Digital Health Bonn

Modern clinical consultations generate rich multimodal data, including spoken dialogue, clinical notes, and structured patient information. However, this information is often fragmented and time-consuming to document. Recent advances in Large Language Models (LLMs) and multimodal AI provide new opportunities to automate clinical documentation while supporting physicians in their daily workflow.

The aim of this thesis is to develop a prototype system that transforms raw clinical consultation data into structured, clinically meaningful, and patient-friendly outputs using LLM-based methods. The project focuses on human-centered AI, where generated content serves as decision support and remains under the supervision of medical professionals.

Your tasks include:

• Generating concise medical summaries from multimodal consultation data
• Extracting and structuring key clinical information, such as diagnosis, severity, and affected anatomical region
• Developing LLM-based methods to generate preliminary treatment and therapy suggestions as physician decision-support drafts
• Creating patient information leaflets in understandable, non-technical language
• Designing and implementing a physician-in-the-loop workflow in which all generated outputs are reviewed and validated by medical experts

Skills and interests:

• Programming experience in Python
• Basic understanding of machine learning concepts
• Interest in AI applications for healthcare
• Experience with prompt engineering or LLM-based applications is beneficial
• Experience with databases, such as PostgreSQL, is a plus
• Familiarity with clinical or biomedical datasets, such as MIMIC-IV, is advantageous

Orthopaedic functional tests are widely used to assess patient mobility, stability, and pain-related movement patterns. Currently, these assessments rely heavily on manual observation by clinicians. Recent advances in computer vision and human pose estimation enable the extraction of skeletal motion data from video, creating new opportunities for automated, objective, and reproducible movement analysis.

The aim of this thesis is to develop a real-time computer vision pipeline that estimates human pose from video streams and automatically evaluates selected orthopaedic functional tests. The resulting system should extract clinically relevant movement parameters and provide structured decision support for clinicians.

Your tasks include:

• Implementing a real-time pose estimation and skeleton tracking pipeline from video streams
• Integrating computer vision frameworks such as MediaPipe and OpenCV for robust motion capture
• Extracting clinically relevant movement features, including joint angles, range of motion, symmetry, and temporal dynamics
• Developing rule-based or machine learning methods for the automated assessment of orthopaedic functional tests
• Designing structured, clinician-friendly reports summarizing test results
• Evaluating and validating the system using recorded and real-world movement data

Skills and interests:

• Programming experience in Python
• Basic understanding of machine learning concepts
• Experience with or interest in computer vision (e.g., OpenCV, MediaPipe, pose estimation)
• Interest in motion data analysis and time-series processing
• Willingness to learn video processing and real-time computer vision pipelines

State-of-the-art pose estimation models achieve impressive accuracy but are often too computationally demanding for real-time use in clinical environments. Applications such as orthopaedic motion analysis require reliable operation on resource-constrained hardware, including laptops, tablets, and embedded systems. This creates a need for optimized AI models that balance accuracy, speed, and energy efficiency.

The aim of this thesis is to optimize modern pose estimation models for fast and efficient inference and deploy them on edge devices for real-world clinical applications. The project will investigate techniques for model optimization while maintaining sufficient accuracy for medical decision support.

Your tasks include:

• Profiling and benchmarking pose estimation models with respect to runtime, memory usage, and prediction accuracy
• Implementing inference optimization techniques, such as hardware acceleration and pipeline optimization
• Applying model compression methods, including quantization, pruning, or knowledge distillation
• Adapting and deploying optimized models on edge devices, such as mobile or embedded systems
• Integrating the optimized model into a lightweight real-time processing pipeline
• Evaluating performance trade-offs between speed, memory consumption, and accuracy in realistic clinical scenarios

Skills and interests:

• Programming experience in Python
• Basic understanding of machine learning concepts
• Interest in model optimization and efficient AI systems
• Interest in healthcare applications of AI
• Experience with or interest in computer vision and pose estimation
• Interest in real-time systems and edge deployment
• Experience with PyTorch, TensorFlow Lite, ONNX, or model compression techniques is beneficial

Patients often face uncertainty about upcoming examinations, treatments, and clinical procedures. Information is typically communicated verbally and can be difficult to remember, leading to confusion and reduced patient engagement. Modern digital systems offer the opportunity to provide personalized, structured, and easy-to-understand guidance throughout the patient’s clinical journey.

The aim of this thesis is to design and develop an interactive patient information system that explains clinical processes and guides patients through their individual treatment pathway. The system should present medical information in a clear, patient-friendly manner while supporting clinicians in communicating complex care pathways.

