Block-Seminar on Deep Learning for Bio-Medical Image Analysis

apl. Prof. Olaf Ronneberger (Google DeepMind)

In this seminar you will learn about relevant bio-medical research fields and the most recent methods (mainly based on deep learning) that have been already applied to bio-medical data, or that have a large potential in this field. For each paper there will be one person, who performs a detailed investigation of a research paper and its background and will give a presentation. The presentation is followed by a discussion with all participants about the merits and limitations of the respective paper. You will learn to read and understand contemporary research papers, to give a good oral presentation, to ask questions, and to openly discuss a research problem.

(2 SWS)
Room: t.b.a.
Places: max. 10 students
Contact person: Özgün Çiçek

Introduction: Thursday, October 18, 2018, 10:15
Room: 52-2-17
Introduction and allocation of seminar topics
Will be held jointly with Seminar on Current Works in Computer Vison

Mid-Semester Meeting: Friday, December 14, 2018, 10:00
Room: t.b.a.
Introduction to Neural Networks by apl. Prof. Olaf Ronneberger (Google DeepMind)

ECTS Credits: 4

Recommended semester:

6 (Bachelor), any (Master)
Requirements: Background in computer vision

Remarks: This course is offered to both Bachelor and Master students. The language of this course is English. All presentations must be given in English.

Topics will be assigned in the first meeting. Please register for the seminar online before the first meeting. If you could not register still come to our introductory meeting to see if there are papers free. If there are more interested students than places, places will be assigned by a mixture of motivation in the first meeting and priority suggestions of the system. The date of registration is NOT important. In particular, we want to avoid that people grab a topic and then jump off during the semester. Please have a coarse look at all available papers to make an informed decision in the first meeting.

There is a related Seminar on Current Works in Computer Vision offered by Prof. Thomas Brox

Please get in contact with your advisor as soon as possible, and at least 4 weeks before your presentation

Submit your presentation outline to your advisor at least 2 weeks before your presentation and meet with your advisor.

Submit your presentation slides to your advisor at least 1 week before your presentation and meet again.

All participants must read all papers and answer a few questions. The questions will be available here. The answers must be sent to the corresponding advisor until 31.01.2019 18:00. We highly recommend to read and understand all papers first, before you start to prepare your presentation.

Papers (please note that we only have 10 places):

Time Paper Presenting student   Advisor Slides
nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation - Winner of the medical image segmentation decathlon at MICCAI 2018 Philipp Jankov Thorsten Falk
Representation Learning with Contrastive Predictive Coding - Training labels are always rare in bio-medical imaging applications. This paper proposes a simple but powerful technique for unsupervised representation learning. New state of the art on several benchmarks. Rebekka Rupprecht Max Argus
Tracking Emerges by Colorizing Videos - Nice way to approach the tracking problem where gt cannot be easily generated Ji Qi Özgün Çiçek
Semi-convolutional operators for instance segmentation - Makes use of position sensitive embeddings to separate objects Lum Birinxhiku Anton Böhm
Large Scale GAN Training for High Fidelity Natural Image Synthesis - Even though we do not want to create fake medical images, this paper provides quite a few insights how to build powerful generative models. Hüseyin Furkan Bozkurt Mohammedreza Zolfaghari
Attentive Neural Processes - Reliable modelling of the uncertainty is the next big challenge in medical image analysis. Attentive neural processes are a promising direction. Joshua Heipel Maxim Tatarchenko
Cardiac MR Segmentation from Undersampled k-space Using Deep Latent Representation Learning - Predicting the segmentation maps directly from the raw dynamic MRI data allows extreme undersampling and consequently very fast imaging. Rolf-David Bergdoll Özgün Çiçek
A Probabilistic U-Net for Segmentation of Ambiguous Images - Medical images are often ambiguous. The proposed probabilistic U-net provides all possible segmentation variants with calibrated probabilities. Julien Siems Osama Makansi