The TUM Distinguished Lecture Series on AI & Healthcare is a bi-weekly lecture series taking place on Mondays from 4-5pm (CET) on Zoom. International experts and pioneers in the field will present their research.
Zoom: https://tum-conf.zoom.us/j/63669475200 (Passcode: 630676)
Subscribe to the mailing list for announcements: https://lists.lrz.de/mailman/listinfo/tum-dls-aih
28.11.2022, 4pm (CET)
Professor Stephen Smith (University of Oxford)
Brain imaging in UK Biobank
Abstract: UK Biobank is a massive prospective epidemiological study, with one major focus being the understanding of later-life disease. There is lifestyle, genetic and cognitive data, and much more, from 500k participants; brain and body imaging is taking place for 100k of these. Health outcomes data is being continuously fed in from NHS records. I will give a brief overview of UK Biobank, and then describe the brain imaging component - the imaging protocol, the processing pipeline that we have created, and some example results from the first 45,000 scans that we have processed and disseminated to the scientific community. I will also show results of looking for genetic associations with the brain imaging measures, and from studying brain aging. Finally, I will describe a sub-project that re-scanned 2000 participants who had already been scanned prior to the COVID-19 epidemic, and who have had the disease in between the two scans.
Speaker Bio: I head the Analysis Group at the Wellcome Centre for Integrative Neuroimaging. We carry out research into new methodologies for the analysis of functional and structural brain imaging data. Our research is also turned into software tools that are available in the FSL package, which is free to all academic and non-profit institutions.
12.12.2022, 4pm (CET)
Professor Tal Arbel (McGill University)
Towards Causal, Explainable, Generalizable Deep Learning Models for Image-Based Personalized Medicine
Abstract: Deep learning models for personalized medicine based on medical images, including predicting future disease progression and responders to different treatments across heterogenous populations, would have an enormous impact on healthcare and drug development. However, reaching this goal requires addressing the unique and open challenges presented to existing deep learning models by medical images acquired from patients in real clinical contexts, and during clinical trial analysis. These include developing explainable, robust and generalizable deep learning models, required for safe and trustworthy clinical deployment.
Speaker Bio: Tal Arbel is a Professor in the Department of Electrical and Computer Engineering, where she is the Director of the Probabilistic Vision Group and Medical Imaging Lab in the Centre for Intelligent Machines, McGill University. She is a Canada CIFAR AI Chair and Associate member of MILA (Montreal Institute for Learning Algorithms) and of the Goodman Cancer Research Centre. Prof. Arbel’s research focuses on development of probabilistic, deep learning methods in computer vision and medical image analysis, for a wide range of real-world applications involving neurological diseases. For example, the machine learning algorithms developed by her team for the detection and segmentation of lesions in brain MRI of patients with Multiple Sclerosis (MS) have been used in the clinical trial analysis of almost all the new MS drugs currently used worldwide. She is a recipient of the 2019 McGill Engineering Christophe Pierre Research Award. She regularly serves on the organizing team of major international conferences in computer vision and in medical image analysis (e.g. MICCAI, MIDL, ICCV, CVPR). She was an Associate Editor for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), and Computer Vision and Image Understanding (CVIU). She is currently the Editor-in-Chief and co-founder of the arXiv overlay journal: Machine Learning for Biomedical Imaging (MELBA).
17.01.2022, 4pm (CET)
Professor Alison Noble (University of Oxford)
Advances in Smart Ultrasound
Abstract: Medical ultrasound is an established tool in clinical radiology, traditionally advanced by progress in ultrasound physics, and known for its portability and utility. However, specialist training is needed to use the technology well. Artificial intelligence, and in particular deep learning is known for its power to learn to automate tasks by learning patterns from large data sets. But how has deep learning impacted medical ultrasound in the last decade? How smart are ultrasound systems today? And how do we make them smarter? I will look at these questions based on my experiences of working in obstetric ultrasound for the last 15 years and illustrated with examples of work from my laboratory, and highlight some of the research and translation challenges ahead if Smart Ultrasound is to fulfil its clinical translation potential.
