Intelligent Imaging Lab (i²LAB), mostly in cooperation with national and international partners.
As the imaging service provider for MOIN CC, our team – which focuses on commercial collaborations – offers key competencies around innovative diagnostic and therapeutic procedures for the detection and treatment of oncological, inflammatory, musculoskeletal and neurological diseases.
Our research on AI began in 2018, and i²LAB is now a recognised player in the field of AI in medicine in northern Germany. AI has the potential to enable efficient translation of basic research into valuable clinical applications, supporting the SBMI’s core goals and areas of work (see Background, Mission & Objectives).
Financing and major projects
The collaborative research projects coordinated by i²LAB are funded with >€3,700,000 of public funding from the BMBF and BMWi (of which >€1,250,000 is for i²LAB), as well as with additional support from Kiel University and its Faculty of Medicine.
The most important ongoing projects include
- ARTEMIS – AI applications for osteoporosis and osteoarthritis
- KI-RAD (part of KI-SIGS) – AI applications in the areas of stroke and skeletal trauma
- RACOON – AI applications in COVID-19
- SOFIA – AI applications for the prediction of bone fractures
The i²LAB team is conducting research on:
- AI methods, development of new AI models and imaging techniques
- AI hardware, building a powerful computing framework centered on the K9 workstation
- AI applications to improve the diagnosis and prognosis of serious diseases, such as stroke, osteoporosis and osteoarthritis, as well as COVID-19
- About AI: perceptions of AI among doctors and patients and methods to improve its trustworthiness. We want to help improve acceptance based on transparent methodology, development and validation.
Computer-assisted quantitative analysis of biomedical images reveals information that is not visible to the human eye. Calibration converts the greyscales of the image into interpretable information about the pathophysiological state of cells, tissues and organs – that is, the state of health of the individual.
These procedures can be significantly improved and facilitated through the use of artificial intelligence methodologies. In the Section for Biomedical Imaging’s Intelligent Imaging Lab, the i²LAB, we are working to make this a reality.
Mission and objectives
The mission of i²LAB is to resolve important medical needs by identifying the most appropriate imaging technology processed using powerful artificial intelligence methods.
Our aim is to develop methodologies
- for automated – that is, objective – operator-independent image analysis
- with the greatest accuracy and precision
- as non-invasively as possible and with minimal health risk
This includes theoretical conception, experimental implementation, preclinical testing, translation into clinical application and epidemiological evaluation.
i²LAB team members usually have a background in physics and computer science, but to be successful, we need close collaboration with SBMI’s multidisciplinary team, which includes doctors and veterinarians, engineers, molecular biologists and biochemists, as well as other disciplines.
Team spirit starts with understanding the mindset of researchers from different backgrounds in order to exploit synergies at the interface between science and medicine.
- Timo Damm, physicist, responsible for IT at MOIN CC and senior scientist for micro-computed tomography, with key projects in the field of musculoskeletal disorders
- Claus-C. Glüer, physicist, head of SBMI, leads research at i²LAB
- Jan-Bernd Hövener, physicist, head of SBMI
- Johannes Köpnick, biomedical engineer, with key projects in gastrointestinal diseases and vessel-wall imaging
- Nicolai Krekiehn, physicist, with key projects in the areas of lung diseases and musculoskeletal disorders
- Eren Yilmaz, computer scientist, specialising in stroke and musculoskeletal disorders
- Niklas Koser, master’s student in computer science, AI topics in the field of musculoskeletal disorders
- Christopher Hansen, PhD student, dental AI
National Consortium for AI in Osteoporosis and Osteoarthritis
ARTEMIS is a multicentre consortium coordinated by i²LAB to develop automated methods for opportunistic screening of CT scans with the aim of identifying patients at high risk of osteoporotic fractures. ARTEMIS-DiaProoF (“Diagnosis and prognosis of osteoporotic fracture risk”) is the sub-project pursued by i²LAB to develop optimised AI-based image analysis techniques for the analysis of quantitative computed tomography of the spine and femur.
ARTEMIS partners include Universitätsklinikum Erlangen, the Universität zu Lübeck, the Ostfalia University of Applied Sciences and Christian-Albrechts-Universität zu Kiel.
