The Diagnostic Image Analysis Group was founded in January 2010. This page provides a brief history of the group, our main research topics, and collaborations.
When mammography screening for breast cancer started in The Netherlands in 1989, the national reference center was established in Nijmegen. Nico Karssemeijer, then a post-doc at Radboudumc, realized that computer analysis of mammograms would become a key area within medical image analysis, especially when digital screening would become a reality. He published his first paper on the automatic detection of abnormalities in mammograms in 1992.
Karssemeijer soon established one of the leading groups in the emerging field of computer-aided detection and diagnosis (CAD). He set up a long collaboration with the Bay Area tech company R2 Technology, which developed the first FDA approved CAD product for detecting breast cancer in mammograms. The core technology for this product is based on Karssemeijer's method of stellate distortion detection.
Mammography CAD remains an active research area within DIAG. Research has expanded into breast density estimation from mammograms, initially together with Matakina Ltd (Wellington, New Zealand), a startup co-founded by Karssemeijer and now named Volpara, and the design of computer-aided detection and diagnosis systems for other modalities such as tomosynthesis, 3D ultrasound and MRI.
Prostate image analysis
Henkjan Huisman, who obtained his PhD in 1998 on ultrasound analysis joined Nico Karssemeijer to expand the field of research to breast and prostate MRI. His first publication with Jelle Barentsz on the topic appeared in 1999. Radboudumc is a worldwide leading center for prostate cancer imaging, and DIAG therefore pioneered the development of prostate CAD and decision support systems.
In 2013, DIAG set up a quality controlled environment (MDD Annex II and ISO 13485) for developing medical device software and CE marked Huisman's ProCAD software. ProCAD was used on a daily basis in clinical practice in several hospitals in Europe and Australia. Huisman now also focuses on real-time AI support during interventions for prostate cancer diagnosis and treatment.
Chest radiography and CT
In 2010, Bram van Ginneken came from University Medical Center Utrecht and joined Nico Karssemeijer at Radboudumc. They adopted the name Diagnostic Image Analysis Group, DIAG for short. In 2011 and 2012, Nico Karssemeijer and Bram van Ginneken were appointed full professor at Radboud University. DIAG was initially a group of about 10 researchers, but the group grew quickly and the overall size increased by around 10 people every year.
Van Ginneken had performed his PhD research on the automatic detection of tuberculosis in chest radiographs at the Image Sciences Institute in Utrecht, The Netherlands from 1997 to 2001. This work continues within DIAG to date, co-funded by the Dutch company Delft Imaging under the name CAD4TB. In 2014, CAD4TB was CE marked by DIAG. The software that analyzes chest radiographs for signs of tuberculosis is now in active use in over 35 countries, mainly in African and Asia, and processes around 5000 chest radiographs per day.
In Utrecht, Van Ginneken started to work on the analysis of chest CT scans in 2003, and together with Mathias Prokop, now head of the Department of Imaging at Radboudumc, began to develop a wide range of techniques and algorithms for detecting and analyzing lung cancer, COPD, and other lung diseases. Eva van Rikxoort was one of the PhD students who graduated in the Utrecht chest CT group of Van Ginneken and Prokop, and after a postdoc at UCLA she joined DIAG in 2011. In the same year, the chest radiologist Cornelia Schaefer-Prokop joined DIAG making chest CT a research focus.
Colin Jacobs did his PhD on early detection of lung cancer in CT and after his graduation he continued to lead the chest CT lung cancer research. Jacobs was lead developer of CIRRUS Lung Screening, software for reading low dose lung cancer screening CT scans with integrated tools for nodule detection, matching, segmentation and characterization. This was commercialized in collaboration with MeVis Medical Solutions as Veolity, and under different brandings this software is market leader for lung cancer screening reading solutions.
Retinal image analysis
Another activity that Bram van Ginneken started in Utrecht and continued in Nijmegen was analysis of retinal images. Clarisa Sánchez joined Bram van Ginneken in Utrecht as a postdoc and went with him to Nijmegen where she is now a faculty member, leading the research on image analysis of ophthalmological images. DIAG is closely collaborating with Radboudumc's Department of Ophthalmology in this area and a first joint paper was published in 2013.