Your tasks include:

• Analyzing and modeling clinical workflows and patient journeys
• Designing and implementing a backend system to manage patient-specific information
• Developing a user-friendly web, mobile, or tablet-based interface for patients
• Integrating structured medical data into understandable, patient-friendly content
• Optionally integrating LLM-based components to generate personalized explanations and answer patient questions
• Evaluating the usability and effectiveness of the system in realistic clinical scenarios

Skills and interests:

• Programming experience in Python, JavaScript, or a similar language
• Interest in software engineering and full-stack development
• Interest in UX/UI design and human-centered software systems
• Interest in digital health and patient-centered applications
• Basic understanding of databases and backend development
• Experience with web frameworks (e.g., React, FastAPI, or Node.js) is beneficial

Patients often leave clinical consultations with unanswered questions or difficulty recalling important information about their diagnosis, treatment, or next steps. Although written materials are helpful, they are static and cannot address individual concerns. Conversational AI systems provide the opportunity to deliver personalized, interactive support that is available whenever patients need it.

The aim of this thesis is to design and develop a conversational AI assistant that helps patients better understand their medical condition, answers questions about their treatment, and guides them through their clinical journey. The system should provide reliable, understandable, and safe responses while supporting—not replacing—medical professionals.

Your tasks include:

• Designing and implementing a conversational user interface for patient interaction
• Integrating Large Language Models (LLMs) for natural language understanding and response generation
• Developing context-aware conversations by incorporating patient information, treatment plans, and clinical documents
• Implementing safety mechanisms and guardrails for trustworthy medical communication
• Evaluating response quality, usability, and patient experience
• Optionally integrating the assistant with clinical information systems or patient data sources

Skills and interests:

• Programming experience in Python and/or JavaScript
• Interest in natural language processing and Large Language Models
• Interest in human-computer interaction and conversational interface design
• Interest in healthcare applications and patient communication
• Basic understanding of software engineering, APIs, and backend systems
• Experience with LLM frameworks (e.g., LangChain), API integration, or prompt engineering is beneficial

In this thesis, we investigate whether eye movements can be reliably tracked using EEG electrodes placed close to the eyes, i.e. by means of electrooculography (EOG). To obtain a ground truth for eye movements, recordings are performed in parallel with eye-tracking glasses (Pupil Labs Neon).

The experimental setup as well as the Matlab-based study software (Psychtoolbox) are already in place and are supported by members of Prof. Krüger’s group.

In the first phase of the thesis, the doctoral candidate will conduct the study with patients in the Department of Epileptology. Each eligible patient may participate in the study on up to five consecutive days.

In the second phase, the recorded data will be analyzed, including the application of machine learning methods to assess how well EOG-based signals can capture relevant eye movement patterns.

Your tasks include:

• Conducting the study with patients in the Department of Epileptology
• Working with an existing experimental setup and Matlab/Psychtoolbox codebase
• Processing and analyzing multimodal eye movement data
• Applying and evaluating machine learning methods for signal analysis

Some programming experience, for example in Python, is helpful, but can also be acquired during the thesis.

Clinical EEG cables are subject to significant mechanical and chemical stress due to repeated use, handling, and cleaning procedures. Undetected cable faults can lead to degraded signal quality or even complete signal loss, which may compromise clinical assessments.

Current procedures for testing and repairing EEG cables are often time-consuming and prone to human error. This thesis aims to address these limitations by developing a dedicated cable testing device.

The goal is to design and implement a system that goes beyond simple conductance testing and enables more advanced diagnostics, such as frequency sweeps, to better characterize cable integrity and performance.

Your tasks include:

• Design and development of a hardware-based cable testing device
• Implementation of measurement procedures (e.g., conductance tests, frequency sweeps)
• Integration of microcontroller-based control and data acquisition
• Development of software interfaces for data analysis and visualization
• Evaluation of the system in a clinical EEG context

Skills and interests:

• PCB design
• Microcontroller programming (µC)
• Programming experience in Python and/or C

Recent advances in radar-based sensing have enabled the contactless measurement of vital parameters such as respiratory rate and even heart rate. A growing number of commercially available modules and development kits promise reliable acquisition of these signals without requiring any physical contact with the patient.

This technology holds significant potential for applications in hospitals, home care, and unobtrusive long-term monitoring scenarios. However, the actual performance, robustness, and clinical applicability of these systems often vary considerably, and some solutions may not meet their advertised capabilities.

The goal of this thesis is to systematically evaluate a range of radar-based sensor modules and development kits. The aim is to identify which systems provide reliable and meaningful measurements under realistic conditions and to distinguish promising technologies from less suitable solutions.

Depending on the focus, the project may also include the development of experimental setups and signal processing pipelines to assess measurement quality and robustness.

Your tasks include:

• Review and selection of commercially available radar-based sensing modules
• Design and implementation of experimental evaluation setups
• Acquisition and analysis of radar-based vital parameter data
• Comparison of different systems with respect to accuracy, robustness, and usability
• Optional: Development of signal processing or machine learning approaches for improved estimation

The project can be conducted in collaboration with Fraunhofer FHR (Wachtberg), providing access to additional expertise and infrastructure in radar technology.

Skills and interests:

• Interest in sensor systems and signal processing
• Experience in programming (e.g., Python, MATLAB, or C/C++)
• Basic knowledge of physics, electronics, or data analysis