Speaker Bio: Professor Alison Noble OBE FRS FREng is the Technikos Professor of Biomedical Engineering at the Institute of Biomedical Engineering (IBME), University of Oxford. Professor Noble’s academic research interests are in ultrasound imaging, and computational (machine-learning based) analysis of images and motivated by clinical unmet needs in western and low-and-middle-income countries healthcare settings. She received the Royal Society Gabor Medal for her inter-disciplinary research contributions in 2019, and the same year received the MICCAI Society Enduring Impact award. Professor Noble co-founded Intelligent Ultrasound Ltd to commercial research from her laboratory which was acquired by MedaPhor Group Plc in 2017 (now called Intelligent Ultrasound Group). Professor Noble is a former president of the MICCAI Society, and her recent UK national roles include Chair of the EPSRC Healthcare Technologies Strategic Advisory Team, and a member of the UK REF 2021 Subpanel 12 (Engineering). She is an active Fellow of the Royal Academy of Engineering and of the Royal Society, an ELLIS Fellow, a Fellow of the MICCAI Society, and a former Trustee of the Institute of Engineering Technology (IET). Professor Noble received an OBE for services to science and engineering in the Queen's Birthday Honours list in June 2013.
|31.01.2022, 4pm (CET)|
Professor Ron Kikinis (Harvard Medical School)
AI in Radiology
Abstract: AI has created a lot of excitement in the field of Radiology a few years ago. Today, we have a more realistic view with some working examples and a better understanding of the strength and limitations of the technology. While this situation is disappointing, it opens opportunities for new research and the hope for improvements for the next generation of algorithms. Significant infrastructure is required in order to train and deploy successful AI algorithms. The presentation will discuss the NCI Imaging Data Commons as an example of such an infrastructure.
Speaker Bio: Dr. Ron Kikinis received his MD degree from the University of Zurich, Switzerland, in 1982. He trained as a resident in radiology at the University Hospital in Zurich, and as a researcher in computer vision at the ETH in Zurich, Switzerland. In 1988, he moved to Brigham & Women's Hospital, and in 1990, founded the Surgical Planning Laboratory, Department of Radiology, Brigham and Women's Hospital, Harvard Medical School, Boston, MA. In 2004, he was appointed Professor of Radiology at Harvard Medical School, and in 2010, was appointed the Robert Greenes Distinguished Director of Biomedical Informatics, Department of Radiology, Brigham and Women's Hospital. From January 2014 through February 2020, he took on a part-time position in Germany as "Institutsleiter" of Fraunhofer MEVIS and Professor of Medical Image Computing at the University of Bremen, while continuing his activities in the U.S. On March 1, 2020, he returned to Boston full time and was appointed the B. Leonard Holman Professor of Radiology, Harvard Medical School, and Vice-Chair for Biomedical Informatics Research, Department of Radiology, Brigham and Women’s Hospital.
Professor Polina Golland (Massachusetts Institute of Technology)
Learning to read xray: applications to heart failure monitoring
Abstract: We propose and demonstrate a novel approach to training image classification models based on large collections of images with limited labels. We take advantage of availability of radiology reports to construct joint multimodal embedding that serves as a basis for classification. We demonstrate the advantages of this approach in application to assessment of pulmonary edema severity in congestive heart failure that motivated the development of the method.
Speaker Bio: Polina Golland is a Henry Ellis Warren (1894) professor of Electrical Engineering and Computer Science at MIT and a principal investigator in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). Her primary research interest is in developing novel techniques for medical image analysis and understanding. With her students, Polina has demonstrated novel approaches to image segmentation, shape analysis, functional image analysis and population studies. She has served as an associate editor of the IEEE Transactions on Medical Imaging and of the IEEE Transactions on Pattern Analysis. Polina is currently on the editorial board of the Journal of Medical Image Analysis. She is a Fellow of the International Society for Medical Image Computing and Computer Assisted Interventions (MICCAI) and of the American Institute for Medical and Biological Engineering (AIMBE).
Professor David Hawkes (University College London)
From Rock-and-Roll to Sailing in the Rain via 46 Years of Fun in Medical Imaging:
Abstract: I will describe how my initial foray into radiobiology with a physics degree got me involved in the early days of CT scanning, and then my determination to move beyond imaging for diagnosis to directly guiding interventions. I will share my experiences of the formation of CISG, CMIC and WEISS, illustrated with the story of image registration and developing image guided interventions in the brain, prostate and liver. I will finish with the exhilaration, promises and pitfalls of the revolution in deep learning and an aside on what dealing with the pandemic teaches us about rapid translation to clinic.
