Generative AI in Medicine
The Generative AI in Medicine group is at the forefront of deep learning research, tackling a wide range of topics involving medical data types such as image, text, and speech. Our work spans large language models, diffusion models, and GANs to ultimately develop foundation multimodal models. With a particular focus on medicine and healthcare, our multidisciplinary team of engineers and healthcare professionals collaborates to advance AI-driven solutions for diagnosis and prognosis.
Publications
Large language models streamline automated machine learning for clinical studies
In: Nature Communications 15 (2024), p. 1603 (non-FAU publication)
ISSN: 2041-1723
DOI: 10.1038/s41467-024-45879-8
BibTeX: Download
, , , , , , :
Using machine learning to reduce the need for contrast agents in breast MRI through synthetic images
In: Radiology 307 (2023), Article No.: e222211 (non-FAU publication)
ISSN: 0033-8419
DOI: 10.1148/radiol.222211
BibTeX: Download
, , , , , , , , , , , :
Benchmarking ChatGPT-4 on a radiation oncology in-training exam and Red Journal Gray Zone cases: potentials and challenges for AI-assisted medical education and decision making in radiation oncology
In: Frontiers in Oncology 13 (2023), p. 1-13
ISSN: 2234-943X
DOI: 10.3389/fonc.2023.1265024
BibTeX: Download
, , , , , , , , , , , , , , :
Open Student Positions
- Master’s thesis/project: Vision-Language Models for Pathology Report Generation from Gigapixel Whole-Slide Images
- Master’s thesis/project: Deep Learning-Based Prostate Cancer Grading from Whole-Slide Images
- Master’s thesis/project: Few-Shot Adaptation of Generalist Vision Models for Gastrointestinal Medical Image Analysis
Steps to pursue your MA/BA thesis at the group
Steps:
-
Contact the supervisor from the PRLab and have a joint meeting with them and your company supervisor (if in collaboration with a company) for discussion and organization, such as:
- Starting and finishing dates
- General tasks and expectations
- Communication media and frequency
-
Prepare the thesis description (1-2 pages) PDF document after discussing it with your company supervisor (if in collaboration with a company)
-
A sample PDF, as well as the LaTeX template, are available from your PRLab supervisor.
- Some examples of online versions: 1, 2
Your description should contain the following parts:
- Motivation and background
- Materials and methods
- Research goals of the thesis
-
-
Read about the general information and requirements for theses at our lab:
- https://lme.tf.fau.de/teaching/thesis/thesis-guidelines-english-version/
- Strictly follow the provided template and guidelines
-
Finalize the binding thesis description, including:
- Agreement of the company supervisor (if in collaboration with a company)
- Agreement of the PRLab supervisor
-
Register your thesis officially at the University (your PRLab supervisor should help you with this). You need to have the following requirements:
- [Usually] Passed at least 70 ECTS for master’s students. Please check this carefully with your degree regulations before taking any steps
- Check with your degree regulations to see if you need to have further thesis reviewers/advisors from other departments/faculties (especially for students pursuing the M.Sc. in Data Science)
- Thesis description
- Enrollment document at FAU
- Registration happens with the secretary of the Pattern Recognition Lab
-
Once your registration is complete, you should be able to see it on your FAU account.
-
The time period for submitting your final thesis:
- For full-time students:
- Minimum 3 months
- Maximum 6 months (strictly binding)
- For part-time students:
- Minimum 6 months
- Maximum 12 months (strictly binding)
- For full-time students:
-
The time period is binding and starts with your official registration
Once you started:
- Participation in the respective colloquium
- Choose the correct colloquium with the help of your PRLab supervisor and start attending. Every colloquium takes place on a weekly or biweekly basis in a hybrid format (Zoom and in person). Please refer to the course list on the chair’s website for the individual day and time of the respective colloquium.
- Examples of suitable colloquia depending on your topic:
- If you are working with speech, audio, and text data: SAGI or TSI colloquia
- If you are working with non-medical images: Computer Vision (CV) Colloquium
- If you are working with medical data: LAMBDA Colloquium
- All your presentations also happen in the colloquium (make an appointment with your PRLab supervisor before the presentation).
- MT/BT intro talk:
- A short presentation of the topic at the beginning of the work
- A 15-minute talk followed by 15 minutes Q&A
- A final presentation (defense) shortly before the end of the thesis:
- About 6 weeks before the thesis finishes, the final presentation takes place. This is a 30-minute talk that should include all the results the candidate has achieved so far. However, it is clear that these results are not necessarily “final”; our experience has shown that while discussing the work with a little more concrete data, different people often come up with interesting ideas. Thus, we find it reasonable to have such a discussion before the actual deadline for handing in the thesis. The templates for the slides are the same as for the short presentation.
Finishing your thesis:
- Finalize the thesis with the help of all your supervisors
- Once all the supervisors agree, you can submit your thesis:
- Source codes, networks, latex code, PDF format, etc., should be archived with the IT administration of the Lab and the PRLab supervisor
- Print the mandatory deposit copy of the PDF of the final thesis and submit it to the secretary of the PRLab.
- Your PRLab supervisor review your final submitted thesis, evaluates your performance during the thesis period, and considers your presentations before recommending a final grade to Prof. Maier.
- Your grade will then be sent to the secretary of the PRLab, who will forward it to the examination office.
- You will be notified of your final grade, similar to other courses from the examination office.
- If you have already passed other requirements for your degree, the date of your MA/BA degree will be your MT/BT defense date.
Supervised Student Theses/Projects
Student | Title | Type | Status |
---|---|---|---|
Behnam Norouzi | Online Retrieval Augmented Generation for Accurate Medical Question Answering | MA thesis | running |
Marc Julian Schwarz | Mainframe Meets AI – Improving Legacy Code Generation Through Fine-tuning of Large Language Models | BA thesis | finished |
Philip Drießlein | AI for IT Operations (AIOps): Leveraging Large Language Models as Support for Process Management in Mainframes | MA thesis | finished |
Student | Title | Type | Status |
---|---|---|---|
Robert Kurin | Report Generation and Evaluation for 3D CT Scans Using Large Vision-Language Models | MA thesis | running |
Master Thesis – Annotation by Speech in Radiology | MA thesis | running | |
Sneha Kumari | Utilizing LLMs for medication data annotation in german medical texts | MA thesis | running |
Rajesh Madhipati | Deep-learning based Long-tailed multi label disease classification | MA thesis | running |
Student | Title | Type | Status |
---|---|---|---|
Himanshi Shah | Shallow Networks and AI Explainability in Context of vDCE for Breast MRI | MA thesis | running |
Vicky Vicky | LLM-Centric Framework for Ontology-Driven SPARQL Query Generation in RAG for DICOM Databases | MA thesis | running |
Colloquia
Computer vision colloquium:
Bhavya Bhatia “Optimizing SegFormer through Quantization for Efficient Semantic Segmentation on Mobile Devices” (Intro)
Speech colloquium: