Keynotes

Full Professor at Heidelberg University

Managing Director of Data Science and Digital Oncology at the German Cancer Research Center (DKFZ)

Title: Machine Learning in Medical Imaging – Current Challenges

Abstract: Despite its vast potential, the actual practice-changing clinical impact of machine learning in medical imaging has so far been rather modest. Why is that? The talk will cover several major challenges that I consider essential in unlocking the full potential of machine learning in medical imaging, and I will present current examples of our ongoing research that address them.

Professor

Institute for Information and System Sciences

School of Mathematics and Statistics

Xi’an Jiaotong University

Title: CT image enhancement through Noise/loss Modeling

Abstract: The loss function used in a conventional machine learning problem is generally specified as an easy fixed form, like L2 norm or L1 norm, which intrinsically assumes the noises contained in data are generated from a simple distribution, like an i.i.d. Gaussian or Laplacian. However, in practical scenarios with complex noise configurations, like low-dose CT image, such modeling inclines to encounter the robustness issue, that is, such modeling manner tends to make the related learning algorithm sensitive to complex noises. In this talk, I will introduce some developments of our research team on noise/loss modeling, which aims to make a machine learning model capable of adaptively learning an appropriate loss function/noise distribution from data, so as to alleviate the robustness issue of generally machine learning regimes. Such loss/noise modeling paradigms have been used on multiple image/video/hyper-spectral image restoration tasks, and achieved state-of-the-art performance on hyper-spectral image denoising, online background subtraction on surveillance videos, low-dose CT image enhancement and video deraining. Such a fundamental regime is expected to inspire useful learning algorithms for more machine learning tasks.

Senior Investigator

Clinical Image Processing Service

Imaging Biomarkers and Computer-Aided Diagnosis Laboratory

Title: Challenges and Opportunities for AI in Abdominal Radiology

Abstract: AI methodology has improved dramatically over just the past five years. Clinical implementation however has moved much more slowly. In this talk, I will provide an overview of how state-of-the-art AI can help radiologists make more accurate and efficient diagnoses and predictions. I will focus on AI applications in the abdomen, both my area of clinical specialty and a relatively less traveled topic area in AI. I will also cover such important topics as explainability, the role of large data sets, and ethics.