ARTEMIS DiaProoF.(‘Diagnosis and prognosis of osteoporotic fracture risk’) is the ARTEMIS subproject conducted by i²LAB. We focus on assessing risks for osteoporotic fractures with a more comprehensive evaluation of CT-based volumetric bone mineral density of the spine and proximal femur. By matching structural features with AI methodologies, we will attempt to predict fragility features that are only visible on high-resolution CT images. The methods will be developed and tested in two independent cohorts, the AGES-Reykjavik study and UKSH data, both of which follow a prospective AI-based study design.
SOFIA is an AI project based on data from the Study of Osteoporotic Fracture (SOF), the most widely published study in the field of osteoporosis with hundreds of publications. In SOFIA, we will pursue two goals:
Evaluate the predictive power of a deep neural network for assessing X-rays of the pelvis and hand regarding the occurrence of hip and other osteoporotic fractures; and artificial intelligence-based identification of features on pelvic X-rays indicating an increased risk of hip fracture.
Insights into the AI black box
A major problem associated with AI is its black box nature: it is not clear how the network arrives at its conclusions. This can have an effect on acceptance. So we also want to improve the interpretability of AI-based fracture risk predictions. We plan to gain insight into the AI black box by using visualisation methods such as heat maps and deep generative methods to better understand the pattern that the AI has learned.
SOFIA is a collaboration between i²LAB and the Institute of Computer Science at Kiel University and US researchers at UCSF, Harvard Medical School and other US universities and research centres.
RACOON is based on a platform architecture of patient data in combination with research structures to analyse these data. All of the radiology departments of German universities are participating by contributing COVID-19 radiological patient data. This will include an extensive database of computed tomography and thoracic radiography data, which will be linked to relevant clinical patient information on the disease.
Structured reporting will enable the pooling of decentralised data collection while maintaining high standards of standardisation and quality assurance. This data will form the basis for high-quality research studies. An early warning dashboard system will help to combat future pandemics.
i²LAB research in RACOON
At i²LAB, we plan to develop AI models to assess disease severity. In particular, we’re going to develop an image-based quantitative COVID-19 score (iqCS) that could be used to predict patient outcomes, especially to identify those at greatest risk for ICU treatment and high long-term morbidity or mortality. We also aim to use AI models to predict CT-based risk scores from chest X-rays.
The severity of disease in COVID-19 patients varies widely, and identifying risk factors for a severe course is important in order to take preventive measures. We will use vertebral fractures and muscle status to assess the contribution of musculoskeletal frailty.
KI-RAD is part of KI-SIGS (“AI Space for Intelligent Healthcare Systems”), northern Germany’s large AI research consortium with partner universities in Bremen, Hamburg and Lübeck (including the KI-SIGS coordination office).
In KI-RAD, we want to develop software that can serve as an intelligent assistant for radiologists and clinicians, specifically for distinguishing haemorrhages from infarctions as the cause of stroke in patients in the emergency department, identifying fresh fractures in trauma patients for rapid establishment of appropriate medical treatment.
The partners in KI-RAD are the University Hospital Schleswig-Holstein via the Department of Radiology and Neuroradiology in Kiel (Prof. Olav Jansen, Dr Sam Sedaghat and Johanna Rümenapp) and the Department of Radiology and Nuclear Medicine in Lübeck (Prof. Jörg Barkhausen), the Institute of Medical Informatics at the University of Lübeck (Prof. Mattias Heinrich), the Institute for Biomedical Imaging at the University Medical Center Hamburg-Eppendorf (Prof. Tobias Knopp), Prof. Carsten Meyer, Ostfalia University of Applied Sciences, Wolfenbüttel, and two companies: Philips Research, Hamburg and mbits, Heidelberg
Research at i²LAB is carried out together with national and international partners.
In Germany, our most important partners are the universities/medical centers in Erlangen, Hamburg, Lübeck, Kiel and Wolfenbüttel.
On an international level, our most important partner is the University of California, San Francisco (UCSF), in particular:
- ci² - Center for Intelligent Imaging in the Department of Radiology and Biomedical Imaging
- San Francisco Coordinating Center
Our Kiel University-funded Intelligent Imaging International project offers a quarterly AI webinar with ci² at UCSF:
- Transatlantic UCSF/Kiel University Webinar on Artificial Intelligence in Biomedical Imaging
International collaborations on large study databases also include
- The MrOS study, Portland, Oregon, USA
- AGES-Reykjavik study