English
Deutsch
Lehre
Lehrstuhl für KI in der Medizin
Lehre und Ausbildung sind ein wesentlicher Bestandteil der Aufgaben unseres Instituts. Alle unsere – stark von unserer Forschung geprägten – Lehrveranstaltungen finden auf Englisch statt. Wir bieten Vorlesungen für Studierende verschiedener Disziplinen, unsere Hauptvorlesungen richten sich jedoch an Informatikstudent*innen:
- KI in der Medizin I (Wintersemester)
- KI in der Medizin II (Sommersemester)
Begleitend zur Hauptvorlesungsserie werden die einzelnen Teilbereiche und Themengebiete anhand von Praktika und Seminaren vertieft. In direktem Austausch mit unseren Dozent*innen können die Studierenden so ihr erlerntes Wissen intensivieren und anwenden. Aktuell bieten wir folgende Seminare und Praktika:
- Master-Seminar – Deep Learning for Inverse Problems
- Master-Seminar – Trustworthy AI in Healthcare
- Practical – Applied Deep Learning in Medicine
Als Wahlfach für Medizinstudierende bieten wir an:
- Computer Science for Medical Students
Dieser Kurs bietet den Studierenden spannende Einblicke in die Welt der KI-Methoden (insbesondere neuronale Netze) und deren Anwendungen in der Medizin. Neben dem Erwerb theoretischen Grundlagenwissens sammeln sie erste praktische Erfahrungen mit Python-Programmierung und bekommen die Möglichkeit, eigene neuronale Netze zu trainieren.
Neben Vorlesungen, Seminaren und Praktika bieten wir eine Reihe Bachelor- und Masterprojekte an. Unsere aktuellen Ausschreibungen für offene Projekte finden Sie hier:
B.Sc. Thesis/Guided Research/IDP: Influence of slice thickness of pulmonary CTs on Semantic Segmentation Algorithms
Description Covid-19 can cause various types of pathologies within the lung. Deep learning is widely used to segment affected areas of the lung using computed tomography images as input. However, these images can be acquired using a large variety of imaging parameters.
B.Sc. Thesis: Collection of a German Biomedical Text Corpus from Public Sources
Recent successes in Natural Language Processing (NLP) are based on pre-training language models on large datasets of unlabelled text. In the medical domain however, such large datasets are hard to acquire.
MSc Thesis: Interpreting age prediction in whole body MR images
Age prediction has been investigated intensively for brain imaging (brain age estimation) [1]. Here, it has been applied for early prediction of neurodegenerative diseases and the assessment of accelerated aging. However, there is very little research on abdominal age prediction which estimates the age of a person based on abdominal organs [2].
MSc Thesis: Cardiac MRI Segmentation using Morphometric Informed Multimodal Self-Supervised Models
Description Medical datasets and especially biobanks, often contain extensive tabular data with rich clinical information in addition to images. In practice, clinicians typically have less data, both in terms of diversity and scale, but still, wish to deploy deep learning solutions.
MSc Thesis: Tabular Feature Selection and Shared Latent Space Explainability in Self-Supervised Multimodal Deep Learning
Description Medical datasets and especially biobanks, often contain extensive tabular data with rich clinical information in addition to images. In practice, clinicians typically have less data, both in terms of diversity and scale, but still, wish to deploy deep learning solutions.
MSc Thesis: An attention-based image denoising network leveraging information of both spatial and frequency domain
Description Image denoising task, in which a clean image is recovered from a noise observation, is a classical inverse problem and still active topic in low-level vision since it is an indispensable step in many practical applications.
MSc Thesis: Motion-Compensated MRI Reconstruction
Long acquisition times in Magnetic Resonance Imaging (MRI) bear the risk of patient motion, which substantially degrades the image quality. Further sources of image degradation are physiological motion, such as periodic respiratory and cardiac motion.
MSc Thesis: Interpreting age prediction in whole body MR images
Age prediction has been investigated intensively for brain imaging (brain age estimation) [1]. Here, it has been applied for early prediction of neurodegenerative diseases and the assessment of accelerated aging. However, there is very little research on abdominal age prediction which estimates the age of a person based on abdominal organs [2].
MSc Thesis: Combining longitudinal volumetric imaging and tabular data for depression prediction
The objective of this project is to combine volumetric imaging and non-imaging longitudinal data to accurately analyze individual patients and provide automated decisions regarding diagnosis and disease prognosis. Physicians consider various medical biomarkers and meta-data to reach a clinical decision.
MSc Thesis: Federated, privacy preserving deep learning for fetal MRI brain segmentation
In medical domain data protection and privacy preservation is highly relevant. Therefore, clinical data of patients is usually securely stored on clinic servers without access from outside. Publication of clinical data is difficult and cumbersome as strict privacy and data protection laws must be obeyed.
MSc Thesis: Non-invasive and accurate prediction of prostate cancer aggressiveness
With estimates of 1 600 000 cases and more than 350 000 deaths annually worldwide, prostate cancer is among the most common cancers in men [1]. Diagnosis of prostate cancer is typically done by using ultrasound-guided needle biopsies.
MSc Thesis: Adversarial attacks in collaborative machine learning
Collaborative machine learning has became the new paradigm-of-choice when it comes to training deep learning models in many fields, including medical image analysis. Due to a number of data protection and governance regulations being introduced, direct data sharing for such training is rendered problematic.
MSc Thesis: Transformer based Graph Extraction
In this Master thesis we aim to address the graph extraction problem using a new powerful class of neural networks - Transformers [2][3]. A comprehensive representation of an image requires understanding objects and their mutual relationship, especially in image-to-graph generation, e.
MSc Thesis: Defending collaborative machine learning through interpretability methods
Collaborative machine learning has became the new paradigm-of-choice when it comes to training deep learning models in many fields, including medical image analysis. Due to a number of data protection and governance regulations being introduced, direct data sharing for such training is rendered problematic.
MSc Thesis: Transformer-based Mesh Analysis for Brain Aneurysm Detection
Detecting aneurysm in 3D volumetric TOF MRA is difficult because a) the variation in shape information of aneurysms is hard to capture in patch-based 3D approach using just intensity information b) processing full 3D volumetric image is computationally demanding.
MSc Thesis: AI against SARS-CoV-2
Viruses interact with cellular proteins to replicate and spread. We aim to gain functional insights into the mode of action of cellular proteins, enabling us to better understand how different viruses like SARS-CoV-2 cause disease.
MSc Thesis: Contrastive Pre-Training for Radiology Reports
In recent years transformer-based language models have proven quite successful in the field of natural language processing (NLP). These models require huge amounts of training data and are therefore typically pre-trained on unlabelled datasets using self-supervised objectives like masked language modelling (MLM) as proposed in BERT [1].
MSc Thesis: Unsupervised deep learning for vessel segmentation in optical coherence tomography angiographs
Optical coherence tomography angiography (OCTA) is an imaging technique that visualizes blood vessels by detecting motion of red blood cells in sequential scans [1]. It has seen initial adoption for the diagnosis and monitoring of clinical conditions that affect the retinal vasculature, such as several different eye diseases or multiple sclerosis [2, 3].
MSc Thesis: Prediction of long-term cognitive outcome in Stroke patients using machine learning
Machine learning, in particular deep learning, has reformed the research in the field of medical imaging, and the focus of this project will be on its use for the prediction of disease progression/ neurological outcome in stroke patients.
MSc Thesis: Machine Learning for Analysis of Sarcoma
Description Early diagnosis of musculoskeletal tumours is crucial for successful therapy and treatment. The sooner a potential malignant growth is detected, the more effective the next steps in therapy and the better a prognosis usually becomes.
MSc Thesis: Distributionally robust neural networks in medical imaging
Deep learning has revolutionized the field of medical imaging. However, the performance of a model drops when the distribution of the test data is different from the distribution of the training data.
MSc Thesis: Out-of-distribution detection using contrastive training for medical imaging
Deep learning has revolutionized the field of medical imaging. However, the performance of a model drops when the distribution of the test data is different from the distribution of the training data.
MSc Thesis: Privacy-preserving Deep Learning in Medical Imaging
Privacy-preserving artificial intelligence techniques such as differential privacy, encryption and multi-party computation can reconcile the needs for data utilisation and data protection in the medical domain, as mandated by legal and ethical requirements.