Radiologists used to look at images on a lightbox. Today they obviously use computer screens; only when you search on google for pictures of a radiologist, you'll still see films and lightboxes. Pathologists, on the other hand, still use their microscopes to look at glass slides. But google for pictures of a pathologist and you start to see computer screens with colorful pathology images next to the microscopes. The reason that pathology is two decades behind radiology in terms of digitization is that scanned tissue slides at the resolution needed by a pathologist are very large. But nowadays, progress in computer technology makes fully digital pathology departments economically feasible. DIAG realized that this transition has the same potential for computer-aided diagnosis as in radiology and started in 2013 with research projects on computational pathology, led by Jeroen van der Laak and later joined by Geert Litjens and Francesco Ciompi.
The Computational Pathology Group has rapidly grown. Key achievements include the organization of the CAMELYON16 challenge with the results published in JAMA and a state-of-the-art implementation for detection of lymph node metastases made available on grand-challenge, the development of an automated method for Gleason grading, published in Lancet Oncology, and a method for neural image compression that can process complete high-resolution whole slide images on a single GPU.
ScreenPoint and Thirona
In 2014, two spin-off companies were established by DIAG members. Nico Karssemeijer founded ScreenPoint, focusing on solutions for computerized interpretation of mammograms and digital breast tomosynthesis. Eva van Rikxoort and Bram van Ginneken started Thirona, a company that provides tools and services for medical image analysis and imaging quantification. Many alumni of DIAG work in these new companies.
ScreenPoint has released Transpara, a product for detecting breast cancers on mammograms and tomosynthesis breast exams. Thirona has developed LungQ, CAD4TB and RetCAD, three products for analysing chest CT scans, chest radiographs and fundus photographs. Thirona also collaborates with DIAG and Delft Imaging in Thira Lab, an ICAI Lab. These labs have a focus on AI technology and are started throughout the Netherlands to stimulate knowledge transfer from public to private partners and, in our case, make the solutions developed at DIAG available to patients worldwide.
Strategic collaboration with Fraunhofer MEVIS
Bremen-based Fraunhofer MEVIS and DIAG have been successfully collaborating as partners in research projects in the area of breast imaging since the mid-nineties, long before MEVIS became part of the Fraunhofer-Gesellschaft. In 2010, this collaboration was deepened and DIAG and Fraunhofer MEVIS established a strategic partnership. There is a continuous exchange of softare, data, and personnel between the two groups. Van Ginneken holds a part-time position at Fraunhofer MEVIS and several Fraunhofer MEVIS researchers have performed or are performing their PhD research at DIAG. MeVisLab has been adopted at DIAG as the software environment used to build all its medical image analysis applications.
DIAG has also expanded its activities to neuro image analysis with MRI and 4D computed tomography led by rashindra-manniesing, and has more recently started new research lines in musculoskelatal image analysis, led by Nikolas Lessmann and Matthieu Rutten, point-of-care ultrasound with real-time integrated deep learning support led by Thomas van den Heuvel and radiotherapy image analysis.
In 2021, the group has grown to 7 faculty members, 6 post-docs, 4 radiologists, 12 software engineers, research scientists and support personnel, and 33 PhD students. DIAG is a diverse and international group; our 68 researchers originate from 15 countries.
In 2007, Bram van Ginneken co-organized the first "grand- challenge" in medical image analysis. The idea of a challange is that a task is defined, for example the segmentation of the liver in a CT scan, a test data set is set aside and team send in the output of their best algorithm and all these results are validated and compared. The concept has been very successful. In 2010, DIAG started to implement a web platform for organizing such challenges, grand-challenge.org.
