Clinical Data Science
The clinical data science group employs advanced statistics, machine learning and computer vision techniques in the context of clinical radiology to enable fast and precise AI-supported diagnosis and prognostication. Our direct embedding into the Department of Radiology provides close cooperation and interdisciplinary interaction with radiologists. This gives us the opportunity to identify and address clinical needs and to develop and evaluate AI solutions directly in the clinical context.
Image data is the most natural type of data in every radiology department. Consequently, a great part of our research is dedicated to image analysis. In particular, we use state-of-the-art deep learning methods to analyze CT and MR images in 2D or 3D.
Computer vision in the field of medicine differs from other application areas since medical image data is typically sparse and labels are very expensive (time and money-wise). This makes computer vision in medicine a challenging, but also very fascinating branch of deep learning research.
Our ongoing computer vision research includes - but is not limited to - the following topics:
- Analyzing medical time series data by feeding CT perfusion images into complex deep neural network architectures
- Predicting the character of potentially cancerous lesions on sparse 3D image datasets without additional information, e.g. segmentation masks or laboratory findings
- Predicting clinical endpoints for 2D and 3D CT images using established convolutional neural network architectures
Our group aims at solving a wide range of problems in the context of clinical radiology. Each task comes with different data and needs to be tackled with different tools. Among the tools we used most often are: classical machine learning algorithms (e.g. random forest), radiomics and classical statistics.
Typically, we apply classical machine learning algorithms and statistics to tabular data in order to solve a clinical task, predict a clinical endpoint or visualize complex data structures.
Radiomics extracts standardized image features from medical images, e.g. CT or MR images, thereby transforming complex information stored in 2D or 3D images into tabular data. The latter can then be analyzed with classical machine learning algorithms and classical statistics.
Recent projects include:
- Radiomics project in which we predict the character of potentially cancerous lesions detected in CT scans
- Survival analysis to study the effects of a reduced chest x-ray volume on intensive care units
Clinical data science within the context of research on radiological topics entails the use of advanced, specialized tools on complex data and consideration of department specifics within the application of these tools.
In order to accommodate the needs of the computer-vision and machine-learning research, using most recent and novel methods, and to answer clinical questions, based on the rich data from heterogeneous sources, we develop, deploy and maintain the tools and infrastructure needed for these tasks.
Our recent efforts include the projects of the like of:
- Structured acquisition of patient data, used for an exploratory analysis of Covid-19 patients undergoing radiological imaging
- Retrieval and pseudonymization of radiological images from our picture archive for a large dataset for supervised deep learning
- Monitoring, visualization and evaluation of the imaging activity of the radiology department
Where to Find Us
Medical Advisor Radiology
Dr. rer. nat. Balthasar Schachtner
Balthasar received his PhD for his research in the field of experimental particle physics. The focus of his postdoctoral research is the facilitation of machine learning in radiology and the development of imaging biomarkers for lung pathologies.
Andreas Mittermeier (M.Sc.)
Andreas has a master’s degree in physics and was interested in medical imaging since his undergraduate studies. His doctoral research project comprises perfusion imaging analysis in the context of novel machine learning methods.
Theresa Stüber (M.Sc.)
Theresa got her master's degree in (bio-)statistics and pursued her great interest for machine learning in medicine already during university studies. In her doctoral research she develops a framework for the combinaton of deep learning with classical statistical modeling.
Philipp Wesp (M.Sc.)
Philipp has a medical physics background and focused primarily on proton computed tomography as a Master student. Today he analyzes CT images using modern machine learning techniques in order to predict valuable clinical endpoints, e.g. the character of potentially malignant lesions.
Patrick Schinke (B.Sc.)
Master Student from LMU’s Faculty of Informatics working on forensic age estimation. Patrick is passionate about his project because it involves solving real world problems and touches on a number of relevant topics outside of computer science, specifically in the field of forensic medicine and law.
Tobias Weber (B.Sc.)
Tobias is working on adapting deep-learning architectures for survival estimation as well as for 3-dimensional CT images for his master thesis in Computer Science. He focuses on mapping complex CT data to latent-space representations and the extraction of features using dedicated survival objectives.
Rueckel J, Kunz WG, Hoppe BF, Patzig M, Notohamiprodjo M, Meinel FG, Cyran CC, Ingrisch M, Ricke J, Sabel BO. Artificial Intelligence Algorithm Detecting Lung Infection in Supine Chest Radiographs of Critically Ill Patients With a Diagnostic Accuracy Similar to Board-Certified Radiologists. Crit Care Med. 2020 Jul;48(7):e574-e583. doi: 10.1097/CCM.0000000000004397. PMID: 32433121.
Rueckel J, Trappmann L, Schachtner B, Wesp P, Hoppe BF, Fink N, Ricke J, Dinkel J, Ingrisch M, Sabel BO. Impact of Confounding Thoracic Tubes and Pleural Dehiscence Extent on Artificial Intelligence Pneumothorax Detection in Chest Radiographs. Invest Radiol. 2020 Jul 15. doi: 10.1097/RLI.0000000000000707. Epub ahead of print. PMID: 32694453.
Nadjiri J, Schachtner B, Bücker A, Heuser L, Morhard D, Landwehr P, Mahnken A, Hoffmann RT, Berlis A, Katoh M, Reimer P, Ingrisch M, Paprottka PM. Availability of Transcatheter Vessel Occlusion Performed by Interventional Radiologists to Treat Bleeding in Germany in the Years 2016 and 2017 - An Analysis of the DeGIR Registry Data. Rofo. 2020 Oct;192(10):952-960. English, German. doi: 10.1055/a-1150-8087. Epub 2020 Jul 7. PMID: 32634837.
Mittermeier A, Ertl-Wagner B, Ricke J, Dietrich O, Ingrisch M. Bayesian pharmacokinetic modeling of dynamic contrast-enhanced magnetic resonance imaging: validation and application. Phys Med Biol. 2019 Sep 17;64(18):18NT02. doi: 10.1088/1361-6560/ab3a5a
Fasler DA, Ingrisch M, Nanz D, Weckbach S, Kyburz D, Fischer DR, Guggenberger R, Andreisek G. Rheumatoid cervical pannus: feasibility of volume and perfusion quantification using dynamic contrast enhanced time resolved MRI. Acta Radiol. 2020 Feb;61(2):227-235. doi: 10.1177/0284185119854200. Epub 2019 Jun 6. PMID: 31169411.
Smith EE, Biessels GJ, De Guio F, de Leeuw FE, Duchesne S, Düring M, Frayne R, Ikram MA, Jouvent E, MacIntosh BJ, Thrippleton MJ, Vernooij MW, Adams H, Backes WH, Ballerini L, Black SE, Chen C, Corriveau R, DeCarli C, Greenberg SM, Gurol ME, Ingrisch M, Job D, Lam BYK, Launer LJ, Linn J, McCreary CR, Mok VCT, Pantoni L, Pike GB, Ramirez J, Reijmer YD, Romero JR, Ropele S, Rost NS, Sachdev PS, Scott CJM, Seshadri S, Sharma M, Sourbron S, Steketee RME, Swartz RH, van Oostenbrugge R, van Osch M, van Rooden S, Viswanathan A, Werring D, Dichgans M, Wardlaw JM. Harmonizing brain magnetic resonance imaging methods for vascular contributions to neurodegeneration. Alzheimers Dement (Amst). 2019 Feb 26;11:191-204. doi: 10.1016/j.dadm.2019.01.002. PMID: 30859119; PMCID: PMC6396326.
Debus C, Floca R, Ingrisch M, Kompan I, Maier-Hein K, Abdollahi A, Nolden M. MITK-ModelFit: A generic open-source framework for model fits and their exploration in medical imaging - design, implementation and application on the example of DCE-MRI. BMC Bioinformatics. 2019 Jan 16;20(1):31. doi: 10.1186/s12859-018-2588-1. PMID: 30651067; PMCID: PMC6335810.
Suchorska B, Schüller U, Biczok A, Lenski M, Albert NL, Giese A, Kreth FW, Ertl-Wagner B, Tonn JC, Ingrisch M. Contrast enhancement is a prognostic factor in IDH1/2 mutant, but not in wild-type WHO grade II/III glioma as confirmed by machine learning. Eur J Cancer. 2019
The Clinical Data Science group gratefully acknowledges research funding by:
Research training group GRK 2274 of the DFG, Deutsche Forschungsgemeinschaft
Deutsches Zentrum für Lungenforschung (DZL)
We can regularly offer Bachelor and Master Theses in the field of Data Science and Machine Learning applied in the context of clinical radiology. If you're interested and wish to learn more about our work feel free to get in touch with us.