Event
Seminar on Transfer Learning in Medical Imaging
Location
Date
About the Seminar
At this Seminar, the speakers will delve into the world of transfer learning in medical imaging. Discover how this cutting-edge technology is revolutionizing the healthcare industry trough these talks:
Are some NLP tasks really solved?
Abstract
Recent years have seen large performance improvements on Natural Language Processing (NLP) benchmarks, in many cases surpassing human performance. In this talk I will take a closer look at what it means to solve an NLP task, and argue that transfer learning is essential. Based on this I will analyze for two fundamental NLP tasks to what extent they are really solved, namely: word segmentation and language identification. I will look at different languages, domains, and algorithms to provide an overview of remaining challenges.
Bio
Rob van der Goot is an associate professor at the ITU. His main interest is in robustness in NLP. He is mostly known for his work on lexical normalization (i.e. converting social media language to “standard” language), multi-task learning, and evaluation.
Model performance prediction during for quality control
By Lisa Koch
Abstract
Monitoring the performance of machine learning (ML) algorithms in medical imaging is crucial for safely integrating them into patient care. Unfortunately, performance cannot be measured directly during test time, when no labels are available. While for example image segmentation algorithms have achieved an average performance that is sufficient for clinical deployment, even the best models are not guaranteed to perform well on all images and may fail silently on individual ones. Recently, methods have emerged to predict performance metrics without requiring ground truth labels. I will present how such approaches can be applied for quality control and and deployment-time monitoring of image segmentation systems.
Bio
I am an assistant professor for data science in diabetes care at the University of Bern, where I lead the machine learning in medicine lab. My long-term goal is to develop certifiably safe, reliable and effective data science tools for patient-specific treatment systems. I have a background in academic research as well as developing data science products for medical devices, with a strong focus on technology that can ultimately benefit patient health in a safe and trustworthy way. After an undergraduate at ETH Zürich, Switzerland, I did a PhD in machine learning for medical image analysis at Imperial College London, UK. After a post-doc at ETH Zürich I joined the Swiss wearable medical device startup Ava, where I eventually became the data science team lead. In this position, I came to appreciate the need for demonstrably safe machine learning in healthcare. In 2021, I returned to academic research to pursue research on this topic as a group leader for machine learning in medical diagnostics at the Hertie AI institute at the University of Tübingen, Germany.
Medical Image Analysis: then, now, and next?
Abstract
TBA
Bio
Clarisa Sánchez is Full Professor of AI & Health at the University of Amsterdam. She has appointments at Faculty of Science (Informatics Institute) and at the University Medical Center Amsterdam (Departments of Biomedical Engineering and Physics), where she leads the Quantitative Healthcare Analysis (qurAI) group, an interfaculty research group embedded in the Faculties of Medicine and Science. She graduated in Telecommunication Engineering at the University of Valladolid, Spain. In 2008, she obtained her Ph.D. (cum laude) at the University of Valladolid with a thesis entitled “Retinal image analysis by mixture model based clustering and discriminant analysis for automatic detection of hard exudates and hemorrhages. A tool for diabetic retinopathy screening”. From 2008 to 2010, she worked as a postdoc at the Image Sciences Institute of the University Medical Center Utrecht. In 2010 she joined the Diagnostic Image Analysis Group (DIAG) at the Radboud University Medical Center, in Nijmegen, where she became Assistant Professor in 2013 and Associate Professor in 2017. At DIAG she led the A-eye research group, a group focused on the development of AI solutions in Ophthalmology for 1) the automated screening of diseases, 2) the implementation of personalized treatment protocols, and 3) the computerized prediction of progression. In 2020 Clarisa moved with her group to the Informatics Institute, Faculty of Science, University of Amsterdam. Her current research aims at enhancing patient care by designing and enabling leading edge AI technologies in healthcare, specially in the fields of ophthalmology, oncology and intensive care.
Sign up and join us for knowledge expansion and networking with professionals in this field.