Speaker Bio: Professor David Hawkes graduated from Oxford with a BA in Natural Sciences (Physics) in 1974 and obtained an MSc in Radiolobiology in Birmingham in 1975. His first post was as a clinical scientist in nuclear medicine at Southampton General Hospital. After obtaining his PhD in 1982 at the University of Surrey and the Institute of Cancer Research he spent seven years working as a clinical scientist at St. George’s Hospital, London, before returning in 1988 to academia at the United Medical and Dental Schools of Guy's and St Thomas' Hospitals (UMDS), where he received a personal Chair in 1998. He was elected FMedSci in 2011, FBIR in 2007, FREng in 2002, FInstP in 1997 and FIPEM in 1993. He became an NIHR Senior Investigator in 2009.
Professor Caroline Essert (University of Strasbourg, CNRS)
Modelling the surgical knowledge for computer-assisted preoperative planning
Abstract: Minimally invasive surgery, and more particularly percutaneous surgical interventions, are very popular alternatives to open surgery. However, the main difficulty with such interventions is the lack of visibility on the surgical site. A thorough preoperative planning is essential for the best efficacy and safety of the procedure, to select the most optimal tools placement while preserving the surrounding sensitive organs. However, an optimal plan is usually very complex to imagine and requires a high expertise. In this talk, we will show how a computer model of the surgical knowledge coupled with patient-specific images and geometric modelling, are used to assist the surgeon to choose the best intervention strategy before performing the surgery.
Speaker Bio: Caroline Essert received her PhD of Computer Science (Computer Graphics) in 2001 and her Habilitation in 2011 from the University of Strasbourg / ICUBE Institute, where she is a professor since 2003. Her main research interests are computer graphics and geometry applied to preoperative surgery planning (more particularly for abdominal surgery and neurosurgery), computer-assisted surgery and navigation, virtual reality and haptics for surgery. She obtained a number of national and international grants on these topics, and authored or co-authored more than 50 scientific articles. She initiated the PILOT software project for preoperative surgical planning. She has been involved in the organization of conferences in the fields of medical imaging or computer graphics for many years, including Eurographics 2014, CARS 2019, and she was the General chair of MICCAI 2021. She is the current president of the MICCAI Society.
Professor Gabor Fichtinger (Queen's University)
Point of care ultrasound interventions in the era of artificial intelligence
Recent advent of AI, and particularly deep learning, in medical image analysis has created great opportunities in image-guided interventions and therapies. This progress, however, is hampered by perennial shortage of training data, as the natural state of interventional data is to be lost, leaving deep learning methods little to learn from. This talk discusses the role of a free open-source of image-guided intervention navigation platform, SlicerIGT (www.SlicerIGT.org) in addressing the “training data shortage” problem, and presents the platform’s abilities through many practical examples.
Speaker Bio: Gabor Fichtinger (IEEE M’04, S’2012, F’2016) received his doctoral degree in computer science from the Technical University of Budapest, Budapest, Hungary, in 1990. He is a Professor and Canada Research Chair in Computer-Integrated Surgery at Queen’s University, Canada, where he directs the Percutaneous Surgery Laboratory (a.k.a. Perk Lab). His research and teaching specialize in computational imaging and robotic guidance for surgery and medical interventions, focusing on the diagnosis and therapy of cancer and musculoskeletal conditions. He has contributed more than 700 publications, with over 14,000 citations. Prof. Fichtinger is an Associate Editor for IEEE Transactions on Biomedical Engineering and Elsevier’s Medical Image Analysis, Deputy Editor for the International Journal of Computer Assisted Surgery Radiology. He has served on the boards of the International Society of Medical Image Computing and Computer Assisted Surgery (MICCAI) and the International Society of Computer Assisted Surgery (ISCAS). Prof. Fichtinger received many honours, including the Tier-1 Canada Research Chair, Cancer Care Ontario Research Chair, IEEE Fellow, AIMBE Fellow, MICCAI Fellow, Marie Curie Fellow of the European Community, Distinguished Speaker of ACM, IEEE EMBS Distinguished Lecturer.
Professor Hervé Delingette (INRIA)
Some Strategies to cope with the cost of annotations in Medical Image Analysis
Abstract: Image annotations such as image labels or organ delineations are required to train supervised learning algorithms to solve various tasks in medical image analysis but also to evaluate their performance. Producing high quality annotations is very time consuming especially when dealing with volumetric images. Furthermore, inter-rater variability when producing those annotations has to be taken into account to reflect the complexity of the tasks. In this lecture, I will present some strategies related to data and models to cope with the cost of annotations. A first set of approaches are data-centric and aim to keep only high quality annotations and to precisely measure the agreement or disagreement between the raters. A second set of methods focused on machine learning models try to minimize the amount of required strong annotations for instance through the use of semi-supervised or mixed-supervised techniques.
Speaker Bio: Hervé Delingette is a Research Director at Inria, Scientific director at the Université Côte d’Azur, a member of the Board of the MICCAI Society, and he is holding a Chair at the AI institute 3IA Côte d’Azur. He received his engineering and Phd degrees from Ecole Centrale Paris. His research focuses on various aspects of artificial intelligence in medical image analysis, computational physiology, and surgery simulation.
Professor Kilian Weinberger (Cornell University)
Learning to Detect Objects from Repeated Traversals of the Same Route
Abstract: Recent progress in autonomous driving has been fueled by improvements in machine learning. Ironically, typical autonomous vehicles do not learn while they are in operation. If a car is used in the same location multiple times, it will act identically every single time. We propose to leverage and learn from repetition by allowing a neural network to save features in a data base that can be retrieved later on. If a vehicle is used in the same location multiple times, it builds up a rich data set of past features that aid object detection in the future. This allows it to recognize objects from afar when they are only perceived by a few pixels or LiDAR points. We further demonstrate that it is in fact possible to completely bootstrap an object detection classifier only based on repetition. Our approach has the potential to drastically improve the accuracy and safety of self-driving cars, enable them for sparsely populated areas, and allow them to adapt naturally to their local environment over time.
Speaker Bio: Kilian Weinberger is a Professor in the Department of Computer Science at Cornell University. He received his Ph.D. from the University of Pennsylvania in Machine Learning and his undergraduate degree in Mathematics and Computing from the University of Oxford. During his career he has won several best paper awards at ICML (2004), CVPR (2004, 2017), AISTATS (2005) and KDD (2014, runner-up award). In 2011 he was awarded the Outstanding AAAI Senior Program Chair Award and in 2012 he received an NSF CAREER award. He was elected co-Program Chair for ICML 2016 and for AAAI 2018 and currently serves as a board member and president-elect of the ICML society. In 2016 he was the recipient of the Daniel M Lazar '29 Excellence in Teaching Award. In 2021 he became a finalist for the Blavatnik National Awards for Young Scientists. Kilian Weinberger's research focuses on Machine Learning and its applications, in particular, metric learning, Gaussian Processes, computer vision, perception for autonomous vehicles, and deep learning.
Professor Marleen de Bruijne (Erasmus MC)
Professor Susan Athey (Stanford University)
Using Neural Sequence Models to Represent Sequences of Human Decisions
Abstract: We adapt neural sequence models, most commonly used in natural language processing, to build predictive models of sequences of transitions of individuals between different states. We customize the model to the problem of modeling transitions where it is most likely that an individual remains in the same state, and where individual characteristics are observed. We apply the model to the problem of worker transitions among occupations. We show how fine-tuning can be used to improve prediction in small, curated, administratively collected representative data. We learn representations using large datasets that are not necessarily representative of the population, and then fine-tune on the administrative data. We also show that a this methodology can be used to understand questions like the extent to which gender gaps in wages can be explained by historical experiences, and where in workers’ careers gaps arise. This methodology may have analogs in medicine, where researchers may have access to smaller datasets with more extensive information, but wish to learn representations based on larger datasets, for example datasets containing health claims.
Speaker Bio: Susan Athey is the Economics of Technology Professor at Stanford Graduate School of Business. She received her bachelor’s degree from Duke University and her Ph.D. from Stanford, and she holds an honorary doctorate from Duke University. She previously taught at the economics departments at MIT, Stanford and Harvard. She is an elected member of the National Academy of Science and is the recipient of the John Bates Clark Medal, awarded by the American Economics Association to the economist under 40 who has made the greatest contributions to thought and knowledge. Her current research focuses on the economics of digitization, marketplace design, and the intersection of econometrics and machine learning.