It became clear that most research projects of DIAG can be seen as a "challenge", as the work involves the collection of training data (input), the development of software trained with that data to produce a certain output, and writing a scientific paper. Gradually, DIAG started to use grand-challenge.org for more and more internal project, while at the same time the number of external users grew quickly. As of mid 2020, over 45,000 researchers have registered on the platform, it hosts 80 public challenges and 180 internal projects and every month about 2500 unique users log in to the site. The software engineering team of DIAG, led by James Meakin has integrated data archives, annotation tools using CIRRUS built with the web toolkit of MeVisLab, and support for online algorithms.
grand-challenge.org is now the official platform for the Radboud Technology Center Deep learning, led by Ajay Patel. The RTC Deep Learning supports image analysis research within DIAG but also wider within Radboudumc and externally. It develops AI solutions and maintains Sol, one of the largest GPU research clusters in The Netherlands.
AI for Health
In June 2019, Radboudumc decided to invest broadly in applied research aiming to integrate artificial intelligence solution in clinical practice. This initiative, AI for Health aligns perfectly with the vision of DIAG, but has a wider scope and also addresses AI application that do not use medical images as input, but process clinical data, genetic data, and text data. AI for Health is also an ICAI lab. The soaring interest in healthcare applications of AI is fueled by the success of deep learning, the technology used in all DIAG projects these days. In the next few years, we may see a close integration of DIAG and AI for Health.
With its focus on creating clinical applications of its algorithmic solution for medical image analysis, DIAG is closely collaborating with a wide range of companies. The following companies directly fund research carried out at DIAG, or have done so in the recent past:
- Mevis Medical Solutions (formerly MeVis Technology AG) was founded in 1997 as an independent spin-off of MeVis Research (today Fraunhofer MEVIS) to act as a commercial partner and to facilitate the transfer of research results into marketable products. DIAG works with MMS in the area of lung CT screening.
- Thirona was founded in 2014 by DIAG members Eva van Rikxoort and Bram van Ginneken. The company funds DIAG research projects in chest CT analysis, fundus imaging, chest x-ray analysis and oncology. Thirona participates in Thira Lab.
- Delft Imaging develops diagnostic imaging equipment and e-health software. DIAG and Delft Imaging have developed CAD4TB, CE marked software for the automated interpretation of chest radiographs, and they work together in the area of maternal health, developing a portable ultrasound device with integrated deep learning called the BabyChker. Delft Imaging is also part of Thira Lab.
- Siemens Healthineers has a close collaboration with the Imaging Department of Radboudumc and works with DIAG in the area of prostate MRI analysis and interventions.
- ScreenPoint collaborates with DIAG in various project in breast image analysis.
- Elekta is partnering with the Department of Radiotherapy in Radboudumc. They work with DIAG on radiotherapy aplication in MRI and cone-beam CT integrated in their linear accelarators.
- Philips is involved in multiple research projects with DIAG in the area of lung cancer interventions and funded research in the Computational Pathology Group on prostate and breast cancer.
- Sectra works with the Computational Pathology Group in DIAG on integrating algorithms into the clinical workflow.
- Smart Reporting runs several projects with DIAG researchers, as part of EU-funded research. One example is the collaboration in CT and chest x-ray image analysis of COVID-19 suspects and patients.
- Microsoft Research and Amazon Web Services have provided and are providing cash and in-kind support for our work on grand-challenge.org.
- Canon is working with DIAG on ultrasound image analysis, is selling DIAG's CAD4TB software solution and was involved in 4D CT image analysis research.
- Riverain Technologies from Dayton, Ohio, develops software to analyze chest radiographs and chest CT scans. DIAG has developed algorithms for Riverain and has carried out observer studies with Riverain's software to investigate the effect of using bone suppression, temporal analysis and nodule detection on clinical practice.
- Matakina is a company co-founded by Nico Karssemeijer that develops software for mammographic image analysis. Matakina’s flagship product is Volpara, an automatic software system that gives a breast density score based on an estimate of the volume of dense tissue in the breast. Matakina has been funding DIAG research towards improvements of breast density estimation methods.
- MedQIA is an imaging contract research organization based in Los Angeles that specializes in quantitative image analysis for clinical trials. DIAG worked with MedQIA and still collaborates with them through Thirona.
- QView Medical, Los Altos, California was co-founded by Nico Karssemeijer and Henkjan Huisman and is working on 3D breast ultrasound image analysis. Together with DIAG new computer-aided detection algorithms